1 /*
2  * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved.
3  * Copyright (c) 2005, 2006 Cisco Systems.  All rights reserved.
4  * Copyright (c) 2013-2014 Mellanox Technologies. All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenIB.org BSD license below:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *	- Redistributions of source code must retain the above
17  *	  copyright notice, this list of conditions and the following
18  *	  disclaimer.
19  *
20  *	- Redistributions in binary form must reproduce the above
21  *	  copyright notice, this list of conditions and the following
22  *	  disclaimer in the documentation and/or other materials
23  *	  provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  */
34 #include <linux/kernel.h>
35 #include <linux/slab.h>
36 #include <linux/delay.h>
37 
38 #include "iscsi_iser.h"
39 
40 #define ISCSI_ISER_MAX_CONN	8
41 #define ISER_MAX_RX_LEN		(ISER_QP_MAX_RECV_DTOS * ISCSI_ISER_MAX_CONN)
42 #define ISER_MAX_TX_LEN		(ISER_QP_MAX_REQ_DTOS  * ISCSI_ISER_MAX_CONN)
43 #define ISER_MAX_CQ_LEN		(ISER_MAX_RX_LEN + ISER_MAX_TX_LEN + \
44 				 ISCSI_ISER_MAX_CONN)
45 
46 static void iser_qp_event_callback(struct ib_event *cause, void *context)
47 {
48 	iser_err("qp event %s (%d)\n",
49 		 ib_event_msg(cause->event), cause->event);
50 }
51 
52 static void iser_event_handler(struct ib_event_handler *handler,
53 				struct ib_event *event)
54 {
55 	iser_err("async event %s (%d) on device %s port %d\n",
56 		 ib_event_msg(event->event), event->event,
57 		dev_name(&event->device->dev), event->element.port_num);
58 }
59 
60 /*
61  * iser_create_device_ib_res - creates Protection Domain (PD), Completion
62  * Queue (CQ), DMA Memory Region (DMA MR) with the device associated with
63  * the adaptor.
64  *
65  * Return: 0 on success, -1 on failure
66  */
67 static int iser_create_device_ib_res(struct iser_device *device)
68 {
69 	struct ib_device *ib_dev = device->ib_device;
70 
71 	if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS)) {
72 		iser_err("IB device does not support memory registrations\n");
73 		return -1;
74 	}
75 
76 	device->pd = ib_alloc_pd(ib_dev,
77 		iser_always_reg ? 0 : IB_PD_UNSAFE_GLOBAL_RKEY);
78 	if (IS_ERR(device->pd))
79 		goto pd_err;
80 
81 	INIT_IB_EVENT_HANDLER(&device->event_handler, ib_dev,
82 			      iser_event_handler);
83 	ib_register_event_handler(&device->event_handler);
84 	return 0;
85 
86 pd_err:
87 	iser_err("failed to allocate an IB resource\n");
88 	return -1;
89 }
90 
91 /*
92  * iser_free_device_ib_res - destroy/dealloc/dereg the DMA MR,
93  * CQ and PD created with the device associated with the adaptor.
94  */
95 static void iser_free_device_ib_res(struct iser_device *device)
96 {
97 	ib_unregister_event_handler(&device->event_handler);
98 	ib_dealloc_pd(device->pd);
99 
100 	device->pd = NULL;
101 }
102 
103 static struct iser_fr_desc *
104 iser_create_fastreg_desc(struct iser_device *device,
105 			 struct ib_pd *pd,
106 			 bool pi_enable,
107 			 unsigned int size)
108 {
109 	struct iser_fr_desc *desc;
110 	struct ib_device *ib_dev = device->ib_device;
111 	enum ib_mr_type mr_type;
112 	int ret;
113 
114 	desc = kzalloc(sizeof(*desc), GFP_KERNEL);
115 	if (!desc)
116 		return ERR_PTR(-ENOMEM);
117 
118 	if (ib_dev->attrs.kernel_cap_flags & IBK_SG_GAPS_REG)
119 		mr_type = IB_MR_TYPE_SG_GAPS;
120 	else
121 		mr_type = IB_MR_TYPE_MEM_REG;
122 
123 	desc->rsc.mr = ib_alloc_mr(pd, mr_type, size);
124 	if (IS_ERR(desc->rsc.mr)) {
125 		ret = PTR_ERR(desc->rsc.mr);
126 		iser_err("Failed to allocate ib_fast_reg_mr err=%d\n", ret);
127 		goto err_alloc_mr;
128 	}
129 
130 	if (pi_enable) {
131 		desc->rsc.sig_mr = ib_alloc_mr_integrity(pd, size, size);
132 		if (IS_ERR(desc->rsc.sig_mr)) {
133 			ret = PTR_ERR(desc->rsc.sig_mr);
134 			iser_err("Failed to allocate sig_mr err=%d\n", ret);
135 			goto err_alloc_mr_integrity;
136 		}
137 	}
138 	desc->rsc.mr_valid = 0;
139 
140 	return desc;
141 
142 err_alloc_mr_integrity:
143 	ib_dereg_mr(desc->rsc.mr);
144 err_alloc_mr:
145 	kfree(desc);
146 
147 	return ERR_PTR(ret);
148 }
149 
150 static void iser_destroy_fastreg_desc(struct iser_fr_desc *desc)
151 {
152 	struct iser_reg_resources *res = &desc->rsc;
153 
154 	ib_dereg_mr(res->mr);
155 	if (res->sig_mr) {
156 		ib_dereg_mr(res->sig_mr);
157 		res->sig_mr = NULL;
158 	}
159 	kfree(desc);
160 }
161 
162 /**
163  * iser_alloc_fastreg_pool - Creates pool of fast_reg descriptors
164  * for fast registration work requests.
165  * @ib_conn: connection RDMA resources
166  * @cmds_max: max number of SCSI commands for this connection
167  * @size: max number of pages per map request
168  *
169  * Return: 0 on success, or errno code on failure
170  */
171 int iser_alloc_fastreg_pool(struct ib_conn *ib_conn,
172 			    unsigned cmds_max,
173 			    unsigned int size)
174 {
175 	struct iser_device *device = ib_conn->device;
176 	struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
177 	struct iser_fr_desc *desc;
178 	int i, ret;
179 
180 	INIT_LIST_HEAD(&fr_pool->list);
181 	INIT_LIST_HEAD(&fr_pool->all_list);
182 	spin_lock_init(&fr_pool->lock);
183 	fr_pool->size = 0;
184 	for (i = 0; i < cmds_max; i++) {
185 		desc = iser_create_fastreg_desc(device, device->pd,
186 						ib_conn->pi_support, size);
187 		if (IS_ERR(desc)) {
188 			ret = PTR_ERR(desc);
189 			goto err;
190 		}
191 
192 		list_add_tail(&desc->list, &fr_pool->list);
193 		list_add_tail(&desc->all_list, &fr_pool->all_list);
194 		fr_pool->size++;
195 	}
196 
197 	return 0;
198 
199 err:
200 	iser_free_fastreg_pool(ib_conn);
201 	return ret;
202 }
203 
204 /**
205  * iser_free_fastreg_pool - releases the pool of fast_reg descriptors
206  * @ib_conn: connection RDMA resources
207  */
208 void iser_free_fastreg_pool(struct ib_conn *ib_conn)
209 {
210 	struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
211 	struct iser_fr_desc *desc, *tmp;
212 	int i = 0;
213 
214 	if (list_empty(&fr_pool->all_list))
215 		return;
216 
217 	iser_info("freeing conn %p fr pool\n", ib_conn);
218 
219 	list_for_each_entry_safe(desc, tmp, &fr_pool->all_list, all_list) {
220 		list_del(&desc->all_list);
221 		iser_destroy_fastreg_desc(desc);
222 		++i;
223 	}
224 
225 	if (i < fr_pool->size)
226 		iser_warn("pool still has %d regions registered\n",
227 			  fr_pool->size - i);
228 }
229 
230 /*
231  * iser_create_ib_conn_res - Queue-Pair (QP)
232  *
233  * Return: 0 on success, -1 on failure
234  */
235 static int iser_create_ib_conn_res(struct ib_conn *ib_conn)
236 {
237 	struct iser_conn *iser_conn = to_iser_conn(ib_conn);
238 	struct iser_device	*device;
239 	struct ib_device	*ib_dev;
240 	struct ib_qp_init_attr	init_attr;
241 	int			ret = -ENOMEM;
242 	unsigned int max_send_wr, cq_size;
243 
244 	BUG_ON(ib_conn->device == NULL);
245 
246 	device = ib_conn->device;
247 	ib_dev = device->ib_device;
248 
249 	/* +1 for drain */
250 	if (ib_conn->pi_support)
251 		max_send_wr = ISER_QP_SIG_MAX_REQ_DTOS + 1;
252 	else
253 		max_send_wr = ISER_QP_MAX_REQ_DTOS + 1;
254 	max_send_wr = min_t(unsigned int, max_send_wr,
255 			    (unsigned int)ib_dev->attrs.max_qp_wr);
256 
257 	cq_size = max_send_wr + ISER_QP_MAX_RECV_DTOS;
258 	ib_conn->cq = ib_cq_pool_get(ib_dev, cq_size, -1, IB_POLL_SOFTIRQ);
259 	if (IS_ERR(ib_conn->cq)) {
260 		ret = PTR_ERR(ib_conn->cq);
261 		goto cq_err;
262 	}
263 	ib_conn->cq_size = cq_size;
264 
265 	memset(&init_attr, 0, sizeof(init_attr));
266 
267 	init_attr.event_handler = iser_qp_event_callback;
268 	init_attr.qp_context = (void *)ib_conn;
269 	init_attr.send_cq = ib_conn->cq;
270 	init_attr.recv_cq = ib_conn->cq;
271 	/* +1 for drain */
272 	init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS + 1;
273 	init_attr.cap.max_send_sge = 2;
274 	init_attr.cap.max_recv_sge = 1;
275 	init_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
276 	init_attr.qp_type = IB_QPT_RC;
277 	init_attr.cap.max_send_wr = max_send_wr;
278 	if (ib_conn->pi_support)
279 		init_attr.create_flags |= IB_QP_CREATE_INTEGRITY_EN;
280 	iser_conn->max_cmds = ISER_GET_MAX_XMIT_CMDS(max_send_wr - 1);
281 
282 	ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr);
283 	if (ret)
284 		goto out_err;
285 
286 	ib_conn->qp = ib_conn->cma_id->qp;
287 	iser_info("setting conn %p cma_id %p qp %p max_send_wr %d\n", ib_conn,
288 		  ib_conn->cma_id, ib_conn->cma_id->qp, max_send_wr);
289 	return ret;
290 
291 out_err:
292 	ib_cq_pool_put(ib_conn->cq, ib_conn->cq_size);
293 cq_err:
294 	iser_err("unable to alloc mem or create resource, err %d\n", ret);
295 
296 	return ret;
297 }
298 
299 /*
300  * based on the resolved device node GUID see if there already allocated
301  * device for this device. If there's no such, create one.
302  */
303 static
304 struct iser_device *iser_device_find_by_ib_device(struct rdma_cm_id *cma_id)
305 {
306 	struct iser_device *device;
307 
308 	mutex_lock(&ig.device_list_mutex);
309 
310 	list_for_each_entry(device, &ig.device_list, ig_list)
311 		/* find if there's a match using the node GUID */
312 		if (device->ib_device->node_guid == cma_id->device->node_guid)
313 			goto inc_refcnt;
314 
315 	device = kzalloc(sizeof *device, GFP_KERNEL);
316 	if (!device)
317 		goto out;
318 
319 	/* assign this device to the device */
320 	device->ib_device = cma_id->device;
321 	/* init the device and link it into ig device list */
322 	if (iser_create_device_ib_res(device)) {
323 		kfree(device);
324 		device = NULL;
325 		goto out;
326 	}
327 	list_add(&device->ig_list, &ig.device_list);
328 
329 inc_refcnt:
330 	device->refcount++;
331 out:
332 	mutex_unlock(&ig.device_list_mutex);
333 	return device;
334 }
335 
336 /* if there's no demand for this device, release it */
337 static void iser_device_try_release(struct iser_device *device)
338 {
339 	mutex_lock(&ig.device_list_mutex);
340 	device->refcount--;
341 	iser_info("device %p refcount %d\n", device, device->refcount);
342 	if (!device->refcount) {
343 		iser_free_device_ib_res(device);
344 		list_del(&device->ig_list);
345 		kfree(device);
346 	}
347 	mutex_unlock(&ig.device_list_mutex);
348 }
349 
350 void iser_release_work(struct work_struct *work)
351 {
352 	struct iser_conn *iser_conn;
353 
354 	iser_conn = container_of(work, struct iser_conn, release_work);
355 
356 	/* Wait for conn_stop to complete */
357 	wait_for_completion(&iser_conn->stop_completion);
358 	/* Wait for IB resouces cleanup to complete */
359 	wait_for_completion(&iser_conn->ib_completion);
360 
361 	mutex_lock(&iser_conn->state_mutex);
362 	iser_conn->state = ISER_CONN_DOWN;
363 	mutex_unlock(&iser_conn->state_mutex);
364 
365 	iser_conn_release(iser_conn);
366 }
367 
368 /**
369  * iser_free_ib_conn_res - release IB related resources
370  * @iser_conn: iser connection struct
371  * @destroy: indicator if we need to try to release the
372  *     iser device and memory regoins pool (only iscsi
373  *     shutdown and DEVICE_REMOVAL will use this).
374  *
375  * This routine is called with the iser state mutex held
376  * so the cm_id removal is out of here. It is Safe to
377  * be invoked multiple times.
378  */
379 static void iser_free_ib_conn_res(struct iser_conn *iser_conn, bool destroy)
380 {
381 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
382 	struct iser_device *device = ib_conn->device;
383 
384 	iser_info("freeing conn %p cma_id %p qp %p\n",
385 		  iser_conn, ib_conn->cma_id, ib_conn->qp);
386 
387 	if (ib_conn->qp) {
388 		rdma_destroy_qp(ib_conn->cma_id);
389 		ib_cq_pool_put(ib_conn->cq, ib_conn->cq_size);
390 		ib_conn->qp = NULL;
391 	}
392 
393 	if (destroy) {
394 		if (iser_conn->rx_descs)
395 			iser_free_rx_descriptors(iser_conn);
396 
397 		if (device) {
398 			iser_device_try_release(device);
399 			ib_conn->device = NULL;
400 		}
401 	}
402 }
403 
404 /**
405  * iser_conn_release - Frees all conn objects and deallocs conn descriptor
406  * @iser_conn: iSER connection context
407  */
408 void iser_conn_release(struct iser_conn *iser_conn)
409 {
410 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
411 
412 	mutex_lock(&ig.connlist_mutex);
413 	list_del(&iser_conn->conn_list);
414 	mutex_unlock(&ig.connlist_mutex);
415 
416 	mutex_lock(&iser_conn->state_mutex);
417 	/* In case we endup here without ep_disconnect being invoked. */
418 	if (iser_conn->state != ISER_CONN_DOWN) {
419 		iser_warn("iser conn %p state %d, expected state down.\n",
420 			  iser_conn, iser_conn->state);
421 		iscsi_destroy_endpoint(iser_conn->ep);
422 		iser_conn->state = ISER_CONN_DOWN;
423 	}
424 	/*
425 	 * In case we never got to bind stage, we still need to
426 	 * release IB resources (which is safe to call more than once).
427 	 */
428 	iser_free_ib_conn_res(iser_conn, true);
429 	mutex_unlock(&iser_conn->state_mutex);
430 
431 	if (ib_conn->cma_id) {
432 		rdma_destroy_id(ib_conn->cma_id);
433 		ib_conn->cma_id = NULL;
434 	}
435 
436 	kfree(iser_conn);
437 }
438 
439 /**
440  * iser_conn_terminate - triggers start of the disconnect procedures and
441  * waits for them to be done
442  * @iser_conn: iSER connection context
443  *
444  * Called with state mutex held
445  */
446 int iser_conn_terminate(struct iser_conn *iser_conn)
447 {
448 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
449 	int err = 0;
450 
451 	lockdep_assert_held(&iser_conn->state_mutex);
452 
453 	/* terminate the iser conn only if the conn state is UP */
454 	if (iser_conn->state != ISER_CONN_UP)
455 		return 0;
456 
457 	iser_conn->state = ISER_CONN_TERMINATING;
458 	iser_info("iser_conn %p state %d\n", iser_conn, iser_conn->state);
459 
460 	/* suspend queuing of new iscsi commands */
461 	if (iser_conn->iscsi_conn)
462 		iscsi_suspend_queue(iser_conn->iscsi_conn);
463 
464 	/*
465 	 * In case we didn't already clean up the cma_id (peer initiated
466 	 * a disconnection), we need to Cause the CMA to change the QP
467 	 * state to ERROR.
468 	 */
469 	if (ib_conn->cma_id) {
470 		err = rdma_disconnect(ib_conn->cma_id);
471 		if (err)
472 			iser_err("Failed to disconnect, conn: 0x%p err %d\n",
473 				 iser_conn, err);
474 
475 		/* block until all flush errors are consumed */
476 		ib_drain_qp(ib_conn->qp);
477 	}
478 
479 	return 1;
480 }
481 
482 /*
483  * Called with state mutex held
484  */
485 static void iser_connect_error(struct rdma_cm_id *cma_id)
486 {
487 	struct iser_conn *iser_conn = cma_id->context;
488 
489 	lockdep_assert_held(&iser_conn->state_mutex);
490 
491 	iser_conn->state = ISER_CONN_TERMINATING;
492 }
493 
494 static void iser_calc_scsi_params(struct iser_conn *iser_conn,
495 				  unsigned int max_sectors)
496 {
497 	struct iser_device *device = iser_conn->ib_conn.device;
498 	struct ib_device_attr *attr = &device->ib_device->attrs;
499 	unsigned short sg_tablesize, sup_sg_tablesize;
500 	unsigned short reserved_mr_pages;
501 	u32 max_num_sg;
502 
503 	/*
504 	 * FRs without SG_GAPS can only map up to a (device) page per entry,
505 	 * but if the first entry is misaligned we'll end up using two entries
506 	 * (head and tail) for a single page worth data, so one additional
507 	 * entry is required.
508 	 */
509 	if (attr->kernel_cap_flags & IBK_SG_GAPS_REG)
510 		reserved_mr_pages = 0;
511 	else
512 		reserved_mr_pages = 1;
513 
514 	if (iser_conn->ib_conn.pi_support)
515 		max_num_sg = attr->max_pi_fast_reg_page_list_len;
516 	else
517 		max_num_sg = attr->max_fast_reg_page_list_len;
518 
519 	sg_tablesize = DIV_ROUND_UP(max_sectors * SECTOR_SIZE, SZ_4K);
520 	sup_sg_tablesize = min_t(uint, ISCSI_ISER_MAX_SG_TABLESIZE,
521 				 max_num_sg - reserved_mr_pages);
522 	iser_conn->scsi_sg_tablesize = min(sg_tablesize, sup_sg_tablesize);
523 	iser_conn->pages_per_mr =
524 		iser_conn->scsi_sg_tablesize + reserved_mr_pages;
525 }
526 
527 /*
528  * Called with state mutex held
529  */
530 static void iser_addr_handler(struct rdma_cm_id *cma_id)
531 {
532 	struct iser_conn *iser_conn = cma_id->context;
533 	struct iser_device *device;
534 	struct ib_conn *ib_conn;
535 	int    ret;
536 
537 	lockdep_assert_held(&iser_conn->state_mutex);
538 
539 	if (iser_conn->state != ISER_CONN_PENDING)
540 		/* bailout */
541 		return;
542 
543 	ib_conn = &iser_conn->ib_conn;
544 	device = iser_device_find_by_ib_device(cma_id);
545 	if (!device) {
546 		iser_err("device lookup/creation failed\n");
547 		iser_connect_error(cma_id);
548 		return;
549 	}
550 
551 	ib_conn->device = device;
552 
553 	/* connection T10-PI support */
554 	if (iser_pi_enable) {
555 		if (!(device->ib_device->attrs.kernel_cap_flags &
556 		      IBK_INTEGRITY_HANDOVER)) {
557 			iser_warn("T10-PI requested but not supported on %s, "
558 				  "continue without T10-PI\n",
559 				  dev_name(&ib_conn->device->ib_device->dev));
560 			ib_conn->pi_support = false;
561 		} else {
562 			ib_conn->pi_support = true;
563 		}
564 	}
565 
566 	iser_calc_scsi_params(iser_conn, iser_max_sectors);
567 
568 	ret = rdma_resolve_route(cma_id, 1000);
569 	if (ret) {
570 		iser_err("resolve route failed: %d\n", ret);
571 		iser_connect_error(cma_id);
572 		return;
573 	}
574 }
575 
576 /*
577  * Called with state mutex held
578  */
579 static void iser_route_handler(struct rdma_cm_id *cma_id)
580 {
581 	struct rdma_conn_param conn_param;
582 	int ret;
583 	struct iser_cm_hdr req_hdr;
584 	struct iser_conn *iser_conn = cma_id->context;
585 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
586 	struct ib_device *ib_dev = ib_conn->device->ib_device;
587 
588 	lockdep_assert_held(&iser_conn->state_mutex);
589 
590 	if (iser_conn->state != ISER_CONN_PENDING)
591 		/* bailout */
592 		return;
593 
594 	ret = iser_create_ib_conn_res(ib_conn);
595 	if (ret)
596 		goto failure;
597 
598 	memset(&conn_param, 0, sizeof conn_param);
599 	conn_param.responder_resources = ib_dev->attrs.max_qp_rd_atom;
600 	conn_param.initiator_depth = 1;
601 	conn_param.retry_count = 7;
602 	conn_param.rnr_retry_count = 6;
603 
604 	memset(&req_hdr, 0, sizeof(req_hdr));
605 	req_hdr.flags = ISER_ZBVA_NOT_SUP;
606 	if (!iser_always_reg)
607 		req_hdr.flags |= ISER_SEND_W_INV_NOT_SUP;
608 	conn_param.private_data	= (void *)&req_hdr;
609 	conn_param.private_data_len = sizeof(struct iser_cm_hdr);
610 
611 	ret = rdma_connect_locked(cma_id, &conn_param);
612 	if (ret) {
613 		iser_err("failure connecting: %d\n", ret);
614 		goto failure;
615 	}
616 
617 	return;
618 failure:
619 	iser_connect_error(cma_id);
620 }
621 
622 /*
623  * Called with state mutex held
624  */
625 static void iser_connected_handler(struct rdma_cm_id *cma_id,
626 				   const void *private_data)
627 {
628 	struct iser_conn *iser_conn = cma_id->context;
629 	struct ib_qp_attr attr;
630 	struct ib_qp_init_attr init_attr;
631 
632 	lockdep_assert_held(&iser_conn->state_mutex);
633 
634 	if (iser_conn->state != ISER_CONN_PENDING)
635 		/* bailout */
636 		return;
637 
638 	(void)ib_query_qp(cma_id->qp, &attr, ~0, &init_attr);
639 	iser_info("remote qpn:%x my qpn:%x\n", attr.dest_qp_num, cma_id->qp->qp_num);
640 
641 	if (private_data) {
642 		u8 flags = *(u8 *)private_data;
643 
644 		iser_conn->snd_w_inv = !(flags & ISER_SEND_W_INV_NOT_SUP);
645 	}
646 
647 	iser_info("conn %p: negotiated %s invalidation\n",
648 		  iser_conn, iser_conn->snd_w_inv ? "remote" : "local");
649 
650 	iser_conn->state = ISER_CONN_UP;
651 	complete(&iser_conn->up_completion);
652 }
653 
654 /*
655  * Called with state mutex held
656  */
657 static void iser_cleanup_handler(struct rdma_cm_id *cma_id,
658 				 bool destroy)
659 {
660 	struct iser_conn *iser_conn = cma_id->context;
661 
662 	lockdep_assert_held(&iser_conn->state_mutex);
663 	/*
664 	 * We are not guaranteed that we visited disconnected_handler
665 	 * by now, call it here to be safe that we handle CM drep
666 	 * and flush errors.
667 	 */
668 	if (iser_conn_terminate(iser_conn)) {
669 		if (iser_conn->iscsi_conn)
670 			iscsi_conn_failure(iser_conn->iscsi_conn,
671 					   ISCSI_ERR_CONN_FAILED);
672 		else
673 			iser_err("iscsi_iser connection isn't bound\n");
674 	}
675 	iser_free_ib_conn_res(iser_conn, destroy);
676 	complete(&iser_conn->ib_completion);
677 }
678 
679 static int iser_cma_handler(struct rdma_cm_id *cma_id,
680 			    struct rdma_cm_event *event)
681 {
682 	struct iser_conn *iser_conn;
683 	int ret = 0;
684 
685 	iser_conn = cma_id->context;
686 	iser_info("%s (%d): status %d conn %p id %p\n",
687 		  rdma_event_msg(event->event), event->event,
688 		  event->status, cma_id->context, cma_id);
689 
690 	mutex_lock(&iser_conn->state_mutex);
691 	switch (event->event) {
692 	case RDMA_CM_EVENT_ADDR_RESOLVED:
693 		iser_addr_handler(cma_id);
694 		break;
695 	case RDMA_CM_EVENT_ROUTE_RESOLVED:
696 		iser_route_handler(cma_id);
697 		break;
698 	case RDMA_CM_EVENT_ESTABLISHED:
699 		iser_connected_handler(cma_id, event->param.conn.private_data);
700 		break;
701 	case RDMA_CM_EVENT_REJECTED:
702 		iser_info("Connection rejected: %s\n",
703 			 rdma_reject_msg(cma_id, event->status));
704 		fallthrough;
705 	case RDMA_CM_EVENT_ADDR_ERROR:
706 	case RDMA_CM_EVENT_ROUTE_ERROR:
707 	case RDMA_CM_EVENT_CONNECT_ERROR:
708 	case RDMA_CM_EVENT_UNREACHABLE:
709 		iser_connect_error(cma_id);
710 		break;
711 	case RDMA_CM_EVENT_DISCONNECTED:
712 	case RDMA_CM_EVENT_ADDR_CHANGE:
713 	case RDMA_CM_EVENT_TIMEWAIT_EXIT:
714 		iser_cleanup_handler(cma_id, false);
715 		break;
716 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
717 		/*
718 		 * we *must* destroy the device as we cannot rely
719 		 * on iscsid to be around to initiate error handling.
720 		 * also if we are not in state DOWN implicitly destroy
721 		 * the cma_id.
722 		 */
723 		iser_cleanup_handler(cma_id, true);
724 		if (iser_conn->state != ISER_CONN_DOWN) {
725 			iser_conn->ib_conn.cma_id = NULL;
726 			ret = 1;
727 		}
728 		break;
729 	default:
730 		iser_err("Unexpected RDMA CM event: %s (%d)\n",
731 			 rdma_event_msg(event->event), event->event);
732 		break;
733 	}
734 	mutex_unlock(&iser_conn->state_mutex);
735 
736 	return ret;
737 }
738 
739 void iser_conn_init(struct iser_conn *iser_conn)
740 {
741 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
742 
743 	iser_conn->state = ISER_CONN_INIT;
744 	init_completion(&iser_conn->stop_completion);
745 	init_completion(&iser_conn->ib_completion);
746 	init_completion(&iser_conn->up_completion);
747 	INIT_LIST_HEAD(&iser_conn->conn_list);
748 	mutex_init(&iser_conn->state_mutex);
749 
750 	ib_conn->reg_cqe.done = iser_reg_comp;
751 }
752 
753 /*
754  * starts the process of connecting to the target
755  * sleeps until the connection is established or rejected
756  */
757 int iser_connect(struct iser_conn *iser_conn, struct sockaddr *src_addr,
758 		 struct sockaddr *dst_addr, int non_blocking)
759 {
760 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
761 	int err = 0;
762 
763 	mutex_lock(&iser_conn->state_mutex);
764 
765 	sprintf(iser_conn->name, "%pISp", dst_addr);
766 
767 	iser_info("connecting to: %s\n", iser_conn->name);
768 
769 	/* the device is known only --after-- address resolution */
770 	ib_conn->device = NULL;
771 
772 	iser_conn->state = ISER_CONN_PENDING;
773 
774 	ib_conn->cma_id = rdma_create_id(&init_net, iser_cma_handler,
775 					 iser_conn, RDMA_PS_TCP, IB_QPT_RC);
776 	if (IS_ERR(ib_conn->cma_id)) {
777 		err = PTR_ERR(ib_conn->cma_id);
778 		iser_err("rdma_create_id failed: %d\n", err);
779 		goto id_failure;
780 	}
781 
782 	err = rdma_resolve_addr(ib_conn->cma_id, src_addr, dst_addr, 1000);
783 	if (err) {
784 		iser_err("rdma_resolve_addr failed: %d\n", err);
785 		goto addr_failure;
786 	}
787 
788 	if (!non_blocking) {
789 		wait_for_completion_interruptible(&iser_conn->up_completion);
790 
791 		if (iser_conn->state != ISER_CONN_UP) {
792 			err =  -EIO;
793 			goto connect_failure;
794 		}
795 	}
796 	mutex_unlock(&iser_conn->state_mutex);
797 
798 	mutex_lock(&ig.connlist_mutex);
799 	list_add(&iser_conn->conn_list, &ig.connlist);
800 	mutex_unlock(&ig.connlist_mutex);
801 	return 0;
802 
803 id_failure:
804 	ib_conn->cma_id = NULL;
805 addr_failure:
806 	iser_conn->state = ISER_CONN_DOWN;
807 connect_failure:
808 	mutex_unlock(&iser_conn->state_mutex);
809 	iser_conn_release(iser_conn);
810 	return err;
811 }
812 
813 int iser_post_recvl(struct iser_conn *iser_conn)
814 {
815 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
816 	struct iser_login_desc *desc = &iser_conn->login_desc;
817 	struct ib_recv_wr wr;
818 	int ret;
819 
820 	desc->sge.addr = desc->rsp_dma;
821 	desc->sge.length = ISER_RX_LOGIN_SIZE;
822 	desc->sge.lkey = ib_conn->device->pd->local_dma_lkey;
823 
824 	desc->cqe.done = iser_login_rsp;
825 	wr.wr_cqe = &desc->cqe;
826 	wr.sg_list = &desc->sge;
827 	wr.num_sge = 1;
828 	wr.next = NULL;
829 
830 	ret = ib_post_recv(ib_conn->qp, &wr, NULL);
831 	if (unlikely(ret))
832 		iser_err("ib_post_recv login failed ret=%d\n", ret);
833 
834 	return ret;
835 }
836 
837 int iser_post_recvm(struct iser_conn *iser_conn, struct iser_rx_desc *rx_desc)
838 {
839 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
840 	struct ib_recv_wr wr;
841 	int ret;
842 
843 	rx_desc->cqe.done = iser_task_rsp;
844 	wr.wr_cqe = &rx_desc->cqe;
845 	wr.sg_list = &rx_desc->rx_sg;
846 	wr.num_sge = 1;
847 	wr.next = NULL;
848 
849 	ret = ib_post_recv(ib_conn->qp, &wr, NULL);
850 	if (unlikely(ret))
851 		iser_err("ib_post_recv failed ret=%d\n", ret);
852 
853 	return ret;
854 }
855 
856 
857 /**
858  * iser_post_send - Initiate a Send DTO operation
859  * @ib_conn: connection RDMA resources
860  * @tx_desc: iSER TX descriptor
861  *
862  * Return: 0 on success, -1 on failure
863  */
864 int iser_post_send(struct ib_conn *ib_conn, struct iser_tx_desc *tx_desc)
865 {
866 	struct ib_send_wr *wr = &tx_desc->send_wr;
867 	struct ib_send_wr *first_wr;
868 	int ret;
869 
870 	ib_dma_sync_single_for_device(ib_conn->device->ib_device,
871 				      tx_desc->dma_addr, ISER_HEADERS_LEN,
872 				      DMA_TO_DEVICE);
873 
874 	wr->next = NULL;
875 	wr->wr_cqe = &tx_desc->cqe;
876 	wr->sg_list = tx_desc->tx_sg;
877 	wr->num_sge = tx_desc->num_sge;
878 	wr->opcode = IB_WR_SEND;
879 	wr->send_flags = IB_SEND_SIGNALED;
880 
881 	if (tx_desc->inv_wr.next)
882 		first_wr = &tx_desc->inv_wr;
883 	else if (tx_desc->reg_wr.wr.next)
884 		first_wr = &tx_desc->reg_wr.wr;
885 	else
886 		first_wr = wr;
887 
888 	ret = ib_post_send(ib_conn->qp, first_wr, NULL);
889 	if (unlikely(ret))
890 		iser_err("ib_post_send failed, ret:%d opcode:%d\n",
891 			 ret, wr->opcode);
892 
893 	return ret;
894 }
895 
896 u8 iser_check_task_pi_status(struct iscsi_iser_task *iser_task,
897 			     enum iser_data_dir cmd_dir, sector_t *sector)
898 {
899 	struct iser_mem_reg *reg = &iser_task->rdma_reg[cmd_dir];
900 	struct iser_fr_desc *desc = reg->desc;
901 	unsigned long sector_size = iser_task->sc->device->sector_size;
902 	struct ib_mr_status mr_status;
903 	int ret;
904 
905 	if (desc && desc->sig_protected) {
906 		desc->sig_protected = false;
907 		ret = ib_check_mr_status(desc->rsc.sig_mr,
908 					 IB_MR_CHECK_SIG_STATUS, &mr_status);
909 		if (ret) {
910 			iser_err("ib_check_mr_status failed, ret %d\n", ret);
911 			/* Not a lot we can do, return ambiguous guard error */
912 			*sector = 0;
913 			return 0x1;
914 		}
915 
916 		if (mr_status.fail_status & IB_MR_CHECK_SIG_STATUS) {
917 			sector_t sector_off = mr_status.sig_err.sig_err_offset;
918 
919 			sector_div(sector_off, sector_size + 8);
920 			*sector = scsi_get_sector(iser_task->sc) + sector_off;
921 
922 			iser_err("PI error found type %d at sector %llx "
923 			       "expected %x vs actual %x\n",
924 			       mr_status.sig_err.err_type,
925 			       (unsigned long long)*sector,
926 			       mr_status.sig_err.expected,
927 			       mr_status.sig_err.actual);
928 
929 			switch (mr_status.sig_err.err_type) {
930 			case IB_SIG_BAD_GUARD:
931 				return 0x1;
932 			case IB_SIG_BAD_REFTAG:
933 				return 0x3;
934 			case IB_SIG_BAD_APPTAG:
935 				return 0x2;
936 			}
937 		}
938 	}
939 
940 	return 0;
941 }
942 
943 void iser_err_comp(struct ib_wc *wc, const char *type)
944 {
945 	if (wc->status != IB_WC_WR_FLUSH_ERR) {
946 		struct iser_conn *iser_conn = to_iser_conn(wc->qp->qp_context);
947 
948 		iser_err("%s failure: %s (%d) vend_err %#x\n", type,
949 			 ib_wc_status_msg(wc->status), wc->status,
950 			 wc->vendor_err);
951 
952 		if (iser_conn->iscsi_conn)
953 			iscsi_conn_failure(iser_conn->iscsi_conn,
954 					   ISCSI_ERR_CONN_FAILED);
955 	} else {
956 		iser_dbg("%s failure: %s (%d)\n", type,
957 			 ib_wc_status_msg(wc->status), wc->status);
958 	}
959 }
960