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 	int i, max_cqe;
72 
73 	if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS)) {
74 		iser_err("IB device does not support memory registrations\n");
75 		return -1;
76 	}
77 
78 	device->comps_used = min_t(int, num_online_cpus(),
79 				 ib_dev->num_comp_vectors);
80 
81 	device->comps = kcalloc(device->comps_used, sizeof(*device->comps),
82 				GFP_KERNEL);
83 	if (!device->comps)
84 		goto comps_err;
85 
86 	max_cqe = min(ISER_MAX_CQ_LEN, ib_dev->attrs.max_cqe);
87 
88 	iser_info("using %d CQs, device %s supports %d vectors max_cqe %d\n",
89 		  device->comps_used, dev_name(&ib_dev->dev),
90 		  ib_dev->num_comp_vectors, max_cqe);
91 
92 	device->pd = ib_alloc_pd(ib_dev,
93 		iser_always_reg ? 0 : IB_PD_UNSAFE_GLOBAL_RKEY);
94 	if (IS_ERR(device->pd))
95 		goto pd_err;
96 
97 	for (i = 0; i < device->comps_used; i++) {
98 		struct iser_comp *comp = &device->comps[i];
99 
100 		comp->cq = ib_alloc_cq(ib_dev, comp, max_cqe, i,
101 				       IB_POLL_SOFTIRQ);
102 		if (IS_ERR(comp->cq)) {
103 			comp->cq = NULL;
104 			goto cq_err;
105 		}
106 	}
107 
108 	INIT_IB_EVENT_HANDLER(&device->event_handler, ib_dev,
109 			      iser_event_handler);
110 	ib_register_event_handler(&device->event_handler);
111 	return 0;
112 
113 cq_err:
114 	for (i = 0; i < device->comps_used; i++) {
115 		struct iser_comp *comp = &device->comps[i];
116 
117 		if (comp->cq)
118 			ib_free_cq(comp->cq);
119 	}
120 	ib_dealloc_pd(device->pd);
121 pd_err:
122 	kfree(device->comps);
123 comps_err:
124 	iser_err("failed to allocate an IB resource\n");
125 	return -1;
126 }
127 
128 /*
129  * iser_free_device_ib_res - destroy/dealloc/dereg the DMA MR,
130  * CQ and PD created with the device associated with the adaptor.
131  */
132 static void iser_free_device_ib_res(struct iser_device *device)
133 {
134 	int i;
135 
136 	for (i = 0; i < device->comps_used; i++) {
137 		struct iser_comp *comp = &device->comps[i];
138 
139 		ib_free_cq(comp->cq);
140 		comp->cq = NULL;
141 	}
142 
143 	ib_unregister_event_handler(&device->event_handler);
144 	ib_dealloc_pd(device->pd);
145 
146 	kfree(device->comps);
147 	device->comps = NULL;
148 	device->pd = NULL;
149 }
150 
151 static struct iser_fr_desc *
152 iser_create_fastreg_desc(struct iser_device *device,
153 			 struct ib_pd *pd,
154 			 bool pi_enable,
155 			 unsigned int size)
156 {
157 	struct iser_fr_desc *desc;
158 	struct ib_device *ib_dev = device->ib_device;
159 	enum ib_mr_type mr_type;
160 	int ret;
161 
162 	desc = kzalloc(sizeof(*desc), GFP_KERNEL);
163 	if (!desc)
164 		return ERR_PTR(-ENOMEM);
165 
166 	if (ib_dev->attrs.device_cap_flags & IB_DEVICE_SG_GAPS_REG)
167 		mr_type = IB_MR_TYPE_SG_GAPS;
168 	else
169 		mr_type = IB_MR_TYPE_MEM_REG;
170 
171 	desc->rsc.mr = ib_alloc_mr(pd, mr_type, size);
172 	if (IS_ERR(desc->rsc.mr)) {
173 		ret = PTR_ERR(desc->rsc.mr);
174 		iser_err("Failed to allocate ib_fast_reg_mr err=%d\n", ret);
175 		goto err_alloc_mr;
176 	}
177 
178 	if (pi_enable) {
179 		desc->rsc.sig_mr = ib_alloc_mr_integrity(pd, size, size);
180 		if (IS_ERR(desc->rsc.sig_mr)) {
181 			ret = PTR_ERR(desc->rsc.sig_mr);
182 			iser_err("Failed to allocate sig_mr err=%d\n", ret);
183 			goto err_alloc_mr_integrity;
184 		}
185 	}
186 	desc->rsc.mr_valid = 0;
187 
188 	return desc;
189 
190 err_alloc_mr_integrity:
191 	ib_dereg_mr(desc->rsc.mr);
192 err_alloc_mr:
193 	kfree(desc);
194 
195 	return ERR_PTR(ret);
196 }
197 
198 static void iser_destroy_fastreg_desc(struct iser_fr_desc *desc)
199 {
200 	struct iser_reg_resources *res = &desc->rsc;
201 
202 	ib_dereg_mr(res->mr);
203 	if (res->sig_mr) {
204 		ib_dereg_mr(res->sig_mr);
205 		res->sig_mr = NULL;
206 	}
207 	kfree(desc);
208 }
209 
210 /**
211  * iser_alloc_fastreg_pool - Creates pool of fast_reg descriptors
212  * for fast registration work requests.
213  * @ib_conn: connection RDMA resources
214  * @cmds_max: max number of SCSI commands for this connection
215  * @size: max number of pages per map request
216  *
217  * Return: 0 on success, or errno code on failure
218  */
219 int iser_alloc_fastreg_pool(struct ib_conn *ib_conn,
220 			    unsigned cmds_max,
221 			    unsigned int size)
222 {
223 	struct iser_device *device = ib_conn->device;
224 	struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
225 	struct iser_fr_desc *desc;
226 	int i, ret;
227 
228 	INIT_LIST_HEAD(&fr_pool->list);
229 	INIT_LIST_HEAD(&fr_pool->all_list);
230 	spin_lock_init(&fr_pool->lock);
231 	fr_pool->size = 0;
232 	for (i = 0; i < cmds_max; i++) {
233 		desc = iser_create_fastreg_desc(device, device->pd,
234 						ib_conn->pi_support, size);
235 		if (IS_ERR(desc)) {
236 			ret = PTR_ERR(desc);
237 			goto err;
238 		}
239 
240 		list_add_tail(&desc->list, &fr_pool->list);
241 		list_add_tail(&desc->all_list, &fr_pool->all_list);
242 		fr_pool->size++;
243 	}
244 
245 	return 0;
246 
247 err:
248 	iser_free_fastreg_pool(ib_conn);
249 	return ret;
250 }
251 
252 /**
253  * iser_free_fastreg_pool - releases the pool of fast_reg descriptors
254  * @ib_conn: connection RDMA resources
255  */
256 void iser_free_fastreg_pool(struct ib_conn *ib_conn)
257 {
258 	struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
259 	struct iser_fr_desc *desc, *tmp;
260 	int i = 0;
261 
262 	if (list_empty(&fr_pool->all_list))
263 		return;
264 
265 	iser_info("freeing conn %p fr pool\n", ib_conn);
266 
267 	list_for_each_entry_safe(desc, tmp, &fr_pool->all_list, all_list) {
268 		list_del(&desc->all_list);
269 		iser_destroy_fastreg_desc(desc);
270 		++i;
271 	}
272 
273 	if (i < fr_pool->size)
274 		iser_warn("pool still has %d regions registered\n",
275 			  fr_pool->size - i);
276 }
277 
278 /*
279  * iser_create_ib_conn_res - Queue-Pair (QP)
280  *
281  * Return: 0 on success, -1 on failure
282  */
283 static int iser_create_ib_conn_res(struct ib_conn *ib_conn)
284 {
285 	struct iser_conn *iser_conn = to_iser_conn(ib_conn);
286 	struct iser_device	*device;
287 	struct ib_device	*ib_dev;
288 	struct ib_qp_init_attr	init_attr;
289 	int			ret = -ENOMEM;
290 	int index, min_index = 0;
291 
292 	BUG_ON(ib_conn->device == NULL);
293 
294 	device = ib_conn->device;
295 	ib_dev = device->ib_device;
296 
297 	memset(&init_attr, 0, sizeof init_attr);
298 
299 	mutex_lock(&ig.connlist_mutex);
300 	/* select the CQ with the minimal number of usages */
301 	for (index = 0; index < device->comps_used; index++) {
302 		if (device->comps[index].active_qps <
303 		    device->comps[min_index].active_qps)
304 			min_index = index;
305 	}
306 	ib_conn->comp = &device->comps[min_index];
307 	ib_conn->comp->active_qps++;
308 	mutex_unlock(&ig.connlist_mutex);
309 	iser_info("cq index %d used for ib_conn %p\n", min_index, ib_conn);
310 
311 	init_attr.event_handler = iser_qp_event_callback;
312 	init_attr.qp_context	= (void *)ib_conn;
313 	init_attr.send_cq	= ib_conn->comp->cq;
314 	init_attr.recv_cq	= ib_conn->comp->cq;
315 	init_attr.cap.max_recv_wr  = ISER_QP_MAX_RECV_DTOS;
316 	init_attr.cap.max_send_sge = 2;
317 	init_attr.cap.max_recv_sge = 1;
318 	init_attr.sq_sig_type	= IB_SIGNAL_REQ_WR;
319 	init_attr.qp_type	= IB_QPT_RC;
320 	if (ib_conn->pi_support) {
321 		init_attr.cap.max_send_wr = ISER_QP_SIG_MAX_REQ_DTOS + 1;
322 		init_attr.create_flags |= IB_QP_CREATE_INTEGRITY_EN;
323 		iser_conn->max_cmds =
324 			ISER_GET_MAX_XMIT_CMDS(ISER_QP_SIG_MAX_REQ_DTOS);
325 	} else {
326 		if (ib_dev->attrs.max_qp_wr > ISER_QP_MAX_REQ_DTOS) {
327 			init_attr.cap.max_send_wr  = ISER_QP_MAX_REQ_DTOS + 1;
328 			iser_conn->max_cmds =
329 				ISER_GET_MAX_XMIT_CMDS(ISER_QP_MAX_REQ_DTOS);
330 		} else {
331 			init_attr.cap.max_send_wr = ib_dev->attrs.max_qp_wr;
332 			iser_conn->max_cmds =
333 				ISER_GET_MAX_XMIT_CMDS(ib_dev->attrs.max_qp_wr);
334 			iser_dbg("device %s supports max_send_wr %d\n",
335 				 dev_name(&device->ib_device->dev),
336 				 ib_dev->attrs.max_qp_wr);
337 		}
338 	}
339 
340 	ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr);
341 	if (ret)
342 		goto out_err;
343 
344 	ib_conn->qp = ib_conn->cma_id->qp;
345 	iser_info("setting conn %p cma_id %p qp %p\n",
346 		  ib_conn, ib_conn->cma_id,
347 		  ib_conn->cma_id->qp);
348 	return ret;
349 
350 out_err:
351 	mutex_lock(&ig.connlist_mutex);
352 	ib_conn->comp->active_qps--;
353 	mutex_unlock(&ig.connlist_mutex);
354 	iser_err("unable to alloc mem or create resource, err %d\n", ret);
355 
356 	return ret;
357 }
358 
359 /*
360  * based on the resolved device node GUID see if there already allocated
361  * device for this device. If there's no such, create one.
362  */
363 static
364 struct iser_device *iser_device_find_by_ib_device(struct rdma_cm_id *cma_id)
365 {
366 	struct iser_device *device;
367 
368 	mutex_lock(&ig.device_list_mutex);
369 
370 	list_for_each_entry(device, &ig.device_list, ig_list)
371 		/* find if there's a match using the node GUID */
372 		if (device->ib_device->node_guid == cma_id->device->node_guid)
373 			goto inc_refcnt;
374 
375 	device = kzalloc(sizeof *device, GFP_KERNEL);
376 	if (device == NULL)
377 		goto out;
378 
379 	/* assign this device to the device */
380 	device->ib_device = cma_id->device;
381 	/* init the device and link it into ig device list */
382 	if (iser_create_device_ib_res(device)) {
383 		kfree(device);
384 		device = NULL;
385 		goto out;
386 	}
387 	list_add(&device->ig_list, &ig.device_list);
388 
389 inc_refcnt:
390 	device->refcount++;
391 out:
392 	mutex_unlock(&ig.device_list_mutex);
393 	return device;
394 }
395 
396 /* if there's no demand for this device, release it */
397 static void iser_device_try_release(struct iser_device *device)
398 {
399 	mutex_lock(&ig.device_list_mutex);
400 	device->refcount--;
401 	iser_info("device %p refcount %d\n", device, device->refcount);
402 	if (!device->refcount) {
403 		iser_free_device_ib_res(device);
404 		list_del(&device->ig_list);
405 		kfree(device);
406 	}
407 	mutex_unlock(&ig.device_list_mutex);
408 }
409 
410 /*
411  * Called with state mutex held
412  */
413 static int iser_conn_state_comp_exch(struct iser_conn *iser_conn,
414 				     enum iser_conn_state comp,
415 				     enum iser_conn_state exch)
416 {
417 	int ret;
418 
419 	ret = (iser_conn->state == comp);
420 	if (ret)
421 		iser_conn->state = exch;
422 
423 	return ret;
424 }
425 
426 void iser_release_work(struct work_struct *work)
427 {
428 	struct iser_conn *iser_conn;
429 
430 	iser_conn = container_of(work, struct iser_conn, release_work);
431 
432 	/* Wait for conn_stop to complete */
433 	wait_for_completion(&iser_conn->stop_completion);
434 	/* Wait for IB resouces cleanup to complete */
435 	wait_for_completion(&iser_conn->ib_completion);
436 
437 	mutex_lock(&iser_conn->state_mutex);
438 	iser_conn->state = ISER_CONN_DOWN;
439 	mutex_unlock(&iser_conn->state_mutex);
440 
441 	iser_conn_release(iser_conn);
442 }
443 
444 /**
445  * iser_free_ib_conn_res - release IB related resources
446  * @iser_conn: iser connection struct
447  * @destroy: indicator if we need to try to release the
448  *     iser device and memory regoins pool (only iscsi
449  *     shutdown and DEVICE_REMOVAL will use this).
450  *
451  * This routine is called with the iser state mutex held
452  * so the cm_id removal is out of here. It is Safe to
453  * be invoked multiple times.
454  */
455 static void iser_free_ib_conn_res(struct iser_conn *iser_conn,
456 				  bool destroy)
457 {
458 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
459 	struct iser_device *device = ib_conn->device;
460 
461 	iser_info("freeing conn %p cma_id %p qp %p\n",
462 		  iser_conn, ib_conn->cma_id, ib_conn->qp);
463 
464 	if (ib_conn->qp != NULL) {
465 		mutex_lock(&ig.connlist_mutex);
466 		ib_conn->comp->active_qps--;
467 		mutex_unlock(&ig.connlist_mutex);
468 		rdma_destroy_qp(ib_conn->cma_id);
469 		ib_conn->qp = NULL;
470 	}
471 
472 	if (destroy) {
473 		if (iser_conn->rx_descs)
474 			iser_free_rx_descriptors(iser_conn);
475 
476 		if (device != NULL) {
477 			iser_device_try_release(device);
478 			ib_conn->device = NULL;
479 		}
480 	}
481 }
482 
483 /**
484  * iser_conn_release - Frees all conn objects and deallocs conn descriptor
485  * @iser_conn: iSER connection context
486  */
487 void iser_conn_release(struct iser_conn *iser_conn)
488 {
489 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
490 
491 	mutex_lock(&ig.connlist_mutex);
492 	list_del(&iser_conn->conn_list);
493 	mutex_unlock(&ig.connlist_mutex);
494 
495 	mutex_lock(&iser_conn->state_mutex);
496 	/* In case we endup here without ep_disconnect being invoked. */
497 	if (iser_conn->state != ISER_CONN_DOWN) {
498 		iser_warn("iser conn %p state %d, expected state down.\n",
499 			  iser_conn, iser_conn->state);
500 		iscsi_destroy_endpoint(iser_conn->ep);
501 		iser_conn->state = ISER_CONN_DOWN;
502 	}
503 	/*
504 	 * In case we never got to bind stage, we still need to
505 	 * release IB resources (which is safe to call more than once).
506 	 */
507 	iser_free_ib_conn_res(iser_conn, true);
508 	mutex_unlock(&iser_conn->state_mutex);
509 
510 	if (ib_conn->cma_id != NULL) {
511 		rdma_destroy_id(ib_conn->cma_id);
512 		ib_conn->cma_id = NULL;
513 	}
514 
515 	kfree(iser_conn);
516 }
517 
518 /**
519  * iser_conn_terminate - triggers start of the disconnect procedures and
520  * waits for them to be done
521  * @iser_conn: iSER connection context
522  *
523  * Called with state mutex held
524  */
525 int iser_conn_terminate(struct iser_conn *iser_conn)
526 {
527 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
528 	int err = 0;
529 
530 	/* terminate the iser conn only if the conn state is UP */
531 	if (!iser_conn_state_comp_exch(iser_conn, ISER_CONN_UP,
532 				       ISER_CONN_TERMINATING))
533 		return 0;
534 
535 	iser_info("iser_conn %p state %d\n", iser_conn, iser_conn->state);
536 
537 	/* suspend queuing of new iscsi commands */
538 	if (iser_conn->iscsi_conn)
539 		iscsi_suspend_queue(iser_conn->iscsi_conn);
540 
541 	/*
542 	 * In case we didn't already clean up the cma_id (peer initiated
543 	 * a disconnection), we need to Cause the CMA to change the QP
544 	 * state to ERROR.
545 	 */
546 	if (ib_conn->cma_id) {
547 		err = rdma_disconnect(ib_conn->cma_id);
548 		if (err)
549 			iser_err("Failed to disconnect, conn: 0x%p err %d\n",
550 				 iser_conn, err);
551 
552 		/* block until all flush errors are consumed */
553 		ib_drain_sq(ib_conn->qp);
554 	}
555 
556 	return 1;
557 }
558 
559 /*
560  * Called with state mutex held
561  */
562 static void iser_connect_error(struct rdma_cm_id *cma_id)
563 {
564 	struct iser_conn *iser_conn;
565 
566 	iser_conn = (struct iser_conn *)cma_id->context;
567 	iser_conn->state = ISER_CONN_TERMINATING;
568 }
569 
570 static void
571 iser_calc_scsi_params(struct iser_conn *iser_conn,
572 		      unsigned int max_sectors)
573 {
574 	struct iser_device *device = iser_conn->ib_conn.device;
575 	struct ib_device_attr *attr = &device->ib_device->attrs;
576 	unsigned short sg_tablesize, sup_sg_tablesize;
577 	unsigned short reserved_mr_pages;
578 	u32 max_num_sg;
579 
580 	/*
581 	 * FRs without SG_GAPS can only map up to a (device) page per entry,
582 	 * but if the first entry is misaligned we'll end up using two entries
583 	 * (head and tail) for a single page worth data, so one additional
584 	 * entry is required.
585 	 */
586 	if (attr->device_cap_flags & IB_DEVICE_SG_GAPS_REG)
587 		reserved_mr_pages = 0;
588 	else
589 		reserved_mr_pages = 1;
590 
591 	if (iser_conn->ib_conn.pi_support)
592 		max_num_sg = attr->max_pi_fast_reg_page_list_len;
593 	else
594 		max_num_sg = attr->max_fast_reg_page_list_len;
595 
596 	sg_tablesize = DIV_ROUND_UP(max_sectors * SECTOR_SIZE, SZ_4K);
597 	sup_sg_tablesize = min_t(uint, ISCSI_ISER_MAX_SG_TABLESIZE,
598 				 max_num_sg - reserved_mr_pages);
599 	iser_conn->scsi_sg_tablesize = min(sg_tablesize, sup_sg_tablesize);
600 	iser_conn->pages_per_mr =
601 		iser_conn->scsi_sg_tablesize + reserved_mr_pages;
602 }
603 
604 /*
605  * Called with state mutex held
606  */
607 static void iser_addr_handler(struct rdma_cm_id *cma_id)
608 {
609 	struct iser_device *device;
610 	struct iser_conn   *iser_conn;
611 	struct ib_conn   *ib_conn;
612 	int    ret;
613 
614 	iser_conn = (struct iser_conn *)cma_id->context;
615 	if (iser_conn->state != ISER_CONN_PENDING)
616 		/* bailout */
617 		return;
618 
619 	ib_conn = &iser_conn->ib_conn;
620 	device = iser_device_find_by_ib_device(cma_id);
621 	if (!device) {
622 		iser_err("device lookup/creation failed\n");
623 		iser_connect_error(cma_id);
624 		return;
625 	}
626 
627 	ib_conn->device = device;
628 
629 	/* connection T10-PI support */
630 	if (iser_pi_enable) {
631 		if (!(device->ib_device->attrs.device_cap_flags &
632 		      IB_DEVICE_INTEGRITY_HANDOVER)) {
633 			iser_warn("T10-PI requested but not supported on %s, "
634 				  "continue without T10-PI\n",
635 				  dev_name(&ib_conn->device->ib_device->dev));
636 			ib_conn->pi_support = false;
637 		} else {
638 			ib_conn->pi_support = true;
639 		}
640 	}
641 
642 	iser_calc_scsi_params(iser_conn, iser_max_sectors);
643 
644 	ret = rdma_resolve_route(cma_id, 1000);
645 	if (ret) {
646 		iser_err("resolve route failed: %d\n", ret);
647 		iser_connect_error(cma_id);
648 		return;
649 	}
650 }
651 
652 /*
653  * Called with state mutex held
654  */
655 static void iser_route_handler(struct rdma_cm_id *cma_id)
656 {
657 	struct rdma_conn_param conn_param;
658 	int    ret;
659 	struct iser_cm_hdr req_hdr;
660 	struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context;
661 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
662 	struct ib_device *ib_dev = ib_conn->device->ib_device;
663 
664 	if (iser_conn->state != ISER_CONN_PENDING)
665 		/* bailout */
666 		return;
667 
668 	ret = iser_create_ib_conn_res(ib_conn);
669 	if (ret)
670 		goto failure;
671 
672 	memset(&conn_param, 0, sizeof conn_param);
673 	conn_param.responder_resources = ib_dev->attrs.max_qp_rd_atom;
674 	conn_param.initiator_depth     = 1;
675 	conn_param.retry_count	       = 7;
676 	conn_param.rnr_retry_count     = 6;
677 
678 	memset(&req_hdr, 0, sizeof(req_hdr));
679 	req_hdr.flags = ISER_ZBVA_NOT_SUP;
680 	if (!iser_always_reg)
681 		req_hdr.flags |= ISER_SEND_W_INV_NOT_SUP;
682 	conn_param.private_data	= (void *)&req_hdr;
683 	conn_param.private_data_len = sizeof(struct iser_cm_hdr);
684 
685 	ret = rdma_connect(cma_id, &conn_param);
686 	if (ret) {
687 		iser_err("failure connecting: %d\n", ret);
688 		goto failure;
689 	}
690 
691 	return;
692 failure:
693 	iser_connect_error(cma_id);
694 }
695 
696 static void iser_connected_handler(struct rdma_cm_id *cma_id,
697 				   const void *private_data)
698 {
699 	struct iser_conn *iser_conn;
700 	struct ib_qp_attr attr;
701 	struct ib_qp_init_attr init_attr;
702 
703 	iser_conn = (struct iser_conn *)cma_id->context;
704 	if (iser_conn->state != ISER_CONN_PENDING)
705 		/* bailout */
706 		return;
707 
708 	(void)ib_query_qp(cma_id->qp, &attr, ~0, &init_attr);
709 	iser_info("remote qpn:%x my qpn:%x\n", attr.dest_qp_num, cma_id->qp->qp_num);
710 
711 	if (private_data) {
712 		u8 flags = *(u8 *)private_data;
713 
714 		iser_conn->snd_w_inv = !(flags & ISER_SEND_W_INV_NOT_SUP);
715 	}
716 
717 	iser_info("conn %p: negotiated %s invalidation\n",
718 		  iser_conn, iser_conn->snd_w_inv ? "remote" : "local");
719 
720 	iser_conn->state = ISER_CONN_UP;
721 	complete(&iser_conn->up_completion);
722 }
723 
724 static void iser_disconnected_handler(struct rdma_cm_id *cma_id)
725 {
726 	struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context;
727 
728 	if (iser_conn_terminate(iser_conn)) {
729 		if (iser_conn->iscsi_conn)
730 			iscsi_conn_failure(iser_conn->iscsi_conn,
731 					   ISCSI_ERR_CONN_FAILED);
732 		else
733 			iser_err("iscsi_iser connection isn't bound\n");
734 	}
735 }
736 
737 static void iser_cleanup_handler(struct rdma_cm_id *cma_id,
738 				 bool destroy)
739 {
740 	struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context;
741 
742 	/*
743 	 * We are not guaranteed that we visited disconnected_handler
744 	 * by now, call it here to be safe that we handle CM drep
745 	 * and flush errors.
746 	 */
747 	iser_disconnected_handler(cma_id);
748 	iser_free_ib_conn_res(iser_conn, destroy);
749 	complete(&iser_conn->ib_completion);
750 };
751 
752 static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
753 {
754 	struct iser_conn *iser_conn;
755 	int ret = 0;
756 
757 	iser_conn = (struct iser_conn *)cma_id->context;
758 	iser_info("%s (%d): status %d conn %p id %p\n",
759 		  rdma_event_msg(event->event), event->event,
760 		  event->status, cma_id->context, cma_id);
761 
762 	mutex_lock(&iser_conn->state_mutex);
763 	switch (event->event) {
764 	case RDMA_CM_EVENT_ADDR_RESOLVED:
765 		iser_addr_handler(cma_id);
766 		break;
767 	case RDMA_CM_EVENT_ROUTE_RESOLVED:
768 		iser_route_handler(cma_id);
769 		break;
770 	case RDMA_CM_EVENT_ESTABLISHED:
771 		iser_connected_handler(cma_id, event->param.conn.private_data);
772 		break;
773 	case RDMA_CM_EVENT_REJECTED:
774 		iser_info("Connection rejected: %s\n",
775 			 rdma_reject_msg(cma_id, event->status));
776 		/* FALLTHROUGH */
777 	case RDMA_CM_EVENT_ADDR_ERROR:
778 	case RDMA_CM_EVENT_ROUTE_ERROR:
779 	case RDMA_CM_EVENT_CONNECT_ERROR:
780 	case RDMA_CM_EVENT_UNREACHABLE:
781 		iser_connect_error(cma_id);
782 		break;
783 	case RDMA_CM_EVENT_DISCONNECTED:
784 	case RDMA_CM_EVENT_ADDR_CHANGE:
785 	case RDMA_CM_EVENT_TIMEWAIT_EXIT:
786 		iser_cleanup_handler(cma_id, false);
787 		break;
788 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
789 		/*
790 		 * we *must* destroy the device as we cannot rely
791 		 * on iscsid to be around to initiate error handling.
792 		 * also if we are not in state DOWN implicitly destroy
793 		 * the cma_id.
794 		 */
795 		iser_cleanup_handler(cma_id, true);
796 		if (iser_conn->state != ISER_CONN_DOWN) {
797 			iser_conn->ib_conn.cma_id = NULL;
798 			ret = 1;
799 		}
800 		break;
801 	default:
802 		iser_err("Unexpected RDMA CM event: %s (%d)\n",
803 			 rdma_event_msg(event->event), event->event);
804 		break;
805 	}
806 	mutex_unlock(&iser_conn->state_mutex);
807 
808 	return ret;
809 }
810 
811 void iser_conn_init(struct iser_conn *iser_conn)
812 {
813 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
814 
815 	iser_conn->state = ISER_CONN_INIT;
816 	init_completion(&iser_conn->stop_completion);
817 	init_completion(&iser_conn->ib_completion);
818 	init_completion(&iser_conn->up_completion);
819 	INIT_LIST_HEAD(&iser_conn->conn_list);
820 	mutex_init(&iser_conn->state_mutex);
821 
822 	ib_conn->post_recv_buf_count = 0;
823 	ib_conn->reg_cqe.done = iser_reg_comp;
824 }
825 
826  /**
827  * starts the process of connecting to the target
828  * sleeps until the connection is established or rejected
829  */
830 int iser_connect(struct iser_conn   *iser_conn,
831 		 struct sockaddr    *src_addr,
832 		 struct sockaddr    *dst_addr,
833 		 int                 non_blocking)
834 {
835 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
836 	int err = 0;
837 
838 	mutex_lock(&iser_conn->state_mutex);
839 
840 	sprintf(iser_conn->name, "%pISp", dst_addr);
841 
842 	iser_info("connecting to: %s\n", iser_conn->name);
843 
844 	/* the device is known only --after-- address resolution */
845 	ib_conn->device = NULL;
846 
847 	iser_conn->state = ISER_CONN_PENDING;
848 
849 	ib_conn->cma_id = rdma_create_id(&init_net, iser_cma_handler,
850 					 (void *)iser_conn,
851 					 RDMA_PS_TCP, IB_QPT_RC);
852 	if (IS_ERR(ib_conn->cma_id)) {
853 		err = PTR_ERR(ib_conn->cma_id);
854 		iser_err("rdma_create_id failed: %d\n", err);
855 		goto id_failure;
856 	}
857 
858 	err = rdma_resolve_addr(ib_conn->cma_id, src_addr, dst_addr, 1000);
859 	if (err) {
860 		iser_err("rdma_resolve_addr failed: %d\n", err);
861 		goto addr_failure;
862 	}
863 
864 	if (!non_blocking) {
865 		wait_for_completion_interruptible(&iser_conn->up_completion);
866 
867 		if (iser_conn->state != ISER_CONN_UP) {
868 			err =  -EIO;
869 			goto connect_failure;
870 		}
871 	}
872 	mutex_unlock(&iser_conn->state_mutex);
873 
874 	mutex_lock(&ig.connlist_mutex);
875 	list_add(&iser_conn->conn_list, &ig.connlist);
876 	mutex_unlock(&ig.connlist_mutex);
877 	return 0;
878 
879 id_failure:
880 	ib_conn->cma_id = NULL;
881 addr_failure:
882 	iser_conn->state = ISER_CONN_DOWN;
883 connect_failure:
884 	mutex_unlock(&iser_conn->state_mutex);
885 	iser_conn_release(iser_conn);
886 	return err;
887 }
888 
889 int iser_post_recvl(struct iser_conn *iser_conn)
890 {
891 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
892 	struct iser_login_desc *desc = &iser_conn->login_desc;
893 	struct ib_recv_wr wr;
894 	int ib_ret;
895 
896 	desc->sge.addr = desc->rsp_dma;
897 	desc->sge.length = ISER_RX_LOGIN_SIZE;
898 	desc->sge.lkey = ib_conn->device->pd->local_dma_lkey;
899 
900 	desc->cqe.done = iser_login_rsp;
901 	wr.wr_cqe = &desc->cqe;
902 	wr.sg_list = &desc->sge;
903 	wr.num_sge = 1;
904 	wr.next = NULL;
905 
906 	ib_conn->post_recv_buf_count++;
907 	ib_ret = ib_post_recv(ib_conn->qp, &wr, NULL);
908 	if (ib_ret) {
909 		iser_err("ib_post_recv failed ret=%d\n", ib_ret);
910 		ib_conn->post_recv_buf_count--;
911 	}
912 
913 	return ib_ret;
914 }
915 
916 int iser_post_recvm(struct iser_conn *iser_conn, int count)
917 {
918 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
919 	unsigned int my_rx_head = iser_conn->rx_desc_head;
920 	struct iser_rx_desc *rx_desc;
921 	struct ib_recv_wr *wr;
922 	int i, ib_ret;
923 
924 	for (wr = ib_conn->rx_wr, i = 0; i < count; i++, wr++) {
925 		rx_desc = &iser_conn->rx_descs[my_rx_head];
926 		rx_desc->cqe.done = iser_task_rsp;
927 		wr->wr_cqe = &rx_desc->cqe;
928 		wr->sg_list = &rx_desc->rx_sg;
929 		wr->num_sge = 1;
930 		wr->next = wr + 1;
931 		my_rx_head = (my_rx_head + 1) & iser_conn->qp_max_recv_dtos_mask;
932 	}
933 
934 	wr--;
935 	wr->next = NULL; /* mark end of work requests list */
936 
937 	ib_conn->post_recv_buf_count += count;
938 	ib_ret = ib_post_recv(ib_conn->qp, ib_conn->rx_wr, NULL);
939 	if (unlikely(ib_ret)) {
940 		iser_err("ib_post_recv failed ret=%d\n", ib_ret);
941 		ib_conn->post_recv_buf_count -= count;
942 	} else
943 		iser_conn->rx_desc_head = my_rx_head;
944 
945 	return ib_ret;
946 }
947 
948 
949 /**
950  * iser_post_send - Initiate a Send DTO operation
951  * @ib_conn: connection RDMA resources
952  * @tx_desc: iSER TX descriptor
953  * @signal: true to send work request as SIGNALED
954  *
955  * Return: 0 on success, -1 on failure
956  */
957 int iser_post_send(struct ib_conn *ib_conn, struct iser_tx_desc *tx_desc,
958 		   bool signal)
959 {
960 	struct ib_send_wr *wr = &tx_desc->send_wr;
961 	struct ib_send_wr *first_wr;
962 	int ib_ret;
963 
964 	ib_dma_sync_single_for_device(ib_conn->device->ib_device,
965 				      tx_desc->dma_addr, ISER_HEADERS_LEN,
966 				      DMA_TO_DEVICE);
967 
968 	wr->next = NULL;
969 	wr->wr_cqe = &tx_desc->cqe;
970 	wr->sg_list = tx_desc->tx_sg;
971 	wr->num_sge = tx_desc->num_sge;
972 	wr->opcode = IB_WR_SEND;
973 	wr->send_flags = signal ? IB_SEND_SIGNALED : 0;
974 
975 	if (tx_desc->inv_wr.next)
976 		first_wr = &tx_desc->inv_wr;
977 	else if (tx_desc->reg_wr.wr.next)
978 		first_wr = &tx_desc->reg_wr.wr;
979 	else
980 		first_wr = wr;
981 
982 	ib_ret = ib_post_send(ib_conn->qp, first_wr, NULL);
983 	if (unlikely(ib_ret))
984 		iser_err("ib_post_send failed, ret:%d opcode:%d\n",
985 			 ib_ret, wr->opcode);
986 
987 	return ib_ret;
988 }
989 
990 u8 iser_check_task_pi_status(struct iscsi_iser_task *iser_task,
991 			     enum iser_data_dir cmd_dir, sector_t *sector)
992 {
993 	struct iser_mem_reg *reg = &iser_task->rdma_reg[cmd_dir];
994 	struct iser_fr_desc *desc = reg->mem_h;
995 	unsigned long sector_size = iser_task->sc->device->sector_size;
996 	struct ib_mr_status mr_status;
997 	int ret;
998 
999 	if (desc && desc->sig_protected) {
1000 		desc->sig_protected = false;
1001 		ret = ib_check_mr_status(desc->rsc.sig_mr,
1002 					 IB_MR_CHECK_SIG_STATUS, &mr_status);
1003 		if (ret) {
1004 			iser_err("ib_check_mr_status failed, ret %d\n", ret);
1005 			/* Not a lot we can do, return ambiguous guard error */
1006 			*sector = 0;
1007 			return 0x1;
1008 		}
1009 
1010 		if (mr_status.fail_status & IB_MR_CHECK_SIG_STATUS) {
1011 			sector_t sector_off = mr_status.sig_err.sig_err_offset;
1012 
1013 			sector_div(sector_off, sector_size + 8);
1014 			*sector = scsi_get_lba(iser_task->sc) + sector_off;
1015 
1016 			iser_err("PI error found type %d at sector %llx "
1017 			       "expected %x vs actual %x\n",
1018 			       mr_status.sig_err.err_type,
1019 			       (unsigned long long)*sector,
1020 			       mr_status.sig_err.expected,
1021 			       mr_status.sig_err.actual);
1022 
1023 			switch (mr_status.sig_err.err_type) {
1024 			case IB_SIG_BAD_GUARD:
1025 				return 0x1;
1026 			case IB_SIG_BAD_REFTAG:
1027 				return 0x3;
1028 			case IB_SIG_BAD_APPTAG:
1029 				return 0x2;
1030 			}
1031 		}
1032 	}
1033 
1034 	return 0;
1035 }
1036 
1037 void iser_err_comp(struct ib_wc *wc, const char *type)
1038 {
1039 	if (wc->status != IB_WC_WR_FLUSH_ERR) {
1040 		struct iser_conn *iser_conn = to_iser_conn(wc->qp->qp_context);
1041 
1042 		iser_err("%s failure: %s (%d) vend_err %#x\n", type,
1043 			 ib_wc_status_msg(wc->status), wc->status,
1044 			 wc->vendor_err);
1045 
1046 		if (iser_conn->iscsi_conn)
1047 			iscsi_conn_failure(iser_conn->iscsi_conn,
1048 					   ISCSI_ERR_CONN_FAILED);
1049 	} else {
1050 		iser_dbg("%s failure: %s (%d)\n", type,
1051 			 ib_wc_status_msg(wc->status), wc->status);
1052 	}
1053 }
1054