xref: /linux/drivers/infiniband/ulp/rtrs/rtrs-srv.c (revision 57390019b68b83f96eb98f490367b9df1f2d77cb)
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 
15 #include "rtrs-srv.h"
16 #include "rtrs-log.h"
17 #include <rdma/ib_cm.h>
18 #include <rdma/ib_verbs.h>
19 #include "rtrs-srv-trace.h"
20 
21 MODULE_DESCRIPTION("RDMA Transport Server");
22 MODULE_LICENSE("GPL");
23 
24 /* Must be power of 2, see mask from mr->page_size in ib_sg_to_pages() */
25 #define DEFAULT_MAX_CHUNK_SIZE (128 << 10)
26 #define DEFAULT_SESS_QUEUE_DEPTH 512
27 #define MAX_HDR_SIZE PAGE_SIZE
28 
29 static struct rtrs_rdma_dev_pd dev_pd;
30 const struct class rtrs_dev_class = {
31 	.name = "rtrs-server",
32 };
33 static struct rtrs_srv_ib_ctx ib_ctx;
34 
35 static int __read_mostly max_chunk_size = DEFAULT_MAX_CHUNK_SIZE;
36 static int __read_mostly sess_queue_depth = DEFAULT_SESS_QUEUE_DEPTH;
37 
38 static bool always_invalidate = true;
39 module_param(always_invalidate, bool, 0444);
40 MODULE_PARM_DESC(always_invalidate,
41 		 "Invalidate memory registration for contiguous memory regions before accessing.");
42 
43 module_param_named(max_chunk_size, max_chunk_size, int, 0444);
44 MODULE_PARM_DESC(max_chunk_size,
45 		 "Max size for each IO request, when change the unit is in byte (default: "
46 		 __stringify(DEFAULT_MAX_CHUNK_SIZE) "KB)");
47 
48 module_param_named(sess_queue_depth, sess_queue_depth, int, 0444);
49 MODULE_PARM_DESC(sess_queue_depth,
50 		 "Number of buffers for pending I/O requests to allocate per session. Maximum: "
51 		 __stringify(MAX_SESS_QUEUE_DEPTH) " (default: "
52 		 __stringify(DEFAULT_SESS_QUEUE_DEPTH) ")");
53 
54 static cpumask_t cq_affinity_mask = { CPU_BITS_ALL };
55 
56 static struct workqueue_struct *rtrs_wq;
57 
58 static inline struct rtrs_srv_con *to_srv_con(struct rtrs_con *c)
59 {
60 	return container_of(c, struct rtrs_srv_con, c);
61 }
62 
63 static bool rtrs_srv_change_state(struct rtrs_srv_path *srv_path,
64 				  enum rtrs_srv_state new_state)
65 {
66 	enum rtrs_srv_state old_state;
67 	bool changed = false;
68 
69 	spin_lock_irq(&srv_path->state_lock);
70 	old_state = srv_path->state;
71 	switch (new_state) {
72 	case RTRS_SRV_CONNECTED:
73 		if (old_state == RTRS_SRV_CONNECTING)
74 			changed = true;
75 		break;
76 	case RTRS_SRV_CLOSING:
77 		if (old_state == RTRS_SRV_CONNECTING ||
78 		    old_state == RTRS_SRV_CONNECTED)
79 			changed = true;
80 		break;
81 	case RTRS_SRV_CLOSED:
82 		if (old_state == RTRS_SRV_CLOSING)
83 			changed = true;
84 		break;
85 	default:
86 		break;
87 	}
88 	if (changed)
89 		srv_path->state = new_state;
90 	spin_unlock_irq(&srv_path->state_lock);
91 
92 	return changed;
93 }
94 
95 static void free_id(struct rtrs_srv_op *id)
96 {
97 	if (!id)
98 		return;
99 	kfree(id);
100 }
101 
102 static void rtrs_srv_free_ops_ids(struct rtrs_srv_path *srv_path)
103 {
104 	struct rtrs_srv_sess *srv = srv_path->srv;
105 	int i;
106 
107 	if (srv_path->ops_ids) {
108 		for (i = 0; i < srv->queue_depth; i++)
109 			free_id(srv_path->ops_ids[i]);
110 		kfree(srv_path->ops_ids);
111 		srv_path->ops_ids = NULL;
112 	}
113 }
114 
115 static void rtrs_srv_rdma_done(struct ib_cq *cq, struct ib_wc *wc);
116 
117 static struct ib_cqe io_comp_cqe = {
118 	.done = rtrs_srv_rdma_done
119 };
120 
121 static inline void rtrs_srv_inflight_ref_release(struct percpu_ref *ref)
122 {
123 	struct rtrs_srv_path *srv_path = container_of(ref,
124 						      struct rtrs_srv_path,
125 						      ids_inflight_ref);
126 
127 	percpu_ref_exit(&srv_path->ids_inflight_ref);
128 	complete(&srv_path->complete_done);
129 }
130 
131 static int rtrs_srv_alloc_ops_ids(struct rtrs_srv_path *srv_path)
132 {
133 	struct rtrs_srv_sess *srv = srv_path->srv;
134 	struct rtrs_srv_op *id;
135 	int i, ret;
136 
137 	srv_path->ops_ids = kcalloc(srv->queue_depth,
138 				    sizeof(*srv_path->ops_ids),
139 				    GFP_KERNEL);
140 	if (!srv_path->ops_ids)
141 		goto err;
142 
143 	for (i = 0; i < srv->queue_depth; ++i) {
144 		id = kzalloc(sizeof(*id), GFP_KERNEL);
145 		if (!id)
146 			goto err;
147 
148 		srv_path->ops_ids[i] = id;
149 	}
150 
151 	ret = percpu_ref_init(&srv_path->ids_inflight_ref,
152 			      rtrs_srv_inflight_ref_release, 0, GFP_KERNEL);
153 	if (ret) {
154 		pr_err("Percpu reference init failed\n");
155 		goto err;
156 	}
157 	init_completion(&srv_path->complete_done);
158 
159 	return 0;
160 
161 err:
162 	rtrs_srv_free_ops_ids(srv_path);
163 	return -ENOMEM;
164 }
165 
166 static inline void rtrs_srv_get_ops_ids(struct rtrs_srv_path *srv_path)
167 {
168 	percpu_ref_get(&srv_path->ids_inflight_ref);
169 }
170 
171 static inline void rtrs_srv_put_ops_ids(struct rtrs_srv_path *srv_path)
172 {
173 	percpu_ref_put(&srv_path->ids_inflight_ref);
174 }
175 
176 static void rtrs_srv_reg_mr_done(struct ib_cq *cq, struct ib_wc *wc)
177 {
178 	struct rtrs_srv_con *con = to_srv_con(wc->qp->qp_context);
179 	struct rtrs_path *s = con->c.path;
180 	struct rtrs_srv_path *srv_path = to_srv_path(s);
181 
182 	if (wc->status != IB_WC_SUCCESS) {
183 		rtrs_err(s, "REG MR failed: %s\n",
184 			  ib_wc_status_msg(wc->status));
185 		close_path(srv_path);
186 		return;
187 	}
188 }
189 
190 static struct ib_cqe local_reg_cqe = {
191 	.done = rtrs_srv_reg_mr_done
192 };
193 
194 static int rdma_write_sg(struct rtrs_srv_op *id)
195 {
196 	struct rtrs_path *s = id->con->c.path;
197 	struct rtrs_srv_path *srv_path = to_srv_path(s);
198 	dma_addr_t dma_addr = srv_path->dma_addr[id->msg_id];
199 	struct rtrs_srv_mr *srv_mr;
200 	struct ib_send_wr inv_wr;
201 	struct ib_rdma_wr imm_wr;
202 	struct ib_rdma_wr *wr = NULL;
203 	enum ib_send_flags flags;
204 	size_t sg_cnt;
205 	int err, offset;
206 	bool need_inval;
207 	u32 rkey = 0;
208 	struct ib_reg_wr rwr;
209 	struct ib_sge *plist;
210 	struct ib_sge list;
211 
212 	sg_cnt = le16_to_cpu(id->rd_msg->sg_cnt);
213 	need_inval = le16_to_cpu(id->rd_msg->flags) & RTRS_MSG_NEED_INVAL_F;
214 	if (sg_cnt != 1)
215 		return -EINVAL;
216 
217 	offset = 0;
218 
219 	wr		= &id->tx_wr;
220 	plist		= &id->tx_sg;
221 	plist->addr	= dma_addr + offset;
222 	plist->length	= le32_to_cpu(id->rd_msg->desc[0].len);
223 
224 	/* WR will fail with length error
225 	 * if this is 0
226 	 */
227 	if (plist->length == 0) {
228 		rtrs_err(s, "Invalid RDMA-Write sg list length 0\n");
229 		return -EINVAL;
230 	}
231 
232 	plist->lkey = srv_path->s.dev->ib_pd->local_dma_lkey;
233 	offset += plist->length;
234 
235 	wr->wr.sg_list	= plist;
236 	wr->wr.num_sge	= 1;
237 	wr->remote_addr	= le64_to_cpu(id->rd_msg->desc[0].addr);
238 	wr->rkey	= le32_to_cpu(id->rd_msg->desc[0].key);
239 	if (rkey == 0)
240 		rkey = wr->rkey;
241 	else
242 		/* Only one key is actually used */
243 		WARN_ON_ONCE(rkey != wr->rkey);
244 
245 	wr->wr.opcode = IB_WR_RDMA_WRITE;
246 	wr->wr.wr_cqe   = &io_comp_cqe;
247 	wr->wr.ex.imm_data = 0;
248 	wr->wr.send_flags  = 0;
249 
250 	if (need_inval && always_invalidate) {
251 		wr->wr.next = &rwr.wr;
252 		rwr.wr.next = &inv_wr;
253 		inv_wr.next = &imm_wr.wr;
254 	} else if (always_invalidate) {
255 		wr->wr.next = &rwr.wr;
256 		rwr.wr.next = &imm_wr.wr;
257 	} else if (need_inval) {
258 		wr->wr.next = &inv_wr;
259 		inv_wr.next = &imm_wr.wr;
260 	} else {
261 		wr->wr.next = &imm_wr.wr;
262 	}
263 	/*
264 	 * From time to time we have to post signaled sends,
265 	 * or send queue will fill up and only QP reset can help.
266 	 */
267 	flags = (atomic_inc_return(&id->con->c.wr_cnt) % s->signal_interval) ?
268 		0 : IB_SEND_SIGNALED;
269 
270 	if (need_inval) {
271 		inv_wr.sg_list = NULL;
272 		inv_wr.num_sge = 0;
273 		inv_wr.opcode = IB_WR_SEND_WITH_INV;
274 		inv_wr.wr_cqe   = &io_comp_cqe;
275 		inv_wr.send_flags = 0;
276 		inv_wr.ex.invalidate_rkey = rkey;
277 	}
278 
279 	imm_wr.wr.next = NULL;
280 	if (always_invalidate) {
281 		struct rtrs_msg_rkey_rsp *msg;
282 
283 		srv_mr = &srv_path->mrs[id->msg_id];
284 		rwr.wr.opcode = IB_WR_REG_MR;
285 		rwr.wr.wr_cqe = &local_reg_cqe;
286 		rwr.wr.num_sge = 0;
287 		rwr.mr = srv_mr->mr;
288 		rwr.wr.send_flags = 0;
289 		rwr.key = srv_mr->mr->rkey;
290 		rwr.access = (IB_ACCESS_LOCAL_WRITE |
291 			      IB_ACCESS_REMOTE_WRITE);
292 		msg = srv_mr->iu->buf;
293 		msg->buf_id = cpu_to_le16(id->msg_id);
294 		msg->type = cpu_to_le16(RTRS_MSG_RKEY_RSP);
295 		msg->rkey = cpu_to_le32(srv_mr->mr->rkey);
296 
297 		list.addr   = srv_mr->iu->dma_addr;
298 		list.length = sizeof(*msg);
299 		list.lkey   = srv_path->s.dev->ib_pd->local_dma_lkey;
300 		imm_wr.wr.sg_list = &list;
301 		imm_wr.wr.num_sge = 1;
302 		imm_wr.wr.opcode = IB_WR_SEND_WITH_IMM;
303 		ib_dma_sync_single_for_device(srv_path->s.dev->ib_dev,
304 					      srv_mr->iu->dma_addr,
305 					      srv_mr->iu->size, DMA_TO_DEVICE);
306 	} else {
307 		imm_wr.wr.sg_list = NULL;
308 		imm_wr.wr.num_sge = 0;
309 		imm_wr.wr.opcode = IB_WR_RDMA_WRITE_WITH_IMM;
310 	}
311 	imm_wr.wr.send_flags = flags;
312 	imm_wr.wr.ex.imm_data = cpu_to_be32(rtrs_to_io_rsp_imm(id->msg_id,
313 							     0, need_inval));
314 
315 	imm_wr.wr.wr_cqe   = &io_comp_cqe;
316 	ib_dma_sync_single_for_device(srv_path->s.dev->ib_dev, dma_addr,
317 				      offset, DMA_BIDIRECTIONAL);
318 
319 	err = ib_post_send(id->con->c.qp, &id->tx_wr.wr, NULL);
320 	if (err)
321 		rtrs_err(s,
322 			  "Posting RDMA-Write-Request to QP failed, err: %d\n",
323 			  err);
324 
325 	return err;
326 }
327 
328 /**
329  * send_io_resp_imm() - respond to client with empty IMM on failed READ/WRITE
330  *                      requests or on successful WRITE request.
331  * @con:	the connection to send back result
332  * @id:		the id associated with the IO
333  * @errno:	the error number of the IO.
334  *
335  * Return 0 on success, errno otherwise.
336  */
337 static int send_io_resp_imm(struct rtrs_srv_con *con, struct rtrs_srv_op *id,
338 			    int errno)
339 {
340 	struct rtrs_path *s = con->c.path;
341 	struct rtrs_srv_path *srv_path = to_srv_path(s);
342 	struct ib_send_wr inv_wr, *wr = NULL;
343 	struct ib_rdma_wr imm_wr;
344 	struct ib_reg_wr rwr;
345 	struct rtrs_srv_mr *srv_mr;
346 	bool need_inval = false;
347 	enum ib_send_flags flags;
348 	u32 imm;
349 	int err;
350 
351 	if (id->dir == READ) {
352 		struct rtrs_msg_rdma_read *rd_msg = id->rd_msg;
353 		size_t sg_cnt;
354 
355 		need_inval = le16_to_cpu(rd_msg->flags) &
356 				RTRS_MSG_NEED_INVAL_F;
357 		sg_cnt = le16_to_cpu(rd_msg->sg_cnt);
358 
359 		if (need_inval) {
360 			if (sg_cnt) {
361 				inv_wr.wr_cqe   = &io_comp_cqe;
362 				inv_wr.sg_list = NULL;
363 				inv_wr.num_sge = 0;
364 				inv_wr.opcode = IB_WR_SEND_WITH_INV;
365 				inv_wr.send_flags = 0;
366 				/* Only one key is actually used */
367 				inv_wr.ex.invalidate_rkey =
368 					le32_to_cpu(rd_msg->desc[0].key);
369 			} else {
370 				WARN_ON_ONCE(1);
371 				need_inval = false;
372 			}
373 		}
374 	}
375 
376 	trace_send_io_resp_imm(id, need_inval, always_invalidate, errno);
377 
378 	if (need_inval && always_invalidate) {
379 		wr = &inv_wr;
380 		inv_wr.next = &rwr.wr;
381 		rwr.wr.next = &imm_wr.wr;
382 	} else if (always_invalidate) {
383 		wr = &rwr.wr;
384 		rwr.wr.next = &imm_wr.wr;
385 	} else if (need_inval) {
386 		wr = &inv_wr;
387 		inv_wr.next = &imm_wr.wr;
388 	} else {
389 		wr = &imm_wr.wr;
390 	}
391 	/*
392 	 * From time to time we have to post signalled sends,
393 	 * or send queue will fill up and only QP reset can help.
394 	 */
395 	flags = (atomic_inc_return(&con->c.wr_cnt) % s->signal_interval) ?
396 		0 : IB_SEND_SIGNALED;
397 	imm = rtrs_to_io_rsp_imm(id->msg_id, errno, need_inval);
398 	imm_wr.wr.next = NULL;
399 	if (always_invalidate) {
400 		struct ib_sge list;
401 		struct rtrs_msg_rkey_rsp *msg;
402 
403 		srv_mr = &srv_path->mrs[id->msg_id];
404 		rwr.wr.next = &imm_wr.wr;
405 		rwr.wr.opcode = IB_WR_REG_MR;
406 		rwr.wr.wr_cqe = &local_reg_cqe;
407 		rwr.wr.num_sge = 0;
408 		rwr.wr.send_flags = 0;
409 		rwr.mr = srv_mr->mr;
410 		rwr.key = srv_mr->mr->rkey;
411 		rwr.access = (IB_ACCESS_LOCAL_WRITE |
412 			      IB_ACCESS_REMOTE_WRITE);
413 		msg = srv_mr->iu->buf;
414 		msg->buf_id = cpu_to_le16(id->msg_id);
415 		msg->type = cpu_to_le16(RTRS_MSG_RKEY_RSP);
416 		msg->rkey = cpu_to_le32(srv_mr->mr->rkey);
417 
418 		list.addr   = srv_mr->iu->dma_addr;
419 		list.length = sizeof(*msg);
420 		list.lkey   = srv_path->s.dev->ib_pd->local_dma_lkey;
421 		imm_wr.wr.sg_list = &list;
422 		imm_wr.wr.num_sge = 1;
423 		imm_wr.wr.opcode = IB_WR_SEND_WITH_IMM;
424 		ib_dma_sync_single_for_device(srv_path->s.dev->ib_dev,
425 					      srv_mr->iu->dma_addr,
426 					      srv_mr->iu->size, DMA_TO_DEVICE);
427 	} else {
428 		imm_wr.wr.sg_list = NULL;
429 		imm_wr.wr.num_sge = 0;
430 		imm_wr.wr.opcode = IB_WR_RDMA_WRITE_WITH_IMM;
431 	}
432 	imm_wr.wr.send_flags = flags;
433 	imm_wr.wr.wr_cqe   = &io_comp_cqe;
434 
435 	imm_wr.wr.ex.imm_data = cpu_to_be32(imm);
436 
437 	err = ib_post_send(id->con->c.qp, wr, NULL);
438 	if (err)
439 		rtrs_err_rl(s, "Posting RDMA-Reply to QP failed, err: %d\n",
440 			     err);
441 
442 	return err;
443 }
444 
445 void close_path(struct rtrs_srv_path *srv_path)
446 {
447 	if (rtrs_srv_change_state(srv_path, RTRS_SRV_CLOSING))
448 		queue_work(rtrs_wq, &srv_path->close_work);
449 	WARN_ON(srv_path->state != RTRS_SRV_CLOSING);
450 }
451 
452 static inline const char *rtrs_srv_state_str(enum rtrs_srv_state state)
453 {
454 	switch (state) {
455 	case RTRS_SRV_CONNECTING:
456 		return "RTRS_SRV_CONNECTING";
457 	case RTRS_SRV_CONNECTED:
458 		return "RTRS_SRV_CONNECTED";
459 	case RTRS_SRV_CLOSING:
460 		return "RTRS_SRV_CLOSING";
461 	case RTRS_SRV_CLOSED:
462 		return "RTRS_SRV_CLOSED";
463 	default:
464 		return "UNKNOWN";
465 	}
466 }
467 
468 /**
469  * rtrs_srv_resp_rdma() - Finish an RDMA request
470  *
471  * @id:		Internal RTRS operation identifier
472  * @status:	Response Code sent to the other side for this operation.
473  *		0 = success, <=0 error
474  * Context: any
475  *
476  * Finish a RDMA operation. A message is sent to the client and the
477  * corresponding memory areas will be released.
478  */
479 bool rtrs_srv_resp_rdma(struct rtrs_srv_op *id, int status)
480 {
481 	struct rtrs_srv_path *srv_path;
482 	struct rtrs_srv_con *con;
483 	struct rtrs_path *s;
484 	int err;
485 
486 	if (WARN_ON(!id))
487 		return true;
488 
489 	con = id->con;
490 	s = con->c.path;
491 	srv_path = to_srv_path(s);
492 
493 	id->status = status;
494 
495 	if (srv_path->state != RTRS_SRV_CONNECTED) {
496 		rtrs_err_rl(s,
497 			    "Sending I/O response failed,  server path %s is disconnected, path state %s\n",
498 			    kobject_name(&srv_path->kobj),
499 			    rtrs_srv_state_str(srv_path->state));
500 		goto out;
501 	}
502 	if (always_invalidate) {
503 		struct rtrs_srv_mr *mr = &srv_path->mrs[id->msg_id];
504 
505 		ib_update_fast_reg_key(mr->mr, ib_inc_rkey(mr->mr->rkey));
506 	}
507 	if (atomic_sub_return(1, &con->c.sq_wr_avail) < 0) {
508 		rtrs_err(s, "IB send queue full: srv_path=%s cid=%d\n",
509 			 kobject_name(&srv_path->kobj),
510 			 con->c.cid);
511 		atomic_add(1, &con->c.sq_wr_avail);
512 		spin_lock(&con->rsp_wr_wait_lock);
513 		list_add_tail(&id->wait_list, &con->rsp_wr_wait_list);
514 		spin_unlock(&con->rsp_wr_wait_lock);
515 		return false;
516 	}
517 
518 	if (status || id->dir == WRITE || !id->rd_msg->sg_cnt)
519 		err = send_io_resp_imm(con, id, status);
520 	else
521 		err = rdma_write_sg(id);
522 
523 	if (err) {
524 		rtrs_err_rl(s, "IO response failed: %d: srv_path=%s\n", err,
525 			    kobject_name(&srv_path->kobj));
526 		close_path(srv_path);
527 	}
528 out:
529 	rtrs_srv_put_ops_ids(srv_path);
530 	return true;
531 }
532 EXPORT_SYMBOL(rtrs_srv_resp_rdma);
533 
534 /**
535  * rtrs_srv_set_sess_priv() - Set private pointer in rtrs_srv.
536  * @srv:	Session pointer
537  * @priv:	The private pointer that is associated with the session.
538  */
539 void rtrs_srv_set_sess_priv(struct rtrs_srv_sess *srv, void *priv)
540 {
541 	srv->priv = priv;
542 }
543 EXPORT_SYMBOL(rtrs_srv_set_sess_priv);
544 
545 static void unmap_cont_bufs(struct rtrs_srv_path *srv_path)
546 {
547 	int i;
548 
549 	for (i = 0; i < srv_path->mrs_num; i++) {
550 		struct rtrs_srv_mr *srv_mr;
551 
552 		srv_mr = &srv_path->mrs[i];
553 		rtrs_iu_free(srv_mr->iu, srv_path->s.dev->ib_dev, 1);
554 		ib_dereg_mr(srv_mr->mr);
555 		ib_dma_unmap_sg(srv_path->s.dev->ib_dev, srv_mr->sgt.sgl,
556 				srv_mr->sgt.nents, DMA_BIDIRECTIONAL);
557 		sg_free_table(&srv_mr->sgt);
558 	}
559 	kfree(srv_path->mrs);
560 }
561 
562 static int map_cont_bufs(struct rtrs_srv_path *srv_path)
563 {
564 	struct rtrs_srv_sess *srv = srv_path->srv;
565 	struct rtrs_path *ss = &srv_path->s;
566 	int i, err, mrs_num;
567 	unsigned int chunk_bits;
568 	int chunks_per_mr = 1;
569 	struct ib_mr *mr;
570 	struct sg_table *sgt;
571 
572 	/*
573 	 * Here we map queue_depth chunks to MR.  Firstly we have to
574 	 * figure out how many chunks can we map per MR.
575 	 */
576 	if (always_invalidate) {
577 		/*
578 		 * in order to do invalidate for each chunks of memory, we needs
579 		 * more memory regions.
580 		 */
581 		mrs_num = srv->queue_depth;
582 	} else {
583 		chunks_per_mr =
584 			srv_path->s.dev->ib_dev->attrs.max_fast_reg_page_list_len;
585 		mrs_num = DIV_ROUND_UP(srv->queue_depth, chunks_per_mr);
586 		chunks_per_mr = DIV_ROUND_UP(srv->queue_depth, mrs_num);
587 	}
588 
589 	srv_path->mrs = kcalloc(mrs_num, sizeof(*srv_path->mrs), GFP_KERNEL);
590 	if (!srv_path->mrs)
591 		return -ENOMEM;
592 
593 	for (srv_path->mrs_num = 0; srv_path->mrs_num < mrs_num;
594 	     srv_path->mrs_num++) {
595 		struct rtrs_srv_mr *srv_mr = &srv_path->mrs[srv_path->mrs_num];
596 		struct scatterlist *s;
597 		int nr, nr_sgt, chunks;
598 
599 		sgt = &srv_mr->sgt;
600 		chunks = chunks_per_mr * srv_path->mrs_num;
601 		if (!always_invalidate)
602 			chunks_per_mr = min_t(int, chunks_per_mr,
603 					      srv->queue_depth - chunks);
604 
605 		err = sg_alloc_table(sgt, chunks_per_mr, GFP_KERNEL);
606 		if (err)
607 			goto err;
608 
609 		for_each_sg(sgt->sgl, s, chunks_per_mr, i)
610 			sg_set_page(s, srv->chunks[chunks + i],
611 				    max_chunk_size, 0);
612 
613 		nr_sgt = ib_dma_map_sg(srv_path->s.dev->ib_dev, sgt->sgl,
614 				   sgt->nents, DMA_BIDIRECTIONAL);
615 		if (!nr_sgt) {
616 			err = -EINVAL;
617 			goto free_sg;
618 		}
619 		mr = ib_alloc_mr(srv_path->s.dev->ib_pd, IB_MR_TYPE_MEM_REG,
620 				 nr_sgt);
621 		if (IS_ERR(mr)) {
622 			err = PTR_ERR(mr);
623 			goto unmap_sg;
624 		}
625 		nr = ib_map_mr_sg(mr, sgt->sgl, nr_sgt,
626 				  NULL, max_chunk_size);
627 		if (nr != nr_sgt) {
628 			err = nr < 0 ? nr : -EINVAL;
629 			goto dereg_mr;
630 		}
631 
632 		if (always_invalidate) {
633 			srv_mr->iu = rtrs_iu_alloc(1,
634 					sizeof(struct rtrs_msg_rkey_rsp),
635 					GFP_KERNEL, srv_path->s.dev->ib_dev,
636 					DMA_TO_DEVICE, rtrs_srv_rdma_done);
637 			if (!srv_mr->iu) {
638 				err = -ENOMEM;
639 				rtrs_err(ss, "rtrs_iu_alloc(), err: %d\n", err);
640 				goto dereg_mr;
641 			}
642 		}
643 		/* Eventually dma addr for each chunk can be cached */
644 		for_each_sg(sgt->sgl, s, nr_sgt, i)
645 			srv_path->dma_addr[chunks + i] = sg_dma_address(s);
646 
647 		ib_update_fast_reg_key(mr, ib_inc_rkey(mr->rkey));
648 		srv_mr->mr = mr;
649 	}
650 
651 	chunk_bits = ilog2(srv->queue_depth - 1) + 1;
652 	srv_path->mem_bits = (MAX_IMM_PAYL_BITS - chunk_bits);
653 
654 	return 0;
655 
656 dereg_mr:
657 	ib_dereg_mr(mr);
658 unmap_sg:
659 	ib_dma_unmap_sg(srv_path->s.dev->ib_dev, sgt->sgl,
660 			sgt->nents, DMA_BIDIRECTIONAL);
661 free_sg:
662 	sg_free_table(sgt);
663 err:
664 	unmap_cont_bufs(srv_path);
665 
666 	return err;
667 }
668 
669 static void rtrs_srv_hb_err_handler(struct rtrs_con *c)
670 {
671 	close_path(to_srv_path(c->path));
672 }
673 
674 static void rtrs_srv_init_hb(struct rtrs_srv_path *srv_path)
675 {
676 	rtrs_init_hb(&srv_path->s, &io_comp_cqe,
677 		      RTRS_HB_INTERVAL_MS,
678 		      RTRS_HB_MISSED_MAX,
679 		      rtrs_srv_hb_err_handler,
680 		      rtrs_wq);
681 }
682 
683 static void rtrs_srv_start_hb(struct rtrs_srv_path *srv_path)
684 {
685 	rtrs_start_hb(&srv_path->s);
686 }
687 
688 static void rtrs_srv_stop_hb(struct rtrs_srv_path *srv_path)
689 {
690 	rtrs_stop_hb(&srv_path->s);
691 }
692 
693 static void rtrs_srv_info_rsp_done(struct ib_cq *cq, struct ib_wc *wc)
694 {
695 	struct rtrs_srv_con *con = to_srv_con(wc->qp->qp_context);
696 	struct rtrs_path *s = con->c.path;
697 	struct rtrs_srv_path *srv_path = to_srv_path(s);
698 	struct rtrs_iu *iu;
699 
700 	iu = container_of(wc->wr_cqe, struct rtrs_iu, cqe);
701 	rtrs_iu_free(iu, srv_path->s.dev->ib_dev, 1);
702 
703 	if (wc->status != IB_WC_SUCCESS) {
704 		rtrs_err(s, "Sess info response send failed: %s\n",
705 			  ib_wc_status_msg(wc->status));
706 		close_path(srv_path);
707 		return;
708 	}
709 	WARN_ON(wc->opcode != IB_WC_SEND);
710 }
711 
712 static void rtrs_srv_path_up(struct rtrs_srv_path *srv_path)
713 {
714 	struct rtrs_srv_sess *srv = srv_path->srv;
715 	struct rtrs_srv_ctx *ctx = srv->ctx;
716 	int up;
717 
718 	mutex_lock(&srv->paths_ev_mutex);
719 	up = ++srv->paths_up;
720 	if (up == 1)
721 		ctx->ops.link_ev(srv, RTRS_SRV_LINK_EV_CONNECTED, NULL);
722 	mutex_unlock(&srv->paths_ev_mutex);
723 
724 	/* Mark session as established */
725 	srv_path->established = true;
726 }
727 
728 static void rtrs_srv_path_down(struct rtrs_srv_path *srv_path)
729 {
730 	struct rtrs_srv_sess *srv = srv_path->srv;
731 	struct rtrs_srv_ctx *ctx = srv->ctx;
732 
733 	if (!srv_path->established)
734 		return;
735 
736 	srv_path->established = false;
737 	mutex_lock(&srv->paths_ev_mutex);
738 	WARN_ON(!srv->paths_up);
739 	if (--srv->paths_up == 0)
740 		ctx->ops.link_ev(srv, RTRS_SRV_LINK_EV_DISCONNECTED, srv->priv);
741 	mutex_unlock(&srv->paths_ev_mutex);
742 }
743 
744 static bool exist_pathname(struct rtrs_srv_ctx *ctx,
745 			   const char *pathname, const uuid_t *path_uuid)
746 {
747 	struct rtrs_srv_sess *srv;
748 	struct rtrs_srv_path *srv_path;
749 	bool found = false;
750 
751 	mutex_lock(&ctx->srv_mutex);
752 	list_for_each_entry(srv, &ctx->srv_list, ctx_list) {
753 		mutex_lock(&srv->paths_mutex);
754 
755 		/* when a client with same uuid and same sessname tried to add a path */
756 		if (uuid_equal(&srv->paths_uuid, path_uuid)) {
757 			mutex_unlock(&srv->paths_mutex);
758 			continue;
759 		}
760 
761 		list_for_each_entry(srv_path, &srv->paths_list, s.entry) {
762 			if (strlen(srv_path->s.sessname) == strlen(pathname) &&
763 			    !strcmp(srv_path->s.sessname, pathname)) {
764 				found = true;
765 				break;
766 			}
767 		}
768 		mutex_unlock(&srv->paths_mutex);
769 		if (found)
770 			break;
771 	}
772 	mutex_unlock(&ctx->srv_mutex);
773 	return found;
774 }
775 
776 static int post_recv_path(struct rtrs_srv_path *srv_path);
777 static int rtrs_rdma_do_reject(struct rdma_cm_id *cm_id, int errno);
778 
779 static int process_info_req(struct rtrs_srv_con *con,
780 			    struct rtrs_msg_info_req *msg)
781 {
782 	struct rtrs_path *s = con->c.path;
783 	struct rtrs_srv_path *srv_path = to_srv_path(s);
784 	struct ib_send_wr *reg_wr = NULL;
785 	struct rtrs_msg_info_rsp *rsp;
786 	struct rtrs_iu *tx_iu;
787 	struct ib_reg_wr *rwr;
788 	int mri, err;
789 	size_t tx_sz;
790 
791 	err = post_recv_path(srv_path);
792 	if (err) {
793 		rtrs_err(s, "post_recv_path(), err: %d\n", err);
794 		return err;
795 	}
796 
797 	if (strchr(msg->pathname, '/') || strchr(msg->pathname, '.')) {
798 		rtrs_err(s, "pathname cannot contain / and .\n");
799 		return -EINVAL;
800 	}
801 
802 	if (exist_pathname(srv_path->srv->ctx,
803 			   msg->pathname, &srv_path->srv->paths_uuid)) {
804 		rtrs_err(s, "pathname is duplicated: %s\n", msg->pathname);
805 		return -EPERM;
806 	}
807 	strscpy(srv_path->s.sessname, msg->pathname,
808 		sizeof(srv_path->s.sessname));
809 
810 	rwr = kcalloc(srv_path->mrs_num, sizeof(*rwr), GFP_KERNEL);
811 	if (!rwr)
812 		return -ENOMEM;
813 
814 	tx_sz  = sizeof(*rsp);
815 	tx_sz += sizeof(rsp->desc[0]) * srv_path->mrs_num;
816 	tx_iu = rtrs_iu_alloc(1, tx_sz, GFP_KERNEL, srv_path->s.dev->ib_dev,
817 			       DMA_TO_DEVICE, rtrs_srv_info_rsp_done);
818 	if (!tx_iu) {
819 		err = -ENOMEM;
820 		goto rwr_free;
821 	}
822 
823 	rsp = tx_iu->buf;
824 	rsp->type = cpu_to_le16(RTRS_MSG_INFO_RSP);
825 	rsp->sg_cnt = cpu_to_le16(srv_path->mrs_num);
826 
827 	for (mri = 0; mri < srv_path->mrs_num; mri++) {
828 		struct ib_mr *mr = srv_path->mrs[mri].mr;
829 
830 		rsp->desc[mri].addr = cpu_to_le64(mr->iova);
831 		rsp->desc[mri].key  = cpu_to_le32(mr->rkey);
832 		rsp->desc[mri].len  = cpu_to_le32(mr->length);
833 
834 		/*
835 		 * Fill in reg MR request and chain them *backwards*
836 		 */
837 		rwr[mri].wr.next = mri ? &rwr[mri - 1].wr : NULL;
838 		rwr[mri].wr.opcode = IB_WR_REG_MR;
839 		rwr[mri].wr.wr_cqe = &local_reg_cqe;
840 		rwr[mri].wr.num_sge = 0;
841 		rwr[mri].wr.send_flags = 0;
842 		rwr[mri].mr = mr;
843 		rwr[mri].key = mr->rkey;
844 		rwr[mri].access = (IB_ACCESS_LOCAL_WRITE |
845 				   IB_ACCESS_REMOTE_WRITE);
846 		reg_wr = &rwr[mri].wr;
847 	}
848 
849 	err = rtrs_srv_create_path_files(srv_path);
850 	if (err)
851 		goto iu_free;
852 	kobject_get(&srv_path->kobj);
853 	get_device(&srv_path->srv->dev);
854 	rtrs_srv_change_state(srv_path, RTRS_SRV_CONNECTED);
855 	rtrs_srv_start_hb(srv_path);
856 
857 	/*
858 	 * We do not account number of established connections at the current
859 	 * moment, we rely on the client, which should send info request when
860 	 * all connections are successfully established.  Thus, simply notify
861 	 * listener with a proper event if we are the first path.
862 	 */
863 	rtrs_srv_path_up(srv_path);
864 
865 	ib_dma_sync_single_for_device(srv_path->s.dev->ib_dev,
866 				      tx_iu->dma_addr,
867 				      tx_iu->size, DMA_TO_DEVICE);
868 
869 	/* Send info response */
870 	err = rtrs_iu_post_send(&con->c, tx_iu, tx_sz, reg_wr);
871 	if (err) {
872 		rtrs_err(s, "rtrs_iu_post_send(), err: %d\n", err);
873 iu_free:
874 		rtrs_iu_free(tx_iu, srv_path->s.dev->ib_dev, 1);
875 	}
876 rwr_free:
877 	kfree(rwr);
878 
879 	return err;
880 }
881 
882 static void rtrs_srv_info_req_done(struct ib_cq *cq, struct ib_wc *wc)
883 {
884 	struct rtrs_srv_con *con = to_srv_con(wc->qp->qp_context);
885 	struct rtrs_path *s = con->c.path;
886 	struct rtrs_srv_path *srv_path = to_srv_path(s);
887 	struct rtrs_msg_info_req *msg;
888 	struct rtrs_iu *iu;
889 	int err;
890 
891 	WARN_ON(con->c.cid);
892 
893 	iu = container_of(wc->wr_cqe, struct rtrs_iu, cqe);
894 	if (wc->status != IB_WC_SUCCESS) {
895 		rtrs_err(s, "Sess info request receive failed: %s\n",
896 			  ib_wc_status_msg(wc->status));
897 		goto close;
898 	}
899 	WARN_ON(wc->opcode != IB_WC_RECV);
900 
901 	if (wc->byte_len < sizeof(*msg)) {
902 		rtrs_err(s, "Sess info request is malformed: size %d\n",
903 			  wc->byte_len);
904 		goto close;
905 	}
906 	ib_dma_sync_single_for_cpu(srv_path->s.dev->ib_dev, iu->dma_addr,
907 				   iu->size, DMA_FROM_DEVICE);
908 	msg = iu->buf;
909 	if (le16_to_cpu(msg->type) != RTRS_MSG_INFO_REQ) {
910 		rtrs_err(s, "Sess info request is malformed: type %d\n",
911 			  le16_to_cpu(msg->type));
912 		goto close;
913 	}
914 	err = process_info_req(con, msg);
915 	if (err)
916 		goto close;
917 
918 out:
919 	rtrs_iu_free(iu, srv_path->s.dev->ib_dev, 1);
920 	return;
921 close:
922 	close_path(srv_path);
923 	goto out;
924 }
925 
926 static int post_recv_info_req(struct rtrs_srv_con *con)
927 {
928 	struct rtrs_path *s = con->c.path;
929 	struct rtrs_srv_path *srv_path = to_srv_path(s);
930 	struct rtrs_iu *rx_iu;
931 	int err;
932 
933 	rx_iu = rtrs_iu_alloc(1, sizeof(struct rtrs_msg_info_req),
934 			       GFP_KERNEL, srv_path->s.dev->ib_dev,
935 			       DMA_FROM_DEVICE, rtrs_srv_info_req_done);
936 	if (!rx_iu)
937 		return -ENOMEM;
938 	/* Prepare for getting info response */
939 	err = rtrs_iu_post_recv(&con->c, rx_iu);
940 	if (err) {
941 		rtrs_err(s, "rtrs_iu_post_recv(), err: %d\n", err);
942 		rtrs_iu_free(rx_iu, srv_path->s.dev->ib_dev, 1);
943 		return err;
944 	}
945 
946 	return 0;
947 }
948 
949 static int post_recv_io(struct rtrs_srv_con *con, size_t q_size)
950 {
951 	int i, err;
952 
953 	for (i = 0; i < q_size; i++) {
954 		err = rtrs_post_recv_empty(&con->c, &io_comp_cqe);
955 		if (err)
956 			return err;
957 	}
958 
959 	return 0;
960 }
961 
962 static int post_recv_path(struct rtrs_srv_path *srv_path)
963 {
964 	struct rtrs_srv_sess *srv = srv_path->srv;
965 	struct rtrs_path *s = &srv_path->s;
966 	size_t q_size;
967 	int err, cid;
968 
969 	for (cid = 0; cid < srv_path->s.con_num; cid++) {
970 		if (cid == 0)
971 			q_size = SERVICE_CON_QUEUE_DEPTH;
972 		else
973 			q_size = srv->queue_depth;
974 
975 		err = post_recv_io(to_srv_con(srv_path->s.con[cid]), q_size);
976 		if (err) {
977 			rtrs_err(s, "post_recv_io(), err: %d\n", err);
978 			return err;
979 		}
980 	}
981 
982 	return 0;
983 }
984 
985 static void process_read(struct rtrs_srv_con *con,
986 			 struct rtrs_msg_rdma_read *msg,
987 			 u32 buf_id, u32 off)
988 {
989 	struct rtrs_path *s = con->c.path;
990 	struct rtrs_srv_path *srv_path = to_srv_path(s);
991 	struct rtrs_srv_sess *srv = srv_path->srv;
992 	struct rtrs_srv_ctx *ctx = srv->ctx;
993 	struct rtrs_srv_op *id;
994 
995 	size_t usr_len, data_len;
996 	void *data;
997 	int ret;
998 
999 	if (srv_path->state != RTRS_SRV_CONNECTED) {
1000 		rtrs_err_rl(s,
1001 			     "Processing read request failed,  session is disconnected, sess state %s\n",
1002 			     rtrs_srv_state_str(srv_path->state));
1003 		return;
1004 	}
1005 	if (msg->sg_cnt != 1 && msg->sg_cnt != 0) {
1006 		rtrs_err_rl(s,
1007 			    "Processing read request failed, invalid message\n");
1008 		return;
1009 	}
1010 	rtrs_srv_get_ops_ids(srv_path);
1011 	rtrs_srv_update_rdma_stats(srv_path->stats, off, READ);
1012 	id = srv_path->ops_ids[buf_id];
1013 	id->con		= con;
1014 	id->dir		= READ;
1015 	id->msg_id	= buf_id;
1016 	id->rd_msg	= msg;
1017 	usr_len = le16_to_cpu(msg->usr_len);
1018 	data_len = off - usr_len;
1019 	data = page_address(srv->chunks[buf_id]);
1020 	ret = ctx->ops.rdma_ev(srv->priv, id, data, data_len,
1021 			   data + data_len, usr_len);
1022 
1023 	if (ret) {
1024 		rtrs_err_rl(s,
1025 			     "Processing read request failed, user module cb reported for msg_id %d, err: %d\n",
1026 			     buf_id, ret);
1027 		goto send_err_msg;
1028 	}
1029 
1030 	return;
1031 
1032 send_err_msg:
1033 	ret = send_io_resp_imm(con, id, ret);
1034 	if (ret < 0) {
1035 		rtrs_err_rl(s,
1036 			     "Sending err msg for failed RDMA-Write-Req failed, msg_id %d, err: %d\n",
1037 			     buf_id, ret);
1038 		close_path(srv_path);
1039 	}
1040 	rtrs_srv_put_ops_ids(srv_path);
1041 }
1042 
1043 static void process_write(struct rtrs_srv_con *con,
1044 			  struct rtrs_msg_rdma_write *req,
1045 			  u32 buf_id, u32 off)
1046 {
1047 	struct rtrs_path *s = con->c.path;
1048 	struct rtrs_srv_path *srv_path = to_srv_path(s);
1049 	struct rtrs_srv_sess *srv = srv_path->srv;
1050 	struct rtrs_srv_ctx *ctx = srv->ctx;
1051 	struct rtrs_srv_op *id;
1052 
1053 	size_t data_len, usr_len;
1054 	void *data;
1055 	int ret;
1056 
1057 	if (srv_path->state != RTRS_SRV_CONNECTED) {
1058 		rtrs_err_rl(s,
1059 			     "Processing write request failed,  session is disconnected, sess state %s\n",
1060 			     rtrs_srv_state_str(srv_path->state));
1061 		return;
1062 	}
1063 	rtrs_srv_get_ops_ids(srv_path);
1064 	rtrs_srv_update_rdma_stats(srv_path->stats, off, WRITE);
1065 	id = srv_path->ops_ids[buf_id];
1066 	id->con    = con;
1067 	id->dir    = WRITE;
1068 	id->msg_id = buf_id;
1069 
1070 	usr_len = le16_to_cpu(req->usr_len);
1071 	data_len = off - usr_len;
1072 	data = page_address(srv->chunks[buf_id]);
1073 	ret = ctx->ops.rdma_ev(srv->priv, id, data, data_len,
1074 			       data + data_len, usr_len);
1075 	if (ret) {
1076 		rtrs_err_rl(s,
1077 			     "Processing write request failed, user module callback reports err: %d\n",
1078 			     ret);
1079 		goto send_err_msg;
1080 	}
1081 
1082 	return;
1083 
1084 send_err_msg:
1085 	ret = send_io_resp_imm(con, id, ret);
1086 	if (ret < 0) {
1087 		rtrs_err_rl(s,
1088 			     "Processing write request failed, sending I/O response failed, msg_id %d, err: %d\n",
1089 			     buf_id, ret);
1090 		close_path(srv_path);
1091 	}
1092 	rtrs_srv_put_ops_ids(srv_path);
1093 }
1094 
1095 static void process_io_req(struct rtrs_srv_con *con, void *msg,
1096 			   u32 id, u32 off)
1097 {
1098 	struct rtrs_path *s = con->c.path;
1099 	struct rtrs_srv_path *srv_path = to_srv_path(s);
1100 	struct rtrs_msg_rdma_hdr *hdr;
1101 	unsigned int type;
1102 
1103 	ib_dma_sync_single_for_cpu(srv_path->s.dev->ib_dev,
1104 				   srv_path->dma_addr[id],
1105 				   max_chunk_size, DMA_BIDIRECTIONAL);
1106 	hdr = msg;
1107 	type = le16_to_cpu(hdr->type);
1108 
1109 	switch (type) {
1110 	case RTRS_MSG_WRITE:
1111 		process_write(con, msg, id, off);
1112 		break;
1113 	case RTRS_MSG_READ:
1114 		process_read(con, msg, id, off);
1115 		break;
1116 	default:
1117 		rtrs_err(s,
1118 			  "Processing I/O request failed, unknown message type received: 0x%02x\n",
1119 			  type);
1120 		goto err;
1121 	}
1122 
1123 	return;
1124 
1125 err:
1126 	close_path(srv_path);
1127 }
1128 
1129 static void rtrs_srv_inv_rkey_done(struct ib_cq *cq, struct ib_wc *wc)
1130 {
1131 	struct rtrs_srv_mr *mr =
1132 		container_of(wc->wr_cqe, typeof(*mr), inv_cqe);
1133 	struct rtrs_srv_con *con = to_srv_con(wc->qp->qp_context);
1134 	struct rtrs_path *s = con->c.path;
1135 	struct rtrs_srv_path *srv_path = to_srv_path(s);
1136 	struct rtrs_srv_sess *srv = srv_path->srv;
1137 	u32 msg_id, off;
1138 	void *data;
1139 
1140 	if (wc->status != IB_WC_SUCCESS) {
1141 		rtrs_err(s, "Failed IB_WR_LOCAL_INV: %s\n",
1142 			  ib_wc_status_msg(wc->status));
1143 		close_path(srv_path);
1144 	}
1145 	msg_id = mr->msg_id;
1146 	off = mr->msg_off;
1147 	data = page_address(srv->chunks[msg_id]) + off;
1148 	process_io_req(con, data, msg_id, off);
1149 }
1150 
1151 static int rtrs_srv_inv_rkey(struct rtrs_srv_con *con,
1152 			      struct rtrs_srv_mr *mr)
1153 {
1154 	struct ib_send_wr wr = {
1155 		.opcode		    = IB_WR_LOCAL_INV,
1156 		.wr_cqe		    = &mr->inv_cqe,
1157 		.send_flags	    = IB_SEND_SIGNALED,
1158 		.ex.invalidate_rkey = mr->mr->rkey,
1159 	};
1160 	mr->inv_cqe.done = rtrs_srv_inv_rkey_done;
1161 
1162 	return ib_post_send(con->c.qp, &wr, NULL);
1163 }
1164 
1165 static void rtrs_rdma_process_wr_wait_list(struct rtrs_srv_con *con)
1166 {
1167 	spin_lock(&con->rsp_wr_wait_lock);
1168 	while (!list_empty(&con->rsp_wr_wait_list)) {
1169 		struct rtrs_srv_op *id;
1170 		int ret;
1171 
1172 		id = list_entry(con->rsp_wr_wait_list.next,
1173 				struct rtrs_srv_op, wait_list);
1174 		list_del(&id->wait_list);
1175 
1176 		spin_unlock(&con->rsp_wr_wait_lock);
1177 		ret = rtrs_srv_resp_rdma(id, id->status);
1178 		spin_lock(&con->rsp_wr_wait_lock);
1179 
1180 		if (!ret) {
1181 			list_add(&id->wait_list, &con->rsp_wr_wait_list);
1182 			break;
1183 		}
1184 	}
1185 	spin_unlock(&con->rsp_wr_wait_lock);
1186 }
1187 
1188 static void rtrs_srv_rdma_done(struct ib_cq *cq, struct ib_wc *wc)
1189 {
1190 	struct rtrs_srv_con *con = to_srv_con(wc->qp->qp_context);
1191 	struct rtrs_path *s = con->c.path;
1192 	struct rtrs_srv_path *srv_path = to_srv_path(s);
1193 	struct rtrs_srv_sess *srv = srv_path->srv;
1194 	u32 imm_type, imm_payload;
1195 	int err;
1196 
1197 	if (wc->status != IB_WC_SUCCESS) {
1198 		if (wc->status != IB_WC_WR_FLUSH_ERR) {
1199 			rtrs_err(s,
1200 				  "%s (wr_cqe: %p, type: %d, vendor_err: 0x%x, len: %u)\n",
1201 				  ib_wc_status_msg(wc->status), wc->wr_cqe,
1202 				  wc->opcode, wc->vendor_err, wc->byte_len);
1203 			close_path(srv_path);
1204 		}
1205 		return;
1206 	}
1207 
1208 	switch (wc->opcode) {
1209 	case IB_WC_RECV_RDMA_WITH_IMM:
1210 		/*
1211 		 * post_recv() RDMA write completions of IO reqs (read/write)
1212 		 * and hb
1213 		 */
1214 		if (WARN_ON(wc->wr_cqe != &io_comp_cqe))
1215 			return;
1216 		err = rtrs_post_recv_empty(&con->c, &io_comp_cqe);
1217 		if (err) {
1218 			rtrs_err(s, "rtrs_post_recv(), err: %d\n", err);
1219 			close_path(srv_path);
1220 			break;
1221 		}
1222 		rtrs_from_imm(be32_to_cpu(wc->ex.imm_data),
1223 			       &imm_type, &imm_payload);
1224 		if (imm_type == RTRS_IO_REQ_IMM) {
1225 			u32 msg_id, off;
1226 			void *data;
1227 
1228 			msg_id = imm_payload >> srv_path->mem_bits;
1229 			off = imm_payload & ((1 << srv_path->mem_bits) - 1);
1230 			if (msg_id >= srv->queue_depth || off >= max_chunk_size) {
1231 				rtrs_err(s, "Wrong msg_id %u, off %u\n",
1232 					  msg_id, off);
1233 				close_path(srv_path);
1234 				return;
1235 			}
1236 			if (always_invalidate) {
1237 				struct rtrs_srv_mr *mr = &srv_path->mrs[msg_id];
1238 
1239 				mr->msg_off = off;
1240 				mr->msg_id = msg_id;
1241 				err = rtrs_srv_inv_rkey(con, mr);
1242 				if (err) {
1243 					rtrs_err(s, "rtrs_post_recv(), err: %d\n",
1244 						  err);
1245 					close_path(srv_path);
1246 					break;
1247 				}
1248 			} else {
1249 				data = page_address(srv->chunks[msg_id]) + off;
1250 				process_io_req(con, data, msg_id, off);
1251 			}
1252 		} else if (imm_type == RTRS_HB_MSG_IMM) {
1253 			WARN_ON(con->c.cid);
1254 			rtrs_send_hb_ack(&srv_path->s);
1255 		} else if (imm_type == RTRS_HB_ACK_IMM) {
1256 			WARN_ON(con->c.cid);
1257 			srv_path->s.hb_missed_cnt = 0;
1258 		} else {
1259 			rtrs_wrn(s, "Unknown IMM type %u\n", imm_type);
1260 		}
1261 		break;
1262 	case IB_WC_RDMA_WRITE:
1263 	case IB_WC_SEND:
1264 		/*
1265 		 * post_send() RDMA write completions of IO reqs (read/write)
1266 		 * and hb.
1267 		 */
1268 		atomic_add(s->signal_interval, &con->c.sq_wr_avail);
1269 
1270 		if (!list_empty_careful(&con->rsp_wr_wait_list))
1271 			rtrs_rdma_process_wr_wait_list(con);
1272 
1273 		break;
1274 	default:
1275 		rtrs_wrn(s, "Unexpected WC type: %d\n", wc->opcode);
1276 		return;
1277 	}
1278 }
1279 
1280 /**
1281  * rtrs_srv_get_path_name() - Get rtrs_srv peer hostname.
1282  * @srv:	Session
1283  * @pathname:	Pathname buffer
1284  * @len:	Length of sessname buffer
1285  */
1286 int rtrs_srv_get_path_name(struct rtrs_srv_sess *srv, char *pathname,
1287 			   size_t len)
1288 {
1289 	struct rtrs_srv_path *srv_path;
1290 	int err = -ENOTCONN;
1291 
1292 	mutex_lock(&srv->paths_mutex);
1293 	list_for_each_entry(srv_path, &srv->paths_list, s.entry) {
1294 		if (srv_path->state != RTRS_SRV_CONNECTED)
1295 			continue;
1296 		strscpy(pathname, srv_path->s.sessname,
1297 			min_t(size_t, sizeof(srv_path->s.sessname), len));
1298 		err = 0;
1299 		break;
1300 	}
1301 	mutex_unlock(&srv->paths_mutex);
1302 
1303 	return err;
1304 }
1305 EXPORT_SYMBOL(rtrs_srv_get_path_name);
1306 
1307 /**
1308  * rtrs_srv_get_queue_depth() - Get rtrs_srv qdepth.
1309  * @srv:	Session
1310  */
1311 int rtrs_srv_get_queue_depth(struct rtrs_srv_sess *srv)
1312 {
1313 	return srv->queue_depth;
1314 }
1315 EXPORT_SYMBOL(rtrs_srv_get_queue_depth);
1316 
1317 static int find_next_bit_ring(struct rtrs_srv_path *srv_path)
1318 {
1319 	struct ib_device *ib_dev = srv_path->s.dev->ib_dev;
1320 	int v;
1321 
1322 	v = cpumask_next(srv_path->cur_cq_vector, &cq_affinity_mask);
1323 	if (v >= nr_cpu_ids || v >= ib_dev->num_comp_vectors)
1324 		v = cpumask_first(&cq_affinity_mask);
1325 	return v;
1326 }
1327 
1328 static int rtrs_srv_get_next_cq_vector(struct rtrs_srv_path *srv_path)
1329 {
1330 	srv_path->cur_cq_vector = find_next_bit_ring(srv_path);
1331 
1332 	return srv_path->cur_cq_vector;
1333 }
1334 
1335 static void rtrs_srv_dev_release(struct device *dev)
1336 {
1337 	struct rtrs_srv_sess *srv = container_of(dev, struct rtrs_srv_sess,
1338 						 dev);
1339 
1340 	kfree(srv);
1341 }
1342 
1343 static void free_srv(struct rtrs_srv_sess *srv)
1344 {
1345 	int i;
1346 
1347 	WARN_ON(refcount_read(&srv->refcount));
1348 	for (i = 0; i < srv->queue_depth; i++)
1349 		__free_pages(srv->chunks[i], get_order(max_chunk_size));
1350 	kfree(srv->chunks);
1351 	mutex_destroy(&srv->paths_mutex);
1352 	mutex_destroy(&srv->paths_ev_mutex);
1353 	/* last put to release the srv structure */
1354 	put_device(&srv->dev);
1355 }
1356 
1357 static struct rtrs_srv_sess *get_or_create_srv(struct rtrs_srv_ctx *ctx,
1358 					  const uuid_t *paths_uuid,
1359 					  bool first_conn)
1360 {
1361 	struct rtrs_srv_sess *srv;
1362 	int i;
1363 
1364 	mutex_lock(&ctx->srv_mutex);
1365 	list_for_each_entry(srv, &ctx->srv_list, ctx_list) {
1366 		if (uuid_equal(&srv->paths_uuid, paths_uuid) &&
1367 		    refcount_inc_not_zero(&srv->refcount)) {
1368 			mutex_unlock(&ctx->srv_mutex);
1369 			return srv;
1370 		}
1371 	}
1372 	mutex_unlock(&ctx->srv_mutex);
1373 	/*
1374 	 * If this request is not the first connection request from the
1375 	 * client for this session then fail and return error.
1376 	 */
1377 	if (!first_conn) {
1378 		pr_err_ratelimited("Error: Not the first connection request for this session\n");
1379 		return ERR_PTR(-ENXIO);
1380 	}
1381 
1382 	/* need to allocate a new srv */
1383 	srv = kzalloc(sizeof(*srv), GFP_KERNEL);
1384 	if  (!srv)
1385 		return ERR_PTR(-ENOMEM);
1386 
1387 	INIT_LIST_HEAD(&srv->paths_list);
1388 	mutex_init(&srv->paths_mutex);
1389 	mutex_init(&srv->paths_ev_mutex);
1390 	uuid_copy(&srv->paths_uuid, paths_uuid);
1391 	srv->queue_depth = sess_queue_depth;
1392 	srv->ctx = ctx;
1393 	device_initialize(&srv->dev);
1394 	srv->dev.release = rtrs_srv_dev_release;
1395 
1396 	srv->chunks = kcalloc(srv->queue_depth, sizeof(*srv->chunks),
1397 			      GFP_KERNEL);
1398 	if (!srv->chunks)
1399 		goto err_free_srv;
1400 
1401 	for (i = 0; i < srv->queue_depth; i++) {
1402 		srv->chunks[i] = alloc_pages(GFP_KERNEL,
1403 					     get_order(max_chunk_size));
1404 		if (!srv->chunks[i])
1405 			goto err_free_chunks;
1406 	}
1407 	refcount_set(&srv->refcount, 1);
1408 	mutex_lock(&ctx->srv_mutex);
1409 	list_add(&srv->ctx_list, &ctx->srv_list);
1410 	mutex_unlock(&ctx->srv_mutex);
1411 
1412 	return srv;
1413 
1414 err_free_chunks:
1415 	while (i--)
1416 		__free_pages(srv->chunks[i], get_order(max_chunk_size));
1417 	kfree(srv->chunks);
1418 
1419 err_free_srv:
1420 	kfree(srv);
1421 	return ERR_PTR(-ENOMEM);
1422 }
1423 
1424 static void put_srv(struct rtrs_srv_sess *srv)
1425 {
1426 	if (refcount_dec_and_test(&srv->refcount)) {
1427 		struct rtrs_srv_ctx *ctx = srv->ctx;
1428 
1429 		WARN_ON(srv->dev.kobj.state_in_sysfs);
1430 
1431 		mutex_lock(&ctx->srv_mutex);
1432 		list_del(&srv->ctx_list);
1433 		mutex_unlock(&ctx->srv_mutex);
1434 		free_srv(srv);
1435 	}
1436 }
1437 
1438 static void __add_path_to_srv(struct rtrs_srv_sess *srv,
1439 			      struct rtrs_srv_path *srv_path)
1440 {
1441 	list_add_tail(&srv_path->s.entry, &srv->paths_list);
1442 	srv->paths_num++;
1443 	WARN_ON(srv->paths_num >= MAX_PATHS_NUM);
1444 }
1445 
1446 static void del_path_from_srv(struct rtrs_srv_path *srv_path)
1447 {
1448 	struct rtrs_srv_sess *srv = srv_path->srv;
1449 
1450 	if (WARN_ON(!srv))
1451 		return;
1452 
1453 	mutex_lock(&srv->paths_mutex);
1454 	list_del(&srv_path->s.entry);
1455 	WARN_ON(!srv->paths_num);
1456 	srv->paths_num--;
1457 	mutex_unlock(&srv->paths_mutex);
1458 }
1459 
1460 /* return true if addresses are the same, error other wise */
1461 static int sockaddr_cmp(const struct sockaddr *a, const struct sockaddr *b)
1462 {
1463 	switch (a->sa_family) {
1464 	case AF_IB:
1465 		return memcmp(&((struct sockaddr_ib *)a)->sib_addr,
1466 			      &((struct sockaddr_ib *)b)->sib_addr,
1467 			      sizeof(struct ib_addr)) &&
1468 			(b->sa_family == AF_IB);
1469 	case AF_INET:
1470 		return memcmp(&((struct sockaddr_in *)a)->sin_addr,
1471 			      &((struct sockaddr_in *)b)->sin_addr,
1472 			      sizeof(struct in_addr)) &&
1473 			(b->sa_family == AF_INET);
1474 	case AF_INET6:
1475 		return memcmp(&((struct sockaddr_in6 *)a)->sin6_addr,
1476 			      &((struct sockaddr_in6 *)b)->sin6_addr,
1477 			      sizeof(struct in6_addr)) &&
1478 			(b->sa_family == AF_INET6);
1479 	default:
1480 		return -ENOENT;
1481 	}
1482 }
1483 
1484 static bool __is_path_w_addr_exists(struct rtrs_srv_sess *srv,
1485 				    struct rdma_addr *addr)
1486 {
1487 	struct rtrs_srv_path *srv_path;
1488 
1489 	list_for_each_entry(srv_path, &srv->paths_list, s.entry)
1490 		if (!sockaddr_cmp((struct sockaddr *)&srv_path->s.dst_addr,
1491 				  (struct sockaddr *)&addr->dst_addr) &&
1492 		    !sockaddr_cmp((struct sockaddr *)&srv_path->s.src_addr,
1493 				  (struct sockaddr *)&addr->src_addr))
1494 			return true;
1495 
1496 	return false;
1497 }
1498 
1499 static void free_path(struct rtrs_srv_path *srv_path)
1500 {
1501 	if (srv_path->kobj.state_in_sysfs) {
1502 		kobject_del(&srv_path->kobj);
1503 		kobject_put(&srv_path->kobj);
1504 	} else {
1505 		free_percpu(srv_path->stats->rdma_stats);
1506 		kfree(srv_path->stats);
1507 		kfree(srv_path);
1508 	}
1509 }
1510 
1511 static void rtrs_srv_close_work(struct work_struct *work)
1512 {
1513 	struct rtrs_srv_path *srv_path;
1514 	struct rtrs_srv_con *con;
1515 	int i;
1516 
1517 	srv_path = container_of(work, typeof(*srv_path), close_work);
1518 
1519 	rtrs_srv_destroy_path_files(srv_path);
1520 	rtrs_srv_stop_hb(srv_path);
1521 
1522 	for (i = 0; i < srv_path->s.con_num; i++) {
1523 		if (!srv_path->s.con[i])
1524 			continue;
1525 		con = to_srv_con(srv_path->s.con[i]);
1526 		rdma_disconnect(con->c.cm_id);
1527 		ib_drain_qp(con->c.qp);
1528 	}
1529 
1530 	/*
1531 	 * Degrade ref count to the usual model with a single shared
1532 	 * atomic_t counter
1533 	 */
1534 	percpu_ref_kill(&srv_path->ids_inflight_ref);
1535 
1536 	/* Wait for all completion */
1537 	wait_for_completion(&srv_path->complete_done);
1538 
1539 	/* Notify upper layer if we are the last path */
1540 	rtrs_srv_path_down(srv_path);
1541 
1542 	unmap_cont_bufs(srv_path);
1543 	rtrs_srv_free_ops_ids(srv_path);
1544 
1545 	for (i = 0; i < srv_path->s.con_num; i++) {
1546 		if (!srv_path->s.con[i])
1547 			continue;
1548 		con = to_srv_con(srv_path->s.con[i]);
1549 		rtrs_cq_qp_destroy(&con->c);
1550 		rdma_destroy_id(con->c.cm_id);
1551 		kfree(con);
1552 	}
1553 	rtrs_ib_dev_put(srv_path->s.dev);
1554 
1555 	del_path_from_srv(srv_path);
1556 	put_srv(srv_path->srv);
1557 	srv_path->srv = NULL;
1558 	rtrs_srv_change_state(srv_path, RTRS_SRV_CLOSED);
1559 
1560 	kfree(srv_path->dma_addr);
1561 	kfree(srv_path->s.con);
1562 	free_path(srv_path);
1563 }
1564 
1565 static int rtrs_rdma_do_accept(struct rtrs_srv_path *srv_path,
1566 			       struct rdma_cm_id *cm_id)
1567 {
1568 	struct rtrs_srv_sess *srv = srv_path->srv;
1569 	struct rtrs_msg_conn_rsp msg;
1570 	struct rdma_conn_param param;
1571 	int err;
1572 
1573 	param = (struct rdma_conn_param) {
1574 		.rnr_retry_count = 7,
1575 		.private_data = &msg,
1576 		.private_data_len = sizeof(msg),
1577 	};
1578 
1579 	msg = (struct rtrs_msg_conn_rsp) {
1580 		.magic = cpu_to_le16(RTRS_MAGIC),
1581 		.version = cpu_to_le16(RTRS_PROTO_VER),
1582 		.queue_depth = cpu_to_le16(srv->queue_depth),
1583 		.max_io_size = cpu_to_le32(max_chunk_size - MAX_HDR_SIZE),
1584 		.max_hdr_size = cpu_to_le32(MAX_HDR_SIZE),
1585 	};
1586 
1587 	if (always_invalidate)
1588 		msg.flags = cpu_to_le32(RTRS_MSG_NEW_RKEY_F);
1589 
1590 	err = rdma_accept(cm_id, &param);
1591 	if (err)
1592 		pr_err("rdma_accept(), err: %d\n", err);
1593 
1594 	return err;
1595 }
1596 
1597 static int rtrs_rdma_do_reject(struct rdma_cm_id *cm_id, int errno)
1598 {
1599 	struct rtrs_msg_conn_rsp msg;
1600 	int err;
1601 
1602 	msg = (struct rtrs_msg_conn_rsp) {
1603 		.magic = cpu_to_le16(RTRS_MAGIC),
1604 		.version = cpu_to_le16(RTRS_PROTO_VER),
1605 		.errno = cpu_to_le16(errno),
1606 	};
1607 
1608 	err = rdma_reject(cm_id, &msg, sizeof(msg), IB_CM_REJ_CONSUMER_DEFINED);
1609 	if (err)
1610 		pr_err("rdma_reject(), err: %d\n", err);
1611 
1612 	/* Bounce errno back */
1613 	return errno;
1614 }
1615 
1616 static struct rtrs_srv_path *
1617 __find_path(struct rtrs_srv_sess *srv, const uuid_t *sess_uuid)
1618 {
1619 	struct rtrs_srv_path *srv_path;
1620 
1621 	list_for_each_entry(srv_path, &srv->paths_list, s.entry) {
1622 		if (uuid_equal(&srv_path->s.uuid, sess_uuid))
1623 			return srv_path;
1624 	}
1625 
1626 	return NULL;
1627 }
1628 
1629 static int create_con(struct rtrs_srv_path *srv_path,
1630 		      struct rdma_cm_id *cm_id,
1631 		      unsigned int cid)
1632 {
1633 	struct rtrs_srv_sess *srv = srv_path->srv;
1634 	struct rtrs_path *s = &srv_path->s;
1635 	struct rtrs_srv_con *con;
1636 
1637 	u32 cq_num, max_send_wr, max_recv_wr, wr_limit;
1638 	int err, cq_vector;
1639 
1640 	con = kzalloc(sizeof(*con), GFP_KERNEL);
1641 	if (!con) {
1642 		err = -ENOMEM;
1643 		goto err;
1644 	}
1645 
1646 	spin_lock_init(&con->rsp_wr_wait_lock);
1647 	INIT_LIST_HEAD(&con->rsp_wr_wait_list);
1648 	con->c.cm_id = cm_id;
1649 	con->c.path = &srv_path->s;
1650 	con->c.cid = cid;
1651 	atomic_set(&con->c.wr_cnt, 1);
1652 	wr_limit = srv_path->s.dev->ib_dev->attrs.max_qp_wr;
1653 
1654 	if (con->c.cid == 0) {
1655 		/*
1656 		 * All receive and all send (each requiring invalidate)
1657 		 * + 2 for drain and heartbeat
1658 		 */
1659 		max_send_wr = min_t(int, wr_limit,
1660 				    SERVICE_CON_QUEUE_DEPTH * 2 + 2);
1661 		max_recv_wr = max_send_wr;
1662 		s->signal_interval = min_not_zero(srv->queue_depth,
1663 						  (size_t)SERVICE_CON_QUEUE_DEPTH);
1664 	} else {
1665 		/* when always_invlaidate enalbed, we need linv+rinv+mr+imm */
1666 		if (always_invalidate)
1667 			max_send_wr =
1668 				min_t(int, wr_limit,
1669 				      srv->queue_depth * (1 + 4) + 1);
1670 		else
1671 			max_send_wr =
1672 				min_t(int, wr_limit,
1673 				      srv->queue_depth * (1 + 2) + 1);
1674 
1675 		max_recv_wr = srv->queue_depth + 1;
1676 	}
1677 	cq_num = max_send_wr + max_recv_wr;
1678 	atomic_set(&con->c.sq_wr_avail, max_send_wr);
1679 	cq_vector = rtrs_srv_get_next_cq_vector(srv_path);
1680 
1681 	/* TODO: SOFTIRQ can be faster, but be careful with softirq context */
1682 	err = rtrs_cq_qp_create(&srv_path->s, &con->c, 1, cq_vector, cq_num,
1683 				 max_send_wr, max_recv_wr,
1684 				 IB_POLL_WORKQUEUE);
1685 	if (err) {
1686 		rtrs_err(s, "rtrs_cq_qp_create(), err: %d\n", err);
1687 		goto free_con;
1688 	}
1689 	if (con->c.cid == 0) {
1690 		err = post_recv_info_req(con);
1691 		if (err)
1692 			goto free_cqqp;
1693 	}
1694 	WARN_ON(srv_path->s.con[cid]);
1695 	srv_path->s.con[cid] = &con->c;
1696 
1697 	/*
1698 	 * Change context from server to current connection.  The other
1699 	 * way is to use cm_id->qp->qp_context, which does not work on OFED.
1700 	 */
1701 	cm_id->context = &con->c;
1702 
1703 	return 0;
1704 
1705 free_cqqp:
1706 	rtrs_cq_qp_destroy(&con->c);
1707 free_con:
1708 	kfree(con);
1709 
1710 err:
1711 	return err;
1712 }
1713 
1714 static struct rtrs_srv_path *__alloc_path(struct rtrs_srv_sess *srv,
1715 					   struct rdma_cm_id *cm_id,
1716 					   unsigned int con_num,
1717 					   unsigned int recon_cnt,
1718 					   const uuid_t *uuid)
1719 {
1720 	struct rtrs_srv_path *srv_path;
1721 	int err = -ENOMEM;
1722 	char str[NAME_MAX];
1723 	struct rtrs_addr path;
1724 
1725 	if (srv->paths_num >= MAX_PATHS_NUM) {
1726 		err = -ECONNRESET;
1727 		goto err;
1728 	}
1729 	if (__is_path_w_addr_exists(srv, &cm_id->route.addr)) {
1730 		err = -EEXIST;
1731 		pr_err("Path with same addr exists\n");
1732 		goto err;
1733 	}
1734 	srv_path = kzalloc(sizeof(*srv_path), GFP_KERNEL);
1735 	if (!srv_path)
1736 		goto err;
1737 
1738 	srv_path->stats = kzalloc(sizeof(*srv_path->stats), GFP_KERNEL);
1739 	if (!srv_path->stats)
1740 		goto err_free_sess;
1741 
1742 	srv_path->stats->rdma_stats = alloc_percpu(struct rtrs_srv_stats_rdma_stats);
1743 	if (!srv_path->stats->rdma_stats)
1744 		goto err_free_stats;
1745 
1746 	srv_path->stats->srv_path = srv_path;
1747 
1748 	srv_path->dma_addr = kcalloc(srv->queue_depth,
1749 				     sizeof(*srv_path->dma_addr),
1750 				     GFP_KERNEL);
1751 	if (!srv_path->dma_addr)
1752 		goto err_free_percpu;
1753 
1754 	srv_path->s.con = kcalloc(con_num, sizeof(*srv_path->s.con),
1755 				  GFP_KERNEL);
1756 	if (!srv_path->s.con)
1757 		goto err_free_dma_addr;
1758 
1759 	srv_path->state = RTRS_SRV_CONNECTING;
1760 	srv_path->srv = srv;
1761 	srv_path->cur_cq_vector = -1;
1762 	srv_path->s.dst_addr = cm_id->route.addr.dst_addr;
1763 	srv_path->s.src_addr = cm_id->route.addr.src_addr;
1764 
1765 	/* temporary until receiving session-name from client */
1766 	path.src = &srv_path->s.src_addr;
1767 	path.dst = &srv_path->s.dst_addr;
1768 	rtrs_addr_to_str(&path, str, sizeof(str));
1769 	strscpy(srv_path->s.sessname, str, sizeof(srv_path->s.sessname));
1770 
1771 	srv_path->s.con_num = con_num;
1772 	srv_path->s.irq_con_num = con_num;
1773 	srv_path->s.recon_cnt = recon_cnt;
1774 	uuid_copy(&srv_path->s.uuid, uuid);
1775 	spin_lock_init(&srv_path->state_lock);
1776 	INIT_WORK(&srv_path->close_work, rtrs_srv_close_work);
1777 	rtrs_srv_init_hb(srv_path);
1778 
1779 	srv_path->s.dev = rtrs_ib_dev_find_or_add(cm_id->device, &dev_pd);
1780 	if (!srv_path->s.dev) {
1781 		err = -ENOMEM;
1782 		goto err_free_con;
1783 	}
1784 	err = map_cont_bufs(srv_path);
1785 	if (err)
1786 		goto err_put_dev;
1787 
1788 	err = rtrs_srv_alloc_ops_ids(srv_path);
1789 	if (err)
1790 		goto err_unmap_bufs;
1791 
1792 	__add_path_to_srv(srv, srv_path);
1793 
1794 	return srv_path;
1795 
1796 err_unmap_bufs:
1797 	unmap_cont_bufs(srv_path);
1798 err_put_dev:
1799 	rtrs_ib_dev_put(srv_path->s.dev);
1800 err_free_con:
1801 	kfree(srv_path->s.con);
1802 err_free_dma_addr:
1803 	kfree(srv_path->dma_addr);
1804 err_free_percpu:
1805 	free_percpu(srv_path->stats->rdma_stats);
1806 err_free_stats:
1807 	kfree(srv_path->stats);
1808 err_free_sess:
1809 	kfree(srv_path);
1810 err:
1811 	return ERR_PTR(err);
1812 }
1813 
1814 static int rtrs_rdma_connect(struct rdma_cm_id *cm_id,
1815 			      const struct rtrs_msg_conn_req *msg,
1816 			      size_t len)
1817 {
1818 	struct rtrs_srv_ctx *ctx = cm_id->context;
1819 	struct rtrs_srv_path *srv_path;
1820 	struct rtrs_srv_sess *srv;
1821 
1822 	u16 version, con_num, cid;
1823 	u16 recon_cnt;
1824 	int err = -ECONNRESET;
1825 
1826 	if (len < sizeof(*msg)) {
1827 		pr_err("Invalid RTRS connection request\n");
1828 		goto reject_w_err;
1829 	}
1830 	if (le16_to_cpu(msg->magic) != RTRS_MAGIC) {
1831 		pr_err("Invalid RTRS magic\n");
1832 		goto reject_w_err;
1833 	}
1834 	version = le16_to_cpu(msg->version);
1835 	if (version >> 8 != RTRS_PROTO_VER_MAJOR) {
1836 		pr_err("Unsupported major RTRS version: %d, expected %d\n",
1837 		       version >> 8, RTRS_PROTO_VER_MAJOR);
1838 		goto reject_w_err;
1839 	}
1840 	con_num = le16_to_cpu(msg->cid_num);
1841 	if (con_num > 4096) {
1842 		/* Sanity check */
1843 		pr_err("Too many connections requested: %d\n", con_num);
1844 		goto reject_w_err;
1845 	}
1846 	cid = le16_to_cpu(msg->cid);
1847 	if (cid >= con_num) {
1848 		/* Sanity check */
1849 		pr_err("Incorrect cid: %d >= %d\n", cid, con_num);
1850 		goto reject_w_err;
1851 	}
1852 	recon_cnt = le16_to_cpu(msg->recon_cnt);
1853 	srv = get_or_create_srv(ctx, &msg->paths_uuid, msg->first_conn);
1854 	if (IS_ERR(srv)) {
1855 		err = PTR_ERR(srv);
1856 		pr_err("get_or_create_srv(), error %d\n", err);
1857 		goto reject_w_err;
1858 	}
1859 	mutex_lock(&srv->paths_mutex);
1860 	srv_path = __find_path(srv, &msg->sess_uuid);
1861 	if (srv_path) {
1862 		struct rtrs_path *s = &srv_path->s;
1863 
1864 		/* Session already holds a reference */
1865 		put_srv(srv);
1866 
1867 		if (srv_path->state != RTRS_SRV_CONNECTING) {
1868 			rtrs_err(s, "Session in wrong state: %s\n",
1869 				  rtrs_srv_state_str(srv_path->state));
1870 			mutex_unlock(&srv->paths_mutex);
1871 			goto reject_w_err;
1872 		}
1873 		/*
1874 		 * Sanity checks
1875 		 */
1876 		if (con_num != s->con_num || cid >= s->con_num) {
1877 			rtrs_err(s, "Incorrect request: %d, %d\n",
1878 				  cid, con_num);
1879 			mutex_unlock(&srv->paths_mutex);
1880 			goto reject_w_err;
1881 		}
1882 		if (s->con[cid]) {
1883 			rtrs_err(s, "Connection already exists: %d\n",
1884 				  cid);
1885 			mutex_unlock(&srv->paths_mutex);
1886 			goto reject_w_err;
1887 		}
1888 	} else {
1889 		srv_path = __alloc_path(srv, cm_id, con_num, recon_cnt,
1890 				    &msg->sess_uuid);
1891 		if (IS_ERR(srv_path)) {
1892 			mutex_unlock(&srv->paths_mutex);
1893 			put_srv(srv);
1894 			err = PTR_ERR(srv_path);
1895 			pr_err("RTRS server session allocation failed: %d\n", err);
1896 			goto reject_w_err;
1897 		}
1898 	}
1899 	err = create_con(srv_path, cm_id, cid);
1900 	if (err) {
1901 		rtrs_err((&srv_path->s), "create_con(), error %d\n", err);
1902 		rtrs_rdma_do_reject(cm_id, err);
1903 		/*
1904 		 * Since session has other connections we follow normal way
1905 		 * through workqueue, but still return an error to tell cma.c
1906 		 * to call rdma_destroy_id() for current connection.
1907 		 */
1908 		goto close_and_return_err;
1909 	}
1910 	err = rtrs_rdma_do_accept(srv_path, cm_id);
1911 	if (err) {
1912 		rtrs_err((&srv_path->s), "rtrs_rdma_do_accept(), error %d\n", err);
1913 		rtrs_rdma_do_reject(cm_id, err);
1914 		/*
1915 		 * Since current connection was successfully added to the
1916 		 * session we follow normal way through workqueue to close the
1917 		 * session, thus return 0 to tell cma.c we call
1918 		 * rdma_destroy_id() ourselves.
1919 		 */
1920 		err = 0;
1921 		goto close_and_return_err;
1922 	}
1923 	mutex_unlock(&srv->paths_mutex);
1924 
1925 	return 0;
1926 
1927 reject_w_err:
1928 	return rtrs_rdma_do_reject(cm_id, err);
1929 
1930 close_and_return_err:
1931 	mutex_unlock(&srv->paths_mutex);
1932 	close_path(srv_path);
1933 
1934 	return err;
1935 }
1936 
1937 static int rtrs_srv_rdma_cm_handler(struct rdma_cm_id *cm_id,
1938 				     struct rdma_cm_event *ev)
1939 {
1940 	struct rtrs_srv_path *srv_path = NULL;
1941 	struct rtrs_path *s = NULL;
1942 	struct rtrs_con *c = NULL;
1943 
1944 	if (ev->event == RDMA_CM_EVENT_CONNECT_REQUEST)
1945 		/*
1946 		 * In case of error cma.c will destroy cm_id,
1947 		 * see cma_process_remove()
1948 		 */
1949 		return rtrs_rdma_connect(cm_id, ev->param.conn.private_data,
1950 					  ev->param.conn.private_data_len);
1951 
1952 	c = cm_id->context;
1953 	s = c->path;
1954 	srv_path = to_srv_path(s);
1955 
1956 	switch (ev->event) {
1957 	case RDMA_CM_EVENT_ESTABLISHED:
1958 		/* Nothing here */
1959 		break;
1960 	case RDMA_CM_EVENT_REJECTED:
1961 	case RDMA_CM_EVENT_CONNECT_ERROR:
1962 	case RDMA_CM_EVENT_UNREACHABLE:
1963 		rtrs_err(s, "CM error (CM event: %s, err: %d)\n",
1964 			  rdma_event_msg(ev->event), ev->status);
1965 		fallthrough;
1966 	case RDMA_CM_EVENT_DISCONNECTED:
1967 	case RDMA_CM_EVENT_ADDR_CHANGE:
1968 	case RDMA_CM_EVENT_TIMEWAIT_EXIT:
1969 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
1970 		close_path(srv_path);
1971 		break;
1972 	default:
1973 		pr_err("Ignoring unexpected CM event %s, err %d\n",
1974 		       rdma_event_msg(ev->event), ev->status);
1975 		break;
1976 	}
1977 
1978 	return 0;
1979 }
1980 
1981 static struct rdma_cm_id *rtrs_srv_cm_init(struct rtrs_srv_ctx *ctx,
1982 					    struct sockaddr *addr,
1983 					    enum rdma_ucm_port_space ps)
1984 {
1985 	struct rdma_cm_id *cm_id;
1986 	int ret;
1987 
1988 	cm_id = rdma_create_id(&init_net, rtrs_srv_rdma_cm_handler,
1989 			       ctx, ps, IB_QPT_RC);
1990 	if (IS_ERR(cm_id)) {
1991 		ret = PTR_ERR(cm_id);
1992 		pr_err("Creating id for RDMA connection failed, err: %d\n",
1993 		       ret);
1994 		goto err_out;
1995 	}
1996 	ret = rdma_bind_addr(cm_id, addr);
1997 	if (ret) {
1998 		pr_err("Binding RDMA address failed, err: %d\n", ret);
1999 		goto err_cm;
2000 	}
2001 	ret = rdma_listen(cm_id, 64);
2002 	if (ret) {
2003 		pr_err("Listening on RDMA connection failed, err: %d\n",
2004 		       ret);
2005 		goto err_cm;
2006 	}
2007 
2008 	return cm_id;
2009 
2010 err_cm:
2011 	rdma_destroy_id(cm_id);
2012 err_out:
2013 
2014 	return ERR_PTR(ret);
2015 }
2016 
2017 static int rtrs_srv_rdma_init(struct rtrs_srv_ctx *ctx, u16 port)
2018 {
2019 	struct sockaddr_in6 sin = {
2020 		.sin6_family	= AF_INET6,
2021 		.sin6_addr	= IN6ADDR_ANY_INIT,
2022 		.sin6_port	= htons(port),
2023 	};
2024 	struct sockaddr_ib sib = {
2025 		.sib_family			= AF_IB,
2026 		.sib_sid	= cpu_to_be64(RDMA_IB_IP_PS_IB | port),
2027 		.sib_sid_mask	= cpu_to_be64(0xffffffffffffffffULL),
2028 		.sib_pkey	= cpu_to_be16(0xffff),
2029 	};
2030 	struct rdma_cm_id *cm_ip, *cm_ib;
2031 	int ret;
2032 
2033 	/*
2034 	 * We accept both IPoIB and IB connections, so we need to keep
2035 	 * two cm id's, one for each socket type and port space.
2036 	 * If the cm initialization of one of the id's fails, we abort
2037 	 * everything.
2038 	 */
2039 	cm_ip = rtrs_srv_cm_init(ctx, (struct sockaddr *)&sin, RDMA_PS_TCP);
2040 	if (IS_ERR(cm_ip))
2041 		return PTR_ERR(cm_ip);
2042 
2043 	cm_ib = rtrs_srv_cm_init(ctx, (struct sockaddr *)&sib, RDMA_PS_IB);
2044 	if (IS_ERR(cm_ib)) {
2045 		ret = PTR_ERR(cm_ib);
2046 		goto free_cm_ip;
2047 	}
2048 
2049 	ctx->cm_id_ip = cm_ip;
2050 	ctx->cm_id_ib = cm_ib;
2051 
2052 	return 0;
2053 
2054 free_cm_ip:
2055 	rdma_destroy_id(cm_ip);
2056 
2057 	return ret;
2058 }
2059 
2060 static struct rtrs_srv_ctx *alloc_srv_ctx(struct rtrs_srv_ops *ops)
2061 {
2062 	struct rtrs_srv_ctx *ctx;
2063 
2064 	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
2065 	if (!ctx)
2066 		return NULL;
2067 
2068 	ctx->ops = *ops;
2069 	mutex_init(&ctx->srv_mutex);
2070 	INIT_LIST_HEAD(&ctx->srv_list);
2071 
2072 	return ctx;
2073 }
2074 
2075 static void free_srv_ctx(struct rtrs_srv_ctx *ctx)
2076 {
2077 	WARN_ON(!list_empty(&ctx->srv_list));
2078 	mutex_destroy(&ctx->srv_mutex);
2079 	kfree(ctx);
2080 }
2081 
2082 static int rtrs_srv_add_one(struct ib_device *device)
2083 {
2084 	struct rtrs_srv_ctx *ctx;
2085 	int ret = 0;
2086 
2087 	mutex_lock(&ib_ctx.ib_dev_mutex);
2088 	if (ib_ctx.ib_dev_count)
2089 		goto out;
2090 
2091 	/*
2092 	 * Since our CM IDs are NOT bound to any ib device we will create them
2093 	 * only once
2094 	 */
2095 	ctx = ib_ctx.srv_ctx;
2096 	ret = rtrs_srv_rdma_init(ctx, ib_ctx.port);
2097 	if (ret) {
2098 		/*
2099 		 * We errored out here.
2100 		 * According to the ib code, if we encounter an error here then the
2101 		 * error code is ignored, and no more calls to our ops are made.
2102 		 */
2103 		pr_err("Failed to initialize RDMA connection");
2104 		goto err_out;
2105 	}
2106 
2107 out:
2108 	/*
2109 	 * Keep a track on the number of ib devices added
2110 	 */
2111 	ib_ctx.ib_dev_count++;
2112 
2113 err_out:
2114 	mutex_unlock(&ib_ctx.ib_dev_mutex);
2115 	return ret;
2116 }
2117 
2118 static void rtrs_srv_remove_one(struct ib_device *device, void *client_data)
2119 {
2120 	struct rtrs_srv_ctx *ctx;
2121 
2122 	mutex_lock(&ib_ctx.ib_dev_mutex);
2123 	ib_ctx.ib_dev_count--;
2124 
2125 	if (ib_ctx.ib_dev_count)
2126 		goto out;
2127 
2128 	/*
2129 	 * Since our CM IDs are NOT bound to any ib device we will remove them
2130 	 * only once, when the last device is removed
2131 	 */
2132 	ctx = ib_ctx.srv_ctx;
2133 	rdma_destroy_id(ctx->cm_id_ip);
2134 	rdma_destroy_id(ctx->cm_id_ib);
2135 
2136 out:
2137 	mutex_unlock(&ib_ctx.ib_dev_mutex);
2138 }
2139 
2140 static struct ib_client rtrs_srv_client = {
2141 	.name	= "rtrs_server",
2142 	.add	= rtrs_srv_add_one,
2143 	.remove	= rtrs_srv_remove_one
2144 };
2145 
2146 /**
2147  * rtrs_srv_open() - open RTRS server context
2148  * @ops:		callback functions
2149  * @port:               port to listen on
2150  *
2151  * Creates server context with specified callbacks.
2152  *
2153  * Return a valid pointer on success otherwise PTR_ERR.
2154  */
2155 struct rtrs_srv_ctx *rtrs_srv_open(struct rtrs_srv_ops *ops, u16 port)
2156 {
2157 	struct rtrs_srv_ctx *ctx;
2158 	int err;
2159 
2160 	ctx = alloc_srv_ctx(ops);
2161 	if (!ctx)
2162 		return ERR_PTR(-ENOMEM);
2163 
2164 	mutex_init(&ib_ctx.ib_dev_mutex);
2165 	ib_ctx.srv_ctx = ctx;
2166 	ib_ctx.port = port;
2167 
2168 	err = ib_register_client(&rtrs_srv_client);
2169 	if (err) {
2170 		free_srv_ctx(ctx);
2171 		return ERR_PTR(err);
2172 	}
2173 
2174 	return ctx;
2175 }
2176 EXPORT_SYMBOL(rtrs_srv_open);
2177 
2178 static void close_paths(struct rtrs_srv_sess *srv)
2179 {
2180 	struct rtrs_srv_path *srv_path;
2181 
2182 	mutex_lock(&srv->paths_mutex);
2183 	list_for_each_entry(srv_path, &srv->paths_list, s.entry)
2184 		close_path(srv_path);
2185 	mutex_unlock(&srv->paths_mutex);
2186 }
2187 
2188 static void close_ctx(struct rtrs_srv_ctx *ctx)
2189 {
2190 	struct rtrs_srv_sess *srv;
2191 
2192 	mutex_lock(&ctx->srv_mutex);
2193 	list_for_each_entry(srv, &ctx->srv_list, ctx_list)
2194 		close_paths(srv);
2195 	mutex_unlock(&ctx->srv_mutex);
2196 	flush_workqueue(rtrs_wq);
2197 }
2198 
2199 /**
2200  * rtrs_srv_close() - close RTRS server context
2201  * @ctx: pointer to server context
2202  *
2203  * Closes RTRS server context with all client sessions.
2204  */
2205 void rtrs_srv_close(struct rtrs_srv_ctx *ctx)
2206 {
2207 	ib_unregister_client(&rtrs_srv_client);
2208 	mutex_destroy(&ib_ctx.ib_dev_mutex);
2209 	close_ctx(ctx);
2210 	free_srv_ctx(ctx);
2211 }
2212 EXPORT_SYMBOL(rtrs_srv_close);
2213 
2214 static int check_module_params(void)
2215 {
2216 	if (sess_queue_depth < 1 || sess_queue_depth > MAX_SESS_QUEUE_DEPTH) {
2217 		pr_err("Invalid sess_queue_depth value %d, has to be >= %d, <= %d.\n",
2218 		       sess_queue_depth, 1, MAX_SESS_QUEUE_DEPTH);
2219 		return -EINVAL;
2220 	}
2221 	if (max_chunk_size < MIN_CHUNK_SIZE || !is_power_of_2(max_chunk_size)) {
2222 		pr_err("Invalid max_chunk_size value %d, has to be >= %d and should be power of two.\n",
2223 		       max_chunk_size, MIN_CHUNK_SIZE);
2224 		return -EINVAL;
2225 	}
2226 
2227 	/*
2228 	 * Check if IB immediate data size is enough to hold the mem_id and the
2229 	 * offset inside the memory chunk
2230 	 */
2231 	if ((ilog2(sess_queue_depth - 1) + 1) +
2232 	    (ilog2(max_chunk_size - 1) + 1) > MAX_IMM_PAYL_BITS) {
2233 		pr_err("RDMA immediate size (%db) not enough to encode %d buffers of size %dB. Reduce 'sess_queue_depth' or 'max_chunk_size' parameters.\n",
2234 		       MAX_IMM_PAYL_BITS, sess_queue_depth, max_chunk_size);
2235 		return -EINVAL;
2236 	}
2237 
2238 	return 0;
2239 }
2240 
2241 static int __init rtrs_server_init(void)
2242 {
2243 	int err;
2244 
2245 	pr_info("Loading module %s, proto %s: (max_chunk_size: %d (pure IO %ld, headers %ld) , sess_queue_depth: %d, always_invalidate: %d)\n",
2246 		KBUILD_MODNAME, RTRS_PROTO_VER_STRING,
2247 		max_chunk_size, max_chunk_size - MAX_HDR_SIZE, MAX_HDR_SIZE,
2248 		sess_queue_depth, always_invalidate);
2249 
2250 	rtrs_rdma_dev_pd_init(0, &dev_pd);
2251 
2252 	err = check_module_params();
2253 	if (err) {
2254 		pr_err("Failed to load module, invalid module parameters, err: %d\n",
2255 		       err);
2256 		return err;
2257 	}
2258 	err = class_register(&rtrs_dev_class);
2259 	if (err)
2260 		goto out_err;
2261 
2262 	rtrs_wq = alloc_workqueue("rtrs_server_wq", 0, 0);
2263 	if (!rtrs_wq) {
2264 		err = -ENOMEM;
2265 		goto out_dev_class;
2266 	}
2267 
2268 	return 0;
2269 
2270 out_dev_class:
2271 	class_unregister(&rtrs_dev_class);
2272 out_err:
2273 	return err;
2274 }
2275 
2276 static void __exit rtrs_server_exit(void)
2277 {
2278 	destroy_workqueue(rtrs_wq);
2279 	class_unregister(&rtrs_dev_class);
2280 	rtrs_rdma_dev_pd_deinit(&dev_pd);
2281 }
2282 
2283 module_init(rtrs_server_init);
2284 module_exit(rtrs_server_exit);
2285