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