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/module.h>
36 #include <linux/slab.h>
37 #include <linux/delay.h>
38 
39 #include "iscsi_iser.h"
40 
41 #define ISCSI_ISER_MAX_CONN	8
42 #define ISER_MAX_RX_LEN		(ISER_QP_MAX_RECV_DTOS * ISCSI_ISER_MAX_CONN)
43 #define ISER_MAX_TX_LEN		(ISER_QP_MAX_REQ_DTOS  * ISCSI_ISER_MAX_CONN)
44 #define ISER_MAX_CQ_LEN		(ISER_MAX_RX_LEN + ISER_MAX_TX_LEN + \
45 				 ISCSI_ISER_MAX_CONN)
46 
47 static void iser_qp_event_callback(struct ib_event *cause, void *context)
48 {
49 	iser_err("qp event %s (%d)\n",
50 		 ib_event_msg(cause->event), cause->event);
51 }
52 
53 static void iser_event_handler(struct ib_event_handler *handler,
54 				struct ib_event *event)
55 {
56 	iser_err("async event %s (%d) on device %s port %d\n",
57 		 ib_event_msg(event->event), event->event,
58 		dev_name(&event->device->dev), event->element.port_num);
59 }
60 
61 /*
62  * iser_create_device_ib_res - creates Protection Domain (PD), Completion
63  * Queue (CQ), DMA Memory Region (DMA MR) with the device associated with
64  * the adaptor.
65  *
66  * Return: 0 on success, -1 on failure
67  */
68 static int iser_create_device_ib_res(struct iser_device *device)
69 {
70 	struct ib_device *ib_dev = device->ib_device;
71 
72 	if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS)) {
73 		iser_err("IB device does not support memory registrations\n");
74 		return -1;
75 	}
76 
77 	device->pd = ib_alloc_pd(ib_dev,
78 		iser_always_reg ? 0 : IB_PD_UNSAFE_GLOBAL_RKEY);
79 	if (IS_ERR(device->pd))
80 		goto pd_err;
81 
82 	INIT_IB_EVENT_HANDLER(&device->event_handler, ib_dev,
83 			      iser_event_handler);
84 	ib_register_event_handler(&device->event_handler);
85 	return 0;
86 
87 pd_err:
88 	iser_err("failed to allocate an IB resource\n");
89 	return -1;
90 }
91 
92 /*
93  * iser_free_device_ib_res - destroy/dealloc/dereg the DMA MR,
94  * CQ and PD created with the device associated with the adaptor.
95  */
96 static void iser_free_device_ib_res(struct iser_device *device)
97 {
98 	ib_unregister_event_handler(&device->event_handler);
99 	ib_dealloc_pd(device->pd);
100 
101 	device->pd = NULL;
102 }
103 
104 static struct iser_fr_desc *
105 iser_create_fastreg_desc(struct iser_device *device,
106 			 struct ib_pd *pd,
107 			 bool pi_enable,
108 			 unsigned int size)
109 {
110 	struct iser_fr_desc *desc;
111 	struct ib_device *ib_dev = device->ib_device;
112 	enum ib_mr_type mr_type;
113 	int ret;
114 
115 	desc = kzalloc(sizeof(*desc), GFP_KERNEL);
116 	if (!desc)
117 		return ERR_PTR(-ENOMEM);
118 
119 	if (ib_dev->attrs.device_cap_flags & IB_DEVICE_SG_GAPS_REG)
120 		mr_type = IB_MR_TYPE_SG_GAPS;
121 	else
122 		mr_type = IB_MR_TYPE_MEM_REG;
123 
124 	desc->rsc.mr = ib_alloc_mr(pd, mr_type, size);
125 	if (IS_ERR(desc->rsc.mr)) {
126 		ret = PTR_ERR(desc->rsc.mr);
127 		iser_err("Failed to allocate ib_fast_reg_mr err=%d\n", ret);
128 		goto err_alloc_mr;
129 	}
130 
131 	if (pi_enable) {
132 		desc->rsc.sig_mr = ib_alloc_mr_integrity(pd, size, size);
133 		if (IS_ERR(desc->rsc.sig_mr)) {
134 			ret = PTR_ERR(desc->rsc.sig_mr);
135 			iser_err("Failed to allocate sig_mr err=%d\n", ret);
136 			goto err_alloc_mr_integrity;
137 		}
138 	}
139 	desc->rsc.mr_valid = 0;
140 
141 	return desc;
142 
143 err_alloc_mr_integrity:
144 	ib_dereg_mr(desc->rsc.mr);
145 err_alloc_mr:
146 	kfree(desc);
147 
148 	return ERR_PTR(ret);
149 }
150 
151 static void iser_destroy_fastreg_desc(struct iser_fr_desc *desc)
152 {
153 	struct iser_reg_resources *res = &desc->rsc;
154 
155 	ib_dereg_mr(res->mr);
156 	if (res->sig_mr) {
157 		ib_dereg_mr(res->sig_mr);
158 		res->sig_mr = NULL;
159 	}
160 	kfree(desc);
161 }
162 
163 /**
164  * iser_alloc_fastreg_pool - Creates pool of fast_reg descriptors
165  * for fast registration work requests.
166  * @ib_conn: connection RDMA resources
167  * @cmds_max: max number of SCSI commands for this connection
168  * @size: max number of pages per map request
169  *
170  * Return: 0 on success, or errno code on failure
171  */
172 int iser_alloc_fastreg_pool(struct ib_conn *ib_conn,
173 			    unsigned cmds_max,
174 			    unsigned int size)
175 {
176 	struct iser_device *device = ib_conn->device;
177 	struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
178 	struct iser_fr_desc *desc;
179 	int i, ret;
180 
181 	INIT_LIST_HEAD(&fr_pool->list);
182 	INIT_LIST_HEAD(&fr_pool->all_list);
183 	spin_lock_init(&fr_pool->lock);
184 	fr_pool->size = 0;
185 	for (i = 0; i < cmds_max; i++) {
186 		desc = iser_create_fastreg_desc(device, device->pd,
187 						ib_conn->pi_support, size);
188 		if (IS_ERR(desc)) {
189 			ret = PTR_ERR(desc);
190 			goto err;
191 		}
192 
193 		list_add_tail(&desc->list, &fr_pool->list);
194 		list_add_tail(&desc->all_list, &fr_pool->all_list);
195 		fr_pool->size++;
196 	}
197 
198 	return 0;
199 
200 err:
201 	iser_free_fastreg_pool(ib_conn);
202 	return ret;
203 }
204 
205 /**
206  * iser_free_fastreg_pool - releases the pool of fast_reg descriptors
207  * @ib_conn: connection RDMA resources
208  */
209 void iser_free_fastreg_pool(struct ib_conn *ib_conn)
210 {
211 	struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
212 	struct iser_fr_desc *desc, *tmp;
213 	int i = 0;
214 
215 	if (list_empty(&fr_pool->all_list))
216 		return;
217 
218 	iser_info("freeing conn %p fr pool\n", ib_conn);
219 
220 	list_for_each_entry_safe(desc, tmp, &fr_pool->all_list, all_list) {
221 		list_del(&desc->all_list);
222 		iser_destroy_fastreg_desc(desc);
223 		++i;
224 	}
225 
226 	if (i < fr_pool->size)
227 		iser_warn("pool still has %d regions registered\n",
228 			  fr_pool->size - i);
229 }
230 
231 /*
232  * iser_create_ib_conn_res - Queue-Pair (QP)
233  *
234  * Return: 0 on success, -1 on failure
235  */
236 static int iser_create_ib_conn_res(struct ib_conn *ib_conn)
237 {
238 	struct iser_conn *iser_conn = to_iser_conn(ib_conn);
239 	struct iser_device	*device;
240 	struct ib_device	*ib_dev;
241 	struct ib_qp_init_attr	init_attr;
242 	int			ret = -ENOMEM;
243 	unsigned int max_send_wr, cq_size;
244 
245 	BUG_ON(ib_conn->device == NULL);
246 
247 	device = ib_conn->device;
248 	ib_dev = device->ib_device;
249 
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 	init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS;
272 	init_attr.cap.max_send_sge = 2;
273 	init_attr.cap.max_recv_sge = 1;
274 	init_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
275 	init_attr.qp_type = IB_QPT_RC;
276 	init_attr.cap.max_send_wr = max_send_wr;
277 	if (ib_conn->pi_support)
278 		init_attr.create_flags |= IB_QP_CREATE_INTEGRITY_EN;
279 	iser_conn->max_cmds = ISER_GET_MAX_XMIT_CMDS(max_send_wr - 1);
280 
281 	ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr);
282 	if (ret)
283 		goto out_err;
284 
285 	ib_conn->qp = ib_conn->cma_id->qp;
286 	iser_info("setting conn %p cma_id %p qp %p max_send_wr %d\n", ib_conn,
287 		  ib_conn->cma_id, ib_conn->cma_id->qp, max_send_wr);
288 	return ret;
289 
290 out_err:
291 	ib_cq_pool_put(ib_conn->cq, ib_conn->cq_size);
292 cq_err:
293 	iser_err("unable to alloc mem or create resource, err %d\n", ret);
294 
295 	return ret;
296 }
297 
298 /*
299  * based on the resolved device node GUID see if there already allocated
300  * device for this device. If there's no such, create one.
301  */
302 static
303 struct iser_device *iser_device_find_by_ib_device(struct rdma_cm_id *cma_id)
304 {
305 	struct iser_device *device;
306 
307 	mutex_lock(&ig.device_list_mutex);
308 
309 	list_for_each_entry(device, &ig.device_list, ig_list)
310 		/* find if there's a match using the node GUID */
311 		if (device->ib_device->node_guid == cma_id->device->node_guid)
312 			goto inc_refcnt;
313 
314 	device = kzalloc(sizeof *device, GFP_KERNEL);
315 	if (!device)
316 		goto out;
317 
318 	/* assign this device to the device */
319 	device->ib_device = cma_id->device;
320 	/* init the device and link it into ig device list */
321 	if (iser_create_device_ib_res(device)) {
322 		kfree(device);
323 		device = NULL;
324 		goto out;
325 	}
326 	list_add(&device->ig_list, &ig.device_list);
327 
328 inc_refcnt:
329 	device->refcount++;
330 out:
331 	mutex_unlock(&ig.device_list_mutex);
332 	return device;
333 }
334 
335 /* if there's no demand for this device, release it */
336 static void iser_device_try_release(struct iser_device *device)
337 {
338 	mutex_lock(&ig.device_list_mutex);
339 	device->refcount--;
340 	iser_info("device %p refcount %d\n", device, device->refcount);
341 	if (!device->refcount) {
342 		iser_free_device_ib_res(device);
343 		list_del(&device->ig_list);
344 		kfree(device);
345 	}
346 	mutex_unlock(&ig.device_list_mutex);
347 }
348 
349 /*
350  * Called with state mutex held
351  */
352 static int iser_conn_state_comp_exch(struct iser_conn *iser_conn,
353 				     enum iser_conn_state comp,
354 				     enum iser_conn_state exch)
355 {
356 	int ret;
357 
358 	ret = (iser_conn->state == comp);
359 	if (ret)
360 		iser_conn->state = exch;
361 
362 	return ret;
363 }
364 
365 void iser_release_work(struct work_struct *work)
366 {
367 	struct iser_conn *iser_conn;
368 
369 	iser_conn = container_of(work, struct iser_conn, release_work);
370 
371 	/* Wait for conn_stop to complete */
372 	wait_for_completion(&iser_conn->stop_completion);
373 	/* Wait for IB resouces cleanup to complete */
374 	wait_for_completion(&iser_conn->ib_completion);
375 
376 	mutex_lock(&iser_conn->state_mutex);
377 	iser_conn->state = ISER_CONN_DOWN;
378 	mutex_unlock(&iser_conn->state_mutex);
379 
380 	iser_conn_release(iser_conn);
381 }
382 
383 /**
384  * iser_free_ib_conn_res - release IB related resources
385  * @iser_conn: iser connection struct
386  * @destroy: indicator if we need to try to release the
387  *     iser device and memory regoins pool (only iscsi
388  *     shutdown and DEVICE_REMOVAL will use this).
389  *
390  * This routine is called with the iser state mutex held
391  * so the cm_id removal is out of here. It is Safe to
392  * be invoked multiple times.
393  */
394 static void iser_free_ib_conn_res(struct iser_conn *iser_conn, bool destroy)
395 {
396 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
397 	struct iser_device *device = ib_conn->device;
398 
399 	iser_info("freeing conn %p cma_id %p qp %p\n",
400 		  iser_conn, ib_conn->cma_id, ib_conn->qp);
401 
402 	if (ib_conn->qp) {
403 		rdma_destroy_qp(ib_conn->cma_id);
404 		ib_cq_pool_put(ib_conn->cq, ib_conn->cq_size);
405 		ib_conn->qp = NULL;
406 	}
407 
408 	if (destroy) {
409 		if (iser_conn->rx_descs)
410 			iser_free_rx_descriptors(iser_conn);
411 
412 		if (device) {
413 			iser_device_try_release(device);
414 			ib_conn->device = NULL;
415 		}
416 	}
417 }
418 
419 /**
420  * iser_conn_release - Frees all conn objects and deallocs conn descriptor
421  * @iser_conn: iSER connection context
422  */
423 void iser_conn_release(struct iser_conn *iser_conn)
424 {
425 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
426 
427 	mutex_lock(&ig.connlist_mutex);
428 	list_del(&iser_conn->conn_list);
429 	mutex_unlock(&ig.connlist_mutex);
430 
431 	mutex_lock(&iser_conn->state_mutex);
432 	/* In case we endup here without ep_disconnect being invoked. */
433 	if (iser_conn->state != ISER_CONN_DOWN) {
434 		iser_warn("iser conn %p state %d, expected state down.\n",
435 			  iser_conn, iser_conn->state);
436 		iscsi_destroy_endpoint(iser_conn->ep);
437 		iser_conn->state = ISER_CONN_DOWN;
438 	}
439 	/*
440 	 * In case we never got to bind stage, we still need to
441 	 * release IB resources (which is safe to call more than once).
442 	 */
443 	iser_free_ib_conn_res(iser_conn, true);
444 	mutex_unlock(&iser_conn->state_mutex);
445 
446 	if (ib_conn->cma_id) {
447 		rdma_destroy_id(ib_conn->cma_id);
448 		ib_conn->cma_id = NULL;
449 	}
450 
451 	kfree(iser_conn);
452 }
453 
454 /**
455  * iser_conn_terminate - triggers start of the disconnect procedures and
456  * waits for them to be done
457  * @iser_conn: iSER connection context
458  *
459  * Called with state mutex held
460  */
461 int iser_conn_terminate(struct iser_conn *iser_conn)
462 {
463 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
464 	int err = 0;
465 
466 	/* terminate the iser conn only if the conn state is UP */
467 	if (!iser_conn_state_comp_exch(iser_conn, ISER_CONN_UP,
468 				       ISER_CONN_TERMINATING))
469 		return 0;
470 
471 	iser_info("iser_conn %p state %d\n", iser_conn, iser_conn->state);
472 
473 	/* suspend queuing of new iscsi commands */
474 	if (iser_conn->iscsi_conn)
475 		iscsi_suspend_queue(iser_conn->iscsi_conn);
476 
477 	/*
478 	 * In case we didn't already clean up the cma_id (peer initiated
479 	 * a disconnection), we need to Cause the CMA to change the QP
480 	 * state to ERROR.
481 	 */
482 	if (ib_conn->cma_id) {
483 		err = rdma_disconnect(ib_conn->cma_id);
484 		if (err)
485 			iser_err("Failed to disconnect, conn: 0x%p err %d\n",
486 				 iser_conn, err);
487 
488 		/* block until all flush errors are consumed */
489 		ib_drain_sq(ib_conn->qp);
490 	}
491 
492 	return 1;
493 }
494 
495 /*
496  * Called with state mutex held
497  */
498 static void iser_connect_error(struct rdma_cm_id *cma_id)
499 {
500 	struct iser_conn *iser_conn;
501 
502 	iser_conn = cma_id->context;
503 	iser_conn->state = ISER_CONN_TERMINATING;
504 }
505 
506 static void iser_calc_scsi_params(struct iser_conn *iser_conn,
507 				  unsigned int max_sectors)
508 {
509 	struct iser_device *device = iser_conn->ib_conn.device;
510 	struct ib_device_attr *attr = &device->ib_device->attrs;
511 	unsigned short sg_tablesize, sup_sg_tablesize;
512 	unsigned short reserved_mr_pages;
513 	u32 max_num_sg;
514 
515 	/*
516 	 * FRs without SG_GAPS can only map up to a (device) page per entry,
517 	 * but if the first entry is misaligned we'll end up using two entries
518 	 * (head and tail) for a single page worth data, so one additional
519 	 * entry is required.
520 	 */
521 	if (attr->device_cap_flags & IB_DEVICE_SG_GAPS_REG)
522 		reserved_mr_pages = 0;
523 	else
524 		reserved_mr_pages = 1;
525 
526 	if (iser_conn->ib_conn.pi_support)
527 		max_num_sg = attr->max_pi_fast_reg_page_list_len;
528 	else
529 		max_num_sg = attr->max_fast_reg_page_list_len;
530 
531 	sg_tablesize = DIV_ROUND_UP(max_sectors * SECTOR_SIZE, SZ_4K);
532 	sup_sg_tablesize = min_t(uint, ISCSI_ISER_MAX_SG_TABLESIZE,
533 				 max_num_sg - reserved_mr_pages);
534 	iser_conn->scsi_sg_tablesize = min(sg_tablesize, sup_sg_tablesize);
535 	iser_conn->pages_per_mr =
536 		iser_conn->scsi_sg_tablesize + reserved_mr_pages;
537 }
538 
539 /*
540  * Called with state mutex held
541  */
542 static void iser_addr_handler(struct rdma_cm_id *cma_id)
543 {
544 	struct iser_device *device;
545 	struct iser_conn *iser_conn;
546 	struct ib_conn *ib_conn;
547 	int    ret;
548 
549 	iser_conn = cma_id->context;
550 	if (iser_conn->state != ISER_CONN_PENDING)
551 		/* bailout */
552 		return;
553 
554 	ib_conn = &iser_conn->ib_conn;
555 	device = iser_device_find_by_ib_device(cma_id);
556 	if (!device) {
557 		iser_err("device lookup/creation failed\n");
558 		iser_connect_error(cma_id);
559 		return;
560 	}
561 
562 	ib_conn->device = device;
563 
564 	/* connection T10-PI support */
565 	if (iser_pi_enable) {
566 		if (!(device->ib_device->attrs.device_cap_flags &
567 		      IB_DEVICE_INTEGRITY_HANDOVER)) {
568 			iser_warn("T10-PI requested but not supported on %s, "
569 				  "continue without T10-PI\n",
570 				  dev_name(&ib_conn->device->ib_device->dev));
571 			ib_conn->pi_support = false;
572 		} else {
573 			ib_conn->pi_support = true;
574 		}
575 	}
576 
577 	iser_calc_scsi_params(iser_conn, iser_max_sectors);
578 
579 	ret = rdma_resolve_route(cma_id, 1000);
580 	if (ret) {
581 		iser_err("resolve route failed: %d\n", ret);
582 		iser_connect_error(cma_id);
583 		return;
584 	}
585 }
586 
587 /*
588  * Called with state mutex held
589  */
590 static void iser_route_handler(struct rdma_cm_id *cma_id)
591 {
592 	struct rdma_conn_param conn_param;
593 	int ret;
594 	struct iser_cm_hdr req_hdr;
595 	struct iser_conn *iser_conn = cma_id->context;
596 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
597 	struct ib_device *ib_dev = ib_conn->device->ib_device;
598 
599 	if (iser_conn->state != ISER_CONN_PENDING)
600 		/* bailout */
601 		return;
602 
603 	ret = iser_create_ib_conn_res(ib_conn);
604 	if (ret)
605 		goto failure;
606 
607 	memset(&conn_param, 0, sizeof conn_param);
608 	conn_param.responder_resources = ib_dev->attrs.max_qp_rd_atom;
609 	conn_param.initiator_depth = 1;
610 	conn_param.retry_count = 7;
611 	conn_param.rnr_retry_count = 6;
612 
613 	memset(&req_hdr, 0, sizeof(req_hdr));
614 	req_hdr.flags = ISER_ZBVA_NOT_SUP;
615 	if (!iser_always_reg)
616 		req_hdr.flags |= ISER_SEND_W_INV_NOT_SUP;
617 	conn_param.private_data	= (void *)&req_hdr;
618 	conn_param.private_data_len = sizeof(struct iser_cm_hdr);
619 
620 	ret = rdma_connect_locked(cma_id, &conn_param);
621 	if (ret) {
622 		iser_err("failure connecting: %d\n", ret);
623 		goto failure;
624 	}
625 
626 	return;
627 failure:
628 	iser_connect_error(cma_id);
629 }
630 
631 static void iser_connected_handler(struct rdma_cm_id *cma_id,
632 				   const void *private_data)
633 {
634 	struct iser_conn *iser_conn;
635 	struct ib_qp_attr attr;
636 	struct ib_qp_init_attr init_attr;
637 
638 	iser_conn = cma_id->context;
639 	if (iser_conn->state != ISER_CONN_PENDING)
640 		/* bailout */
641 		return;
642 
643 	(void)ib_query_qp(cma_id->qp, &attr, ~0, &init_attr);
644 	iser_info("remote qpn:%x my qpn:%x\n", attr.dest_qp_num, cma_id->qp->qp_num);
645 
646 	if (private_data) {
647 		u8 flags = *(u8 *)private_data;
648 
649 		iser_conn->snd_w_inv = !(flags & ISER_SEND_W_INV_NOT_SUP);
650 	}
651 
652 	iser_info("conn %p: negotiated %s invalidation\n",
653 		  iser_conn, iser_conn->snd_w_inv ? "remote" : "local");
654 
655 	iser_conn->state = ISER_CONN_UP;
656 	complete(&iser_conn->up_completion);
657 }
658 
659 static void iser_disconnected_handler(struct rdma_cm_id *cma_id)
660 {
661 	struct iser_conn *iser_conn = cma_id->context;
662 
663 	if (iser_conn_terminate(iser_conn)) {
664 		if (iser_conn->iscsi_conn)
665 			iscsi_conn_failure(iser_conn->iscsi_conn,
666 					   ISCSI_ERR_CONN_FAILED);
667 		else
668 			iser_err("iscsi_iser connection isn't bound\n");
669 	}
670 }
671 
672 static void iser_cleanup_handler(struct rdma_cm_id *cma_id,
673 				 bool destroy)
674 {
675 	struct iser_conn *iser_conn = cma_id->context;
676 
677 	/*
678 	 * We are not guaranteed that we visited disconnected_handler
679 	 * by now, call it here to be safe that we handle CM drep
680 	 * and flush errors.
681 	 */
682 	iser_disconnected_handler(cma_id);
683 	iser_free_ib_conn_res(iser_conn, destroy);
684 	complete(&iser_conn->ib_completion);
685 }
686 
687 static int iser_cma_handler(struct rdma_cm_id *cma_id,
688 			    struct rdma_cm_event *event)
689 {
690 	struct iser_conn *iser_conn;
691 	int ret = 0;
692 
693 	iser_conn = cma_id->context;
694 	iser_info("%s (%d): status %d conn %p id %p\n",
695 		  rdma_event_msg(event->event), event->event,
696 		  event->status, cma_id->context, cma_id);
697 
698 	mutex_lock(&iser_conn->state_mutex);
699 	switch (event->event) {
700 	case RDMA_CM_EVENT_ADDR_RESOLVED:
701 		iser_addr_handler(cma_id);
702 		break;
703 	case RDMA_CM_EVENT_ROUTE_RESOLVED:
704 		iser_route_handler(cma_id);
705 		break;
706 	case RDMA_CM_EVENT_ESTABLISHED:
707 		iser_connected_handler(cma_id, event->param.conn.private_data);
708 		break;
709 	case RDMA_CM_EVENT_REJECTED:
710 		iser_info("Connection rejected: %s\n",
711 			 rdma_reject_msg(cma_id, event->status));
712 		fallthrough;
713 	case RDMA_CM_EVENT_ADDR_ERROR:
714 	case RDMA_CM_EVENT_ROUTE_ERROR:
715 	case RDMA_CM_EVENT_CONNECT_ERROR:
716 	case RDMA_CM_EVENT_UNREACHABLE:
717 		iser_connect_error(cma_id);
718 		break;
719 	case RDMA_CM_EVENT_DISCONNECTED:
720 	case RDMA_CM_EVENT_ADDR_CHANGE:
721 	case RDMA_CM_EVENT_TIMEWAIT_EXIT:
722 		iser_cleanup_handler(cma_id, false);
723 		break;
724 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
725 		/*
726 		 * we *must* destroy the device as we cannot rely
727 		 * on iscsid to be around to initiate error handling.
728 		 * also if we are not in state DOWN implicitly destroy
729 		 * the cma_id.
730 		 */
731 		iser_cleanup_handler(cma_id, true);
732 		if (iser_conn->state != ISER_CONN_DOWN) {
733 			iser_conn->ib_conn.cma_id = NULL;
734 			ret = 1;
735 		}
736 		break;
737 	default:
738 		iser_err("Unexpected RDMA CM event: %s (%d)\n",
739 			 rdma_event_msg(event->event), event->event);
740 		break;
741 	}
742 	mutex_unlock(&iser_conn->state_mutex);
743 
744 	return ret;
745 }
746 
747 void iser_conn_init(struct iser_conn *iser_conn)
748 {
749 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
750 
751 	iser_conn->state = ISER_CONN_INIT;
752 	init_completion(&iser_conn->stop_completion);
753 	init_completion(&iser_conn->ib_completion);
754 	init_completion(&iser_conn->up_completion);
755 	INIT_LIST_HEAD(&iser_conn->conn_list);
756 	mutex_init(&iser_conn->state_mutex);
757 
758 	ib_conn->reg_cqe.done = iser_reg_comp;
759 }
760 
761 /*
762  * starts the process of connecting to the target
763  * sleeps until the connection is established or rejected
764  */
765 int iser_connect(struct iser_conn *iser_conn, struct sockaddr *src_addr,
766 		 struct sockaddr *dst_addr, int non_blocking)
767 {
768 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
769 	int err = 0;
770 
771 	mutex_lock(&iser_conn->state_mutex);
772 
773 	sprintf(iser_conn->name, "%pISp", dst_addr);
774 
775 	iser_info("connecting to: %s\n", iser_conn->name);
776 
777 	/* the device is known only --after-- address resolution */
778 	ib_conn->device = NULL;
779 
780 	iser_conn->state = ISER_CONN_PENDING;
781 
782 	ib_conn->cma_id = rdma_create_id(&init_net, iser_cma_handler,
783 					 iser_conn, RDMA_PS_TCP, IB_QPT_RC);
784 	if (IS_ERR(ib_conn->cma_id)) {
785 		err = PTR_ERR(ib_conn->cma_id);
786 		iser_err("rdma_create_id failed: %d\n", err);
787 		goto id_failure;
788 	}
789 
790 	err = rdma_resolve_addr(ib_conn->cma_id, src_addr, dst_addr, 1000);
791 	if (err) {
792 		iser_err("rdma_resolve_addr failed: %d\n", err);
793 		goto addr_failure;
794 	}
795 
796 	if (!non_blocking) {
797 		wait_for_completion_interruptible(&iser_conn->up_completion);
798 
799 		if (iser_conn->state != ISER_CONN_UP) {
800 			err =  -EIO;
801 			goto connect_failure;
802 		}
803 	}
804 	mutex_unlock(&iser_conn->state_mutex);
805 
806 	mutex_lock(&ig.connlist_mutex);
807 	list_add(&iser_conn->conn_list, &ig.connlist);
808 	mutex_unlock(&ig.connlist_mutex);
809 	return 0;
810 
811 id_failure:
812 	ib_conn->cma_id = NULL;
813 addr_failure:
814 	iser_conn->state = ISER_CONN_DOWN;
815 connect_failure:
816 	mutex_unlock(&iser_conn->state_mutex);
817 	iser_conn_release(iser_conn);
818 	return err;
819 }
820 
821 int iser_post_recvl(struct iser_conn *iser_conn)
822 {
823 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
824 	struct iser_login_desc *desc = &iser_conn->login_desc;
825 	struct ib_recv_wr wr;
826 	int ret;
827 
828 	desc->sge.addr = desc->rsp_dma;
829 	desc->sge.length = ISER_RX_LOGIN_SIZE;
830 	desc->sge.lkey = ib_conn->device->pd->local_dma_lkey;
831 
832 	desc->cqe.done = iser_login_rsp;
833 	wr.wr_cqe = &desc->cqe;
834 	wr.sg_list = &desc->sge;
835 	wr.num_sge = 1;
836 	wr.next = NULL;
837 
838 	ret = ib_post_recv(ib_conn->qp, &wr, NULL);
839 	if (unlikely(ret))
840 		iser_err("ib_post_recv login failed ret=%d\n", ret);
841 
842 	return ret;
843 }
844 
845 int iser_post_recvm(struct iser_conn *iser_conn, struct iser_rx_desc *rx_desc)
846 {
847 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
848 	struct ib_recv_wr wr;
849 	int ret;
850 
851 	rx_desc->cqe.done = iser_task_rsp;
852 	wr.wr_cqe = &rx_desc->cqe;
853 	wr.sg_list = &rx_desc->rx_sg;
854 	wr.num_sge = 1;
855 	wr.next = NULL;
856 
857 	ret = ib_post_recv(ib_conn->qp, &wr, NULL);
858 	if (unlikely(ret))
859 		iser_err("ib_post_recv failed ret=%d\n", ret);
860 
861 	return ret;
862 }
863 
864 
865 /**
866  * iser_post_send - Initiate a Send DTO operation
867  * @ib_conn: connection RDMA resources
868  * @tx_desc: iSER TX descriptor
869  *
870  * Return: 0 on success, -1 on failure
871  */
872 int iser_post_send(struct ib_conn *ib_conn, struct iser_tx_desc *tx_desc)
873 {
874 	struct ib_send_wr *wr = &tx_desc->send_wr;
875 	struct ib_send_wr *first_wr;
876 	int ret;
877 
878 	ib_dma_sync_single_for_device(ib_conn->device->ib_device,
879 				      tx_desc->dma_addr, ISER_HEADERS_LEN,
880 				      DMA_TO_DEVICE);
881 
882 	wr->next = NULL;
883 	wr->wr_cqe = &tx_desc->cqe;
884 	wr->sg_list = tx_desc->tx_sg;
885 	wr->num_sge = tx_desc->num_sge;
886 	wr->opcode = IB_WR_SEND;
887 	wr->send_flags = IB_SEND_SIGNALED;
888 
889 	if (tx_desc->inv_wr.next)
890 		first_wr = &tx_desc->inv_wr;
891 	else if (tx_desc->reg_wr.wr.next)
892 		first_wr = &tx_desc->reg_wr.wr;
893 	else
894 		first_wr = wr;
895 
896 	ret = ib_post_send(ib_conn->qp, first_wr, NULL);
897 	if (unlikely(ret))
898 		iser_err("ib_post_send failed, ret:%d opcode:%d\n",
899 			 ret, wr->opcode);
900 
901 	return ret;
902 }
903 
904 u8 iser_check_task_pi_status(struct iscsi_iser_task *iser_task,
905 			     enum iser_data_dir cmd_dir, sector_t *sector)
906 {
907 	struct iser_mem_reg *reg = &iser_task->rdma_reg[cmd_dir];
908 	struct iser_fr_desc *desc = reg->mem_h;
909 	unsigned long sector_size = iser_task->sc->device->sector_size;
910 	struct ib_mr_status mr_status;
911 	int ret;
912 
913 	if (desc && desc->sig_protected) {
914 		desc->sig_protected = false;
915 		ret = ib_check_mr_status(desc->rsc.sig_mr,
916 					 IB_MR_CHECK_SIG_STATUS, &mr_status);
917 		if (ret) {
918 			iser_err("ib_check_mr_status failed, ret %d\n", ret);
919 			/* Not a lot we can do, return ambiguous guard error */
920 			*sector = 0;
921 			return 0x1;
922 		}
923 
924 		if (mr_status.fail_status & IB_MR_CHECK_SIG_STATUS) {
925 			sector_t sector_off = mr_status.sig_err.sig_err_offset;
926 
927 			sector_div(sector_off, sector_size + 8);
928 			*sector = scsi_get_sector(iser_task->sc) + sector_off;
929 
930 			iser_err("PI error found type %d at sector %llx "
931 			       "expected %x vs actual %x\n",
932 			       mr_status.sig_err.err_type,
933 			       (unsigned long long)*sector,
934 			       mr_status.sig_err.expected,
935 			       mr_status.sig_err.actual);
936 
937 			switch (mr_status.sig_err.err_type) {
938 			case IB_SIG_BAD_GUARD:
939 				return 0x1;
940 			case IB_SIG_BAD_REFTAG:
941 				return 0x3;
942 			case IB_SIG_BAD_APPTAG:
943 				return 0x2;
944 			}
945 		}
946 	}
947 
948 	return 0;
949 }
950 
951 void iser_err_comp(struct ib_wc *wc, const char *type)
952 {
953 	if (wc->status != IB_WC_WR_FLUSH_ERR) {
954 		struct iser_conn *iser_conn = to_iser_conn(wc->qp->qp_context);
955 
956 		iser_err("%s failure: %s (%d) vend_err %#x\n", type,
957 			 ib_wc_status_msg(wc->status), wc->status,
958 			 wc->vendor_err);
959 
960 		if (iser_conn->iscsi_conn)
961 			iscsi_conn_failure(iser_conn->iscsi_conn,
962 					   ISCSI_ERR_CONN_FAILED);
963 	} else {
964 		iser_dbg("%s failure: %s (%d)\n", type,
965 			 ib_wc_status_msg(wc->status), wc->status);
966 	}
967 }
968