1 /*******************************************************************************
2  * This file contains iSCSI extentions for RDMA (iSER) Verbs
3  *
4  * (c) Copyright 2013 Datera, Inc.
5  *
6  * Nicholas A. Bellinger <nab@linux-iscsi.org>
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; either version 2 of the License, or
11  * (at your option) any later version.
12  *
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17  ****************************************************************************/
18 
19 #include <linux/string.h>
20 #include <linux/module.h>
21 #include <linux/scatterlist.h>
22 #include <linux/socket.h>
23 #include <linux/in.h>
24 #include <linux/in6.h>
25 #include <rdma/ib_verbs.h>
26 #include <rdma/rdma_cm.h>
27 #include <target/target_core_base.h>
28 #include <target/target_core_fabric.h>
29 #include <target/iscsi/iscsi_transport.h>
30 #include <linux/semaphore.h>
31 
32 #include "ib_isert.h"
33 
34 #define	ISERT_MAX_CONN		8
35 #define ISER_MAX_RX_CQ_LEN	(ISERT_QP_MAX_RECV_DTOS * ISERT_MAX_CONN)
36 #define ISER_MAX_TX_CQ_LEN \
37 	((ISERT_QP_MAX_REQ_DTOS + ISCSI_DEF_XMIT_CMDS_MAX) * ISERT_MAX_CONN)
38 #define ISER_MAX_CQ_LEN		(ISER_MAX_RX_CQ_LEN + ISER_MAX_TX_CQ_LEN + \
39 				 ISERT_MAX_CONN)
40 
41 static int isert_debug_level;
42 module_param_named(debug_level, isert_debug_level, int, 0644);
43 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0 (default:0)");
44 
45 static DEFINE_MUTEX(device_list_mutex);
46 static LIST_HEAD(device_list);
47 static struct workqueue_struct *isert_comp_wq;
48 static struct workqueue_struct *isert_release_wq;
49 
50 static int
51 isert_put_response(struct iscsi_conn *conn, struct iscsi_cmd *cmd);
52 static int
53 isert_login_post_recv(struct isert_conn *isert_conn);
54 static int
55 isert_rdma_accept(struct isert_conn *isert_conn);
56 struct rdma_cm_id *isert_setup_id(struct isert_np *isert_np);
57 
58 static void isert_release_work(struct work_struct *work);
59 static void isert_recv_done(struct ib_cq *cq, struct ib_wc *wc);
60 static void isert_send_done(struct ib_cq *cq, struct ib_wc *wc);
61 static void isert_login_recv_done(struct ib_cq *cq, struct ib_wc *wc);
62 static void isert_login_send_done(struct ib_cq *cq, struct ib_wc *wc);
63 
64 static inline bool
65 isert_prot_cmd(struct isert_conn *conn, struct se_cmd *cmd)
66 {
67 	return (conn->pi_support &&
68 		cmd->prot_op != TARGET_PROT_NORMAL);
69 }
70 
71 
72 static void
73 isert_qp_event_callback(struct ib_event *e, void *context)
74 {
75 	struct isert_conn *isert_conn = context;
76 
77 	isert_err("%s (%d): conn %p\n",
78 		  ib_event_msg(e->event), e->event, isert_conn);
79 
80 	switch (e->event) {
81 	case IB_EVENT_COMM_EST:
82 		rdma_notify(isert_conn->cm_id, IB_EVENT_COMM_EST);
83 		break;
84 	case IB_EVENT_QP_LAST_WQE_REACHED:
85 		isert_warn("Reached TX IB_EVENT_QP_LAST_WQE_REACHED\n");
86 		break;
87 	default:
88 		break;
89 	}
90 }
91 
92 static struct isert_comp *
93 isert_comp_get(struct isert_conn *isert_conn)
94 {
95 	struct isert_device *device = isert_conn->device;
96 	struct isert_comp *comp;
97 	int i, min = 0;
98 
99 	mutex_lock(&device_list_mutex);
100 	for (i = 0; i < device->comps_used; i++)
101 		if (device->comps[i].active_qps <
102 		    device->comps[min].active_qps)
103 			min = i;
104 	comp = &device->comps[min];
105 	comp->active_qps++;
106 	mutex_unlock(&device_list_mutex);
107 
108 	isert_info("conn %p, using comp %p min_index: %d\n",
109 		   isert_conn, comp, min);
110 
111 	return comp;
112 }
113 
114 static void
115 isert_comp_put(struct isert_comp *comp)
116 {
117 	mutex_lock(&device_list_mutex);
118 	comp->active_qps--;
119 	mutex_unlock(&device_list_mutex);
120 }
121 
122 static struct ib_qp *
123 isert_create_qp(struct isert_conn *isert_conn,
124 		struct isert_comp *comp,
125 		struct rdma_cm_id *cma_id)
126 {
127 	struct isert_device *device = isert_conn->device;
128 	struct ib_qp_init_attr attr;
129 	int ret;
130 
131 	memset(&attr, 0, sizeof(struct ib_qp_init_attr));
132 	attr.event_handler = isert_qp_event_callback;
133 	attr.qp_context = isert_conn;
134 	attr.send_cq = comp->cq;
135 	attr.recv_cq = comp->cq;
136 	attr.cap.max_send_wr = ISERT_QP_MAX_REQ_DTOS + 1;
137 	attr.cap.max_recv_wr = ISERT_QP_MAX_RECV_DTOS + 1;
138 	attr.cap.max_rdma_ctxs = ISCSI_DEF_XMIT_CMDS_MAX;
139 	attr.cap.max_send_sge = device->ib_device->attrs.max_sge;
140 	isert_conn->max_sge = min(device->ib_device->attrs.max_sge,
141 				  device->ib_device->attrs.max_sge_rd);
142 	attr.cap.max_recv_sge = 1;
143 	attr.sq_sig_type = IB_SIGNAL_REQ_WR;
144 	attr.qp_type = IB_QPT_RC;
145 	if (device->pi_capable)
146 		attr.create_flags |= IB_QP_CREATE_SIGNATURE_EN;
147 
148 	ret = rdma_create_qp(cma_id, device->pd, &attr);
149 	if (ret) {
150 		isert_err("rdma_create_qp failed for cma_id %d\n", ret);
151 		return ERR_PTR(ret);
152 	}
153 
154 	return cma_id->qp;
155 }
156 
157 static int
158 isert_conn_setup_qp(struct isert_conn *isert_conn, struct rdma_cm_id *cma_id)
159 {
160 	struct isert_comp *comp;
161 	int ret;
162 
163 	comp = isert_comp_get(isert_conn);
164 	isert_conn->qp = isert_create_qp(isert_conn, comp, cma_id);
165 	if (IS_ERR(isert_conn->qp)) {
166 		ret = PTR_ERR(isert_conn->qp);
167 		goto err;
168 	}
169 
170 	return 0;
171 err:
172 	isert_comp_put(comp);
173 	return ret;
174 }
175 
176 static int
177 isert_alloc_rx_descriptors(struct isert_conn *isert_conn)
178 {
179 	struct isert_device *device = isert_conn->device;
180 	struct ib_device *ib_dev = device->ib_device;
181 	struct iser_rx_desc *rx_desc;
182 	struct ib_sge *rx_sg;
183 	u64 dma_addr;
184 	int i, j;
185 
186 	isert_conn->rx_descs = kzalloc(ISERT_QP_MAX_RECV_DTOS *
187 				sizeof(struct iser_rx_desc), GFP_KERNEL);
188 	if (!isert_conn->rx_descs)
189 		goto fail;
190 
191 	rx_desc = isert_conn->rx_descs;
192 
193 	for (i = 0; i < ISERT_QP_MAX_RECV_DTOS; i++, rx_desc++)  {
194 		dma_addr = ib_dma_map_single(ib_dev, (void *)rx_desc,
195 					ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
196 		if (ib_dma_mapping_error(ib_dev, dma_addr))
197 			goto dma_map_fail;
198 
199 		rx_desc->dma_addr = dma_addr;
200 
201 		rx_sg = &rx_desc->rx_sg;
202 		rx_sg->addr = rx_desc->dma_addr;
203 		rx_sg->length = ISER_RX_PAYLOAD_SIZE;
204 		rx_sg->lkey = device->pd->local_dma_lkey;
205 		rx_desc->rx_cqe.done = isert_recv_done;
206 	}
207 
208 	return 0;
209 
210 dma_map_fail:
211 	rx_desc = isert_conn->rx_descs;
212 	for (j = 0; j < i; j++, rx_desc++) {
213 		ib_dma_unmap_single(ib_dev, rx_desc->dma_addr,
214 				    ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
215 	}
216 	kfree(isert_conn->rx_descs);
217 	isert_conn->rx_descs = NULL;
218 fail:
219 	isert_err("conn %p failed to allocate rx descriptors\n", isert_conn);
220 
221 	return -ENOMEM;
222 }
223 
224 static void
225 isert_free_rx_descriptors(struct isert_conn *isert_conn)
226 {
227 	struct ib_device *ib_dev = isert_conn->device->ib_device;
228 	struct iser_rx_desc *rx_desc;
229 	int i;
230 
231 	if (!isert_conn->rx_descs)
232 		return;
233 
234 	rx_desc = isert_conn->rx_descs;
235 	for (i = 0; i < ISERT_QP_MAX_RECV_DTOS; i++, rx_desc++)  {
236 		ib_dma_unmap_single(ib_dev, rx_desc->dma_addr,
237 				    ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
238 	}
239 
240 	kfree(isert_conn->rx_descs);
241 	isert_conn->rx_descs = NULL;
242 }
243 
244 static void
245 isert_free_comps(struct isert_device *device)
246 {
247 	int i;
248 
249 	for (i = 0; i < device->comps_used; i++) {
250 		struct isert_comp *comp = &device->comps[i];
251 
252 		if (comp->cq)
253 			ib_free_cq(comp->cq);
254 	}
255 	kfree(device->comps);
256 }
257 
258 static int
259 isert_alloc_comps(struct isert_device *device)
260 {
261 	int i, max_cqe, ret = 0;
262 
263 	device->comps_used = min(ISERT_MAX_CQ, min_t(int, num_online_cpus(),
264 				 device->ib_device->num_comp_vectors));
265 
266 	isert_info("Using %d CQs, %s supports %d vectors support "
267 		   "pi_capable %d\n",
268 		   device->comps_used, device->ib_device->name,
269 		   device->ib_device->num_comp_vectors,
270 		   device->pi_capable);
271 
272 	device->comps = kcalloc(device->comps_used, sizeof(struct isert_comp),
273 				GFP_KERNEL);
274 	if (!device->comps) {
275 		isert_err("Unable to allocate completion contexts\n");
276 		return -ENOMEM;
277 	}
278 
279 	max_cqe = min(ISER_MAX_CQ_LEN, device->ib_device->attrs.max_cqe);
280 
281 	for (i = 0; i < device->comps_used; i++) {
282 		struct isert_comp *comp = &device->comps[i];
283 
284 		comp->device = device;
285 		comp->cq = ib_alloc_cq(device->ib_device, comp, max_cqe, i,
286 				IB_POLL_WORKQUEUE);
287 		if (IS_ERR(comp->cq)) {
288 			isert_err("Unable to allocate cq\n");
289 			ret = PTR_ERR(comp->cq);
290 			comp->cq = NULL;
291 			goto out_cq;
292 		}
293 	}
294 
295 	return 0;
296 out_cq:
297 	isert_free_comps(device);
298 	return ret;
299 }
300 
301 static int
302 isert_create_device_ib_res(struct isert_device *device)
303 {
304 	struct ib_device *ib_dev = device->ib_device;
305 	int ret;
306 
307 	isert_dbg("devattr->max_sge: %d\n", ib_dev->attrs.max_sge);
308 	isert_dbg("devattr->max_sge_rd: %d\n", ib_dev->attrs.max_sge_rd);
309 
310 	ret = isert_alloc_comps(device);
311 	if (ret)
312 		goto out;
313 
314 	device->pd = ib_alloc_pd(ib_dev);
315 	if (IS_ERR(device->pd)) {
316 		ret = PTR_ERR(device->pd);
317 		isert_err("failed to allocate pd, device %p, ret=%d\n",
318 			  device, ret);
319 		goto out_cq;
320 	}
321 
322 	/* Check signature cap */
323 	device->pi_capable = ib_dev->attrs.device_cap_flags &
324 			     IB_DEVICE_SIGNATURE_HANDOVER ? true : false;
325 
326 	return 0;
327 
328 out_cq:
329 	isert_free_comps(device);
330 out:
331 	if (ret > 0)
332 		ret = -EINVAL;
333 	return ret;
334 }
335 
336 static void
337 isert_free_device_ib_res(struct isert_device *device)
338 {
339 	isert_info("device %p\n", device);
340 
341 	ib_dealloc_pd(device->pd);
342 	isert_free_comps(device);
343 }
344 
345 static void
346 isert_device_put(struct isert_device *device)
347 {
348 	mutex_lock(&device_list_mutex);
349 	device->refcount--;
350 	isert_info("device %p refcount %d\n", device, device->refcount);
351 	if (!device->refcount) {
352 		isert_free_device_ib_res(device);
353 		list_del(&device->dev_node);
354 		kfree(device);
355 	}
356 	mutex_unlock(&device_list_mutex);
357 }
358 
359 static struct isert_device *
360 isert_device_get(struct rdma_cm_id *cma_id)
361 {
362 	struct isert_device *device;
363 	int ret;
364 
365 	mutex_lock(&device_list_mutex);
366 	list_for_each_entry(device, &device_list, dev_node) {
367 		if (device->ib_device->node_guid == cma_id->device->node_guid) {
368 			device->refcount++;
369 			isert_info("Found iser device %p refcount %d\n",
370 				   device, device->refcount);
371 			mutex_unlock(&device_list_mutex);
372 			return device;
373 		}
374 	}
375 
376 	device = kzalloc(sizeof(struct isert_device), GFP_KERNEL);
377 	if (!device) {
378 		mutex_unlock(&device_list_mutex);
379 		return ERR_PTR(-ENOMEM);
380 	}
381 
382 	INIT_LIST_HEAD(&device->dev_node);
383 
384 	device->ib_device = cma_id->device;
385 	ret = isert_create_device_ib_res(device);
386 	if (ret) {
387 		kfree(device);
388 		mutex_unlock(&device_list_mutex);
389 		return ERR_PTR(ret);
390 	}
391 
392 	device->refcount++;
393 	list_add_tail(&device->dev_node, &device_list);
394 	isert_info("Created a new iser device %p refcount %d\n",
395 		   device, device->refcount);
396 	mutex_unlock(&device_list_mutex);
397 
398 	return device;
399 }
400 
401 static void
402 isert_init_conn(struct isert_conn *isert_conn)
403 {
404 	isert_conn->state = ISER_CONN_INIT;
405 	INIT_LIST_HEAD(&isert_conn->node);
406 	init_completion(&isert_conn->login_comp);
407 	init_completion(&isert_conn->login_req_comp);
408 	kref_init(&isert_conn->kref);
409 	mutex_init(&isert_conn->mutex);
410 	INIT_WORK(&isert_conn->release_work, isert_release_work);
411 }
412 
413 static void
414 isert_free_login_buf(struct isert_conn *isert_conn)
415 {
416 	struct ib_device *ib_dev = isert_conn->device->ib_device;
417 
418 	ib_dma_unmap_single(ib_dev, isert_conn->login_rsp_dma,
419 			    ISER_RX_PAYLOAD_SIZE, DMA_TO_DEVICE);
420 	kfree(isert_conn->login_rsp_buf);
421 
422 	ib_dma_unmap_single(ib_dev, isert_conn->login_req_dma,
423 			    ISER_RX_PAYLOAD_SIZE,
424 			    DMA_FROM_DEVICE);
425 	kfree(isert_conn->login_req_buf);
426 }
427 
428 static int
429 isert_alloc_login_buf(struct isert_conn *isert_conn,
430 		      struct ib_device *ib_dev)
431 {
432 	int ret;
433 
434 	isert_conn->login_req_buf = kzalloc(sizeof(*isert_conn->login_req_buf),
435 			GFP_KERNEL);
436 	if (!isert_conn->login_req_buf) {
437 		isert_err("Unable to allocate isert_conn->login_buf\n");
438 		return -ENOMEM;
439 	}
440 
441 	isert_conn->login_req_dma = ib_dma_map_single(ib_dev,
442 				isert_conn->login_req_buf,
443 				ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
444 	ret = ib_dma_mapping_error(ib_dev, isert_conn->login_req_dma);
445 	if (ret) {
446 		isert_err("login_req_dma mapping error: %d\n", ret);
447 		isert_conn->login_req_dma = 0;
448 		goto out_free_login_req_buf;
449 	}
450 
451 	isert_conn->login_rsp_buf = kzalloc(ISER_RX_PAYLOAD_SIZE, GFP_KERNEL);
452 	if (!isert_conn->login_rsp_buf) {
453 		isert_err("Unable to allocate isert_conn->login_rspbuf\n");
454 		goto out_unmap_login_req_buf;
455 	}
456 
457 	isert_conn->login_rsp_dma = ib_dma_map_single(ib_dev,
458 					isert_conn->login_rsp_buf,
459 					ISER_RX_PAYLOAD_SIZE, DMA_TO_DEVICE);
460 	ret = ib_dma_mapping_error(ib_dev, isert_conn->login_rsp_dma);
461 	if (ret) {
462 		isert_err("login_rsp_dma mapping error: %d\n", ret);
463 		isert_conn->login_rsp_dma = 0;
464 		goto out_free_login_rsp_buf;
465 	}
466 
467 	return 0;
468 
469 out_free_login_rsp_buf:
470 	kfree(isert_conn->login_rsp_buf);
471 out_unmap_login_req_buf:
472 	ib_dma_unmap_single(ib_dev, isert_conn->login_req_dma,
473 			    ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
474 out_free_login_req_buf:
475 	kfree(isert_conn->login_req_buf);
476 	return ret;
477 }
478 
479 static void
480 isert_set_nego_params(struct isert_conn *isert_conn,
481 		      struct rdma_conn_param *param)
482 {
483 	struct ib_device_attr *attr = &isert_conn->device->ib_device->attrs;
484 
485 	/* Set max inflight RDMA READ requests */
486 	isert_conn->initiator_depth = min_t(u8, param->initiator_depth,
487 				attr->max_qp_init_rd_atom);
488 	isert_dbg("Using initiator_depth: %u\n", isert_conn->initiator_depth);
489 
490 	if (param->private_data) {
491 		u8 flags = *(u8 *)param->private_data;
492 
493 		/*
494 		 * use remote invalidation if the both initiator
495 		 * and the HCA support it
496 		 */
497 		isert_conn->snd_w_inv = !(flags & ISER_SEND_W_INV_NOT_SUP) &&
498 					  (attr->device_cap_flags &
499 					   IB_DEVICE_MEM_MGT_EXTENSIONS);
500 		if (isert_conn->snd_w_inv)
501 			isert_info("Using remote invalidation\n");
502 	}
503 }
504 
505 static int
506 isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
507 {
508 	struct isert_np *isert_np = cma_id->context;
509 	struct iscsi_np *np = isert_np->np;
510 	struct isert_conn *isert_conn;
511 	struct isert_device *device;
512 	int ret = 0;
513 
514 	spin_lock_bh(&np->np_thread_lock);
515 	if (!np->enabled) {
516 		spin_unlock_bh(&np->np_thread_lock);
517 		isert_dbg("iscsi_np is not enabled, reject connect request\n");
518 		return rdma_reject(cma_id, NULL, 0);
519 	}
520 	spin_unlock_bh(&np->np_thread_lock);
521 
522 	isert_dbg("cma_id: %p, portal: %p\n",
523 		 cma_id, cma_id->context);
524 
525 	isert_conn = kzalloc(sizeof(struct isert_conn), GFP_KERNEL);
526 	if (!isert_conn)
527 		return -ENOMEM;
528 
529 	isert_init_conn(isert_conn);
530 	isert_conn->cm_id = cma_id;
531 
532 	ret = isert_alloc_login_buf(isert_conn, cma_id->device);
533 	if (ret)
534 		goto out;
535 
536 	device = isert_device_get(cma_id);
537 	if (IS_ERR(device)) {
538 		ret = PTR_ERR(device);
539 		goto out_rsp_dma_map;
540 	}
541 	isert_conn->device = device;
542 
543 	isert_set_nego_params(isert_conn, &event->param.conn);
544 
545 	ret = isert_conn_setup_qp(isert_conn, cma_id);
546 	if (ret)
547 		goto out_conn_dev;
548 
549 	ret = isert_login_post_recv(isert_conn);
550 	if (ret)
551 		goto out_conn_dev;
552 
553 	ret = isert_rdma_accept(isert_conn);
554 	if (ret)
555 		goto out_conn_dev;
556 
557 	mutex_lock(&isert_np->mutex);
558 	list_add_tail(&isert_conn->node, &isert_np->accepted);
559 	mutex_unlock(&isert_np->mutex);
560 
561 	return 0;
562 
563 out_conn_dev:
564 	isert_device_put(device);
565 out_rsp_dma_map:
566 	isert_free_login_buf(isert_conn);
567 out:
568 	kfree(isert_conn);
569 	rdma_reject(cma_id, NULL, 0);
570 	return ret;
571 }
572 
573 static void
574 isert_connect_release(struct isert_conn *isert_conn)
575 {
576 	struct isert_device *device = isert_conn->device;
577 
578 	isert_dbg("conn %p\n", isert_conn);
579 
580 	BUG_ON(!device);
581 
582 	isert_free_rx_descriptors(isert_conn);
583 	if (isert_conn->cm_id)
584 		rdma_destroy_id(isert_conn->cm_id);
585 
586 	if (isert_conn->qp) {
587 		struct isert_comp *comp = isert_conn->qp->recv_cq->cq_context;
588 
589 		isert_comp_put(comp);
590 		ib_destroy_qp(isert_conn->qp);
591 	}
592 
593 	if (isert_conn->login_req_buf)
594 		isert_free_login_buf(isert_conn);
595 
596 	isert_device_put(device);
597 
598 	kfree(isert_conn);
599 }
600 
601 static void
602 isert_connected_handler(struct rdma_cm_id *cma_id)
603 {
604 	struct isert_conn *isert_conn = cma_id->qp->qp_context;
605 	struct isert_np *isert_np = cma_id->context;
606 
607 	isert_info("conn %p\n", isert_conn);
608 
609 	mutex_lock(&isert_conn->mutex);
610 	isert_conn->state = ISER_CONN_UP;
611 	kref_get(&isert_conn->kref);
612 	mutex_unlock(&isert_conn->mutex);
613 
614 	mutex_lock(&isert_np->mutex);
615 	list_move_tail(&isert_conn->node, &isert_np->pending);
616 	mutex_unlock(&isert_np->mutex);
617 
618 	isert_info("np %p: Allow accept_np to continue\n", isert_np);
619 	up(&isert_np->sem);
620 }
621 
622 static void
623 isert_release_kref(struct kref *kref)
624 {
625 	struct isert_conn *isert_conn = container_of(kref,
626 				struct isert_conn, kref);
627 
628 	isert_info("conn %p final kref %s/%d\n", isert_conn, current->comm,
629 		   current->pid);
630 
631 	isert_connect_release(isert_conn);
632 }
633 
634 static void
635 isert_put_conn(struct isert_conn *isert_conn)
636 {
637 	kref_put(&isert_conn->kref, isert_release_kref);
638 }
639 
640 static void
641 isert_handle_unbound_conn(struct isert_conn *isert_conn)
642 {
643 	struct isert_np *isert_np = isert_conn->cm_id->context;
644 
645 	mutex_lock(&isert_np->mutex);
646 	if (!list_empty(&isert_conn->node)) {
647 		/*
648 		 * This means iscsi doesn't know this connection
649 		 * so schedule a cleanup ourselves
650 		 */
651 		list_del_init(&isert_conn->node);
652 		isert_put_conn(isert_conn);
653 		queue_work(isert_release_wq, &isert_conn->release_work);
654 	}
655 	mutex_unlock(&isert_np->mutex);
656 }
657 
658 /**
659  * isert_conn_terminate() - Initiate connection termination
660  * @isert_conn: isert connection struct
661  *
662  * Notes:
663  * In case the connection state is BOUND, move state
664  * to TEMINATING and start teardown sequence (rdma_disconnect).
665  * In case the connection state is UP, complete flush as well.
666  *
667  * This routine must be called with mutex held. Thus it is
668  * safe to call multiple times.
669  */
670 static void
671 isert_conn_terminate(struct isert_conn *isert_conn)
672 {
673 	int err;
674 
675 	if (isert_conn->state >= ISER_CONN_TERMINATING)
676 		return;
677 
678 	isert_info("Terminating conn %p state %d\n",
679 		   isert_conn, isert_conn->state);
680 	isert_conn->state = ISER_CONN_TERMINATING;
681 	err = rdma_disconnect(isert_conn->cm_id);
682 	if (err)
683 		isert_warn("Failed rdma_disconnect isert_conn %p\n",
684 			   isert_conn);
685 }
686 
687 static int
688 isert_np_cma_handler(struct isert_np *isert_np,
689 		     enum rdma_cm_event_type event)
690 {
691 	isert_dbg("%s (%d): isert np %p\n",
692 		  rdma_event_msg(event), event, isert_np);
693 
694 	switch (event) {
695 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
696 		isert_np->cm_id = NULL;
697 		break;
698 	case RDMA_CM_EVENT_ADDR_CHANGE:
699 		isert_np->cm_id = isert_setup_id(isert_np);
700 		if (IS_ERR(isert_np->cm_id)) {
701 			isert_err("isert np %p setup id failed: %ld\n",
702 				  isert_np, PTR_ERR(isert_np->cm_id));
703 			isert_np->cm_id = NULL;
704 		}
705 		break;
706 	default:
707 		isert_err("isert np %p Unexpected event %d\n",
708 			  isert_np, event);
709 	}
710 
711 	return -1;
712 }
713 
714 static int
715 isert_disconnected_handler(struct rdma_cm_id *cma_id,
716 			   enum rdma_cm_event_type event)
717 {
718 	struct isert_conn *isert_conn = cma_id->qp->qp_context;
719 
720 	mutex_lock(&isert_conn->mutex);
721 	switch (isert_conn->state) {
722 	case ISER_CONN_TERMINATING:
723 		break;
724 	case ISER_CONN_UP:
725 		isert_conn_terminate(isert_conn);
726 		ib_drain_qp(isert_conn->qp);
727 		isert_handle_unbound_conn(isert_conn);
728 		break;
729 	case ISER_CONN_BOUND:
730 	case ISER_CONN_FULL_FEATURE: /* FALLTHRU */
731 		iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
732 		break;
733 	default:
734 		isert_warn("conn %p teminating in state %d\n",
735 			   isert_conn, isert_conn->state);
736 	}
737 	mutex_unlock(&isert_conn->mutex);
738 
739 	return 0;
740 }
741 
742 static int
743 isert_connect_error(struct rdma_cm_id *cma_id)
744 {
745 	struct isert_conn *isert_conn = cma_id->qp->qp_context;
746 
747 	list_del_init(&isert_conn->node);
748 	isert_conn->cm_id = NULL;
749 	isert_put_conn(isert_conn);
750 
751 	return -1;
752 }
753 
754 static int
755 isert_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
756 {
757 	struct isert_np *isert_np = cma_id->context;
758 	int ret = 0;
759 
760 	isert_info("%s (%d): status %d id %p np %p\n",
761 		   rdma_event_msg(event->event), event->event,
762 		   event->status, cma_id, cma_id->context);
763 
764 	if (isert_np->cm_id == cma_id)
765 		return isert_np_cma_handler(cma_id->context, event->event);
766 
767 	switch (event->event) {
768 	case RDMA_CM_EVENT_CONNECT_REQUEST:
769 		ret = isert_connect_request(cma_id, event);
770 		if (ret)
771 			isert_err("failed handle connect request %d\n", ret);
772 		break;
773 	case RDMA_CM_EVENT_ESTABLISHED:
774 		isert_connected_handler(cma_id);
775 		break;
776 	case RDMA_CM_EVENT_ADDR_CHANGE:    /* FALLTHRU */
777 	case RDMA_CM_EVENT_DISCONNECTED:   /* FALLTHRU */
778 	case RDMA_CM_EVENT_DEVICE_REMOVAL: /* FALLTHRU */
779 	case RDMA_CM_EVENT_TIMEWAIT_EXIT:  /* FALLTHRU */
780 		ret = isert_disconnected_handler(cma_id, event->event);
781 		break;
782 	case RDMA_CM_EVENT_REJECTED:       /* FALLTHRU */
783 	case RDMA_CM_EVENT_UNREACHABLE:    /* FALLTHRU */
784 	case RDMA_CM_EVENT_CONNECT_ERROR:
785 		ret = isert_connect_error(cma_id);
786 		break;
787 	default:
788 		isert_err("Unhandled RDMA CMA event: %d\n", event->event);
789 		break;
790 	}
791 
792 	return ret;
793 }
794 
795 static int
796 isert_post_recvm(struct isert_conn *isert_conn, u32 count)
797 {
798 	struct ib_recv_wr *rx_wr, *rx_wr_failed;
799 	int i, ret;
800 	struct iser_rx_desc *rx_desc;
801 
802 	for (rx_wr = isert_conn->rx_wr, i = 0; i < count; i++, rx_wr++) {
803 		rx_desc = &isert_conn->rx_descs[i];
804 
805 		rx_wr->wr_cqe = &rx_desc->rx_cqe;
806 		rx_wr->sg_list = &rx_desc->rx_sg;
807 		rx_wr->num_sge = 1;
808 		rx_wr->next = rx_wr + 1;
809 	}
810 	rx_wr--;
811 	rx_wr->next = NULL; /* mark end of work requests list */
812 
813 	ret = ib_post_recv(isert_conn->qp, isert_conn->rx_wr,
814 			   &rx_wr_failed);
815 	if (ret)
816 		isert_err("ib_post_recv() failed with ret: %d\n", ret);
817 
818 	return ret;
819 }
820 
821 static int
822 isert_post_recv(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc)
823 {
824 	struct ib_recv_wr *rx_wr_failed, rx_wr;
825 	int ret;
826 
827 	rx_wr.wr_cqe = &rx_desc->rx_cqe;
828 	rx_wr.sg_list = &rx_desc->rx_sg;
829 	rx_wr.num_sge = 1;
830 	rx_wr.next = NULL;
831 
832 	ret = ib_post_recv(isert_conn->qp, &rx_wr, &rx_wr_failed);
833 	if (ret)
834 		isert_err("ib_post_recv() failed with ret: %d\n", ret);
835 
836 	return ret;
837 }
838 
839 static int
840 isert_login_post_send(struct isert_conn *isert_conn, struct iser_tx_desc *tx_desc)
841 {
842 	struct ib_device *ib_dev = isert_conn->cm_id->device;
843 	struct ib_send_wr send_wr, *send_wr_failed;
844 	int ret;
845 
846 	ib_dma_sync_single_for_device(ib_dev, tx_desc->dma_addr,
847 				      ISER_HEADERS_LEN, DMA_TO_DEVICE);
848 
849 	tx_desc->tx_cqe.done = isert_login_send_done;
850 
851 	send_wr.next	= NULL;
852 	send_wr.wr_cqe	= &tx_desc->tx_cqe;
853 	send_wr.sg_list	= tx_desc->tx_sg;
854 	send_wr.num_sge	= tx_desc->num_sge;
855 	send_wr.opcode	= IB_WR_SEND;
856 	send_wr.send_flags = IB_SEND_SIGNALED;
857 
858 	ret = ib_post_send(isert_conn->qp, &send_wr, &send_wr_failed);
859 	if (ret)
860 		isert_err("ib_post_send() failed, ret: %d\n", ret);
861 
862 	return ret;
863 }
864 
865 static void
866 isert_create_send_desc(struct isert_conn *isert_conn,
867 		       struct isert_cmd *isert_cmd,
868 		       struct iser_tx_desc *tx_desc)
869 {
870 	struct isert_device *device = isert_conn->device;
871 	struct ib_device *ib_dev = device->ib_device;
872 
873 	ib_dma_sync_single_for_cpu(ib_dev, tx_desc->dma_addr,
874 				   ISER_HEADERS_LEN, DMA_TO_DEVICE);
875 
876 	memset(&tx_desc->iser_header, 0, sizeof(struct iser_ctrl));
877 	tx_desc->iser_header.flags = ISCSI_CTRL;
878 
879 	tx_desc->num_sge = 1;
880 
881 	if (tx_desc->tx_sg[0].lkey != device->pd->local_dma_lkey) {
882 		tx_desc->tx_sg[0].lkey = device->pd->local_dma_lkey;
883 		isert_dbg("tx_desc %p lkey mismatch, fixing\n", tx_desc);
884 	}
885 }
886 
887 static int
888 isert_init_tx_hdrs(struct isert_conn *isert_conn,
889 		   struct iser_tx_desc *tx_desc)
890 {
891 	struct isert_device *device = isert_conn->device;
892 	struct ib_device *ib_dev = device->ib_device;
893 	u64 dma_addr;
894 
895 	dma_addr = ib_dma_map_single(ib_dev, (void *)tx_desc,
896 			ISER_HEADERS_LEN, DMA_TO_DEVICE);
897 	if (ib_dma_mapping_error(ib_dev, dma_addr)) {
898 		isert_err("ib_dma_mapping_error() failed\n");
899 		return -ENOMEM;
900 	}
901 
902 	tx_desc->dma_addr = dma_addr;
903 	tx_desc->tx_sg[0].addr	= tx_desc->dma_addr;
904 	tx_desc->tx_sg[0].length = ISER_HEADERS_LEN;
905 	tx_desc->tx_sg[0].lkey = device->pd->local_dma_lkey;
906 
907 	isert_dbg("Setup tx_sg[0].addr: 0x%llx length: %u lkey: 0x%x\n",
908 		  tx_desc->tx_sg[0].addr, tx_desc->tx_sg[0].length,
909 		  tx_desc->tx_sg[0].lkey);
910 
911 	return 0;
912 }
913 
914 static void
915 isert_init_send_wr(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
916 		   struct ib_send_wr *send_wr)
917 {
918 	struct iser_tx_desc *tx_desc = &isert_cmd->tx_desc;
919 
920 	tx_desc->tx_cqe.done = isert_send_done;
921 	send_wr->wr_cqe = &tx_desc->tx_cqe;
922 
923 	if (isert_conn->snd_w_inv && isert_cmd->inv_rkey) {
924 		send_wr->opcode  = IB_WR_SEND_WITH_INV;
925 		send_wr->ex.invalidate_rkey = isert_cmd->inv_rkey;
926 	} else {
927 		send_wr->opcode = IB_WR_SEND;
928 	}
929 
930 	send_wr->sg_list = &tx_desc->tx_sg[0];
931 	send_wr->num_sge = isert_cmd->tx_desc.num_sge;
932 	send_wr->send_flags = IB_SEND_SIGNALED;
933 }
934 
935 static int
936 isert_login_post_recv(struct isert_conn *isert_conn)
937 {
938 	struct ib_recv_wr rx_wr, *rx_wr_fail;
939 	struct ib_sge sge;
940 	int ret;
941 
942 	memset(&sge, 0, sizeof(struct ib_sge));
943 	sge.addr = isert_conn->login_req_dma;
944 	sge.length = ISER_RX_PAYLOAD_SIZE;
945 	sge.lkey = isert_conn->device->pd->local_dma_lkey;
946 
947 	isert_dbg("Setup sge: addr: %llx length: %d 0x%08x\n",
948 		sge.addr, sge.length, sge.lkey);
949 
950 	isert_conn->login_req_buf->rx_cqe.done = isert_login_recv_done;
951 
952 	memset(&rx_wr, 0, sizeof(struct ib_recv_wr));
953 	rx_wr.wr_cqe = &isert_conn->login_req_buf->rx_cqe;
954 	rx_wr.sg_list = &sge;
955 	rx_wr.num_sge = 1;
956 
957 	ret = ib_post_recv(isert_conn->qp, &rx_wr, &rx_wr_fail);
958 	if (ret)
959 		isert_err("ib_post_recv() failed: %d\n", ret);
960 
961 	return ret;
962 }
963 
964 static int
965 isert_put_login_tx(struct iscsi_conn *conn, struct iscsi_login *login,
966 		   u32 length)
967 {
968 	struct isert_conn *isert_conn = conn->context;
969 	struct isert_device *device = isert_conn->device;
970 	struct ib_device *ib_dev = device->ib_device;
971 	struct iser_tx_desc *tx_desc = &isert_conn->login_tx_desc;
972 	int ret;
973 
974 	isert_create_send_desc(isert_conn, NULL, tx_desc);
975 
976 	memcpy(&tx_desc->iscsi_header, &login->rsp[0],
977 	       sizeof(struct iscsi_hdr));
978 
979 	isert_init_tx_hdrs(isert_conn, tx_desc);
980 
981 	if (length > 0) {
982 		struct ib_sge *tx_dsg = &tx_desc->tx_sg[1];
983 
984 		ib_dma_sync_single_for_cpu(ib_dev, isert_conn->login_rsp_dma,
985 					   length, DMA_TO_DEVICE);
986 
987 		memcpy(isert_conn->login_rsp_buf, login->rsp_buf, length);
988 
989 		ib_dma_sync_single_for_device(ib_dev, isert_conn->login_rsp_dma,
990 					      length, DMA_TO_DEVICE);
991 
992 		tx_dsg->addr	= isert_conn->login_rsp_dma;
993 		tx_dsg->length	= length;
994 		tx_dsg->lkey	= isert_conn->device->pd->local_dma_lkey;
995 		tx_desc->num_sge = 2;
996 	}
997 	if (!login->login_failed) {
998 		if (login->login_complete) {
999 			ret = isert_alloc_rx_descriptors(isert_conn);
1000 			if (ret)
1001 				return ret;
1002 
1003 			ret = isert_post_recvm(isert_conn,
1004 					       ISERT_QP_MAX_RECV_DTOS);
1005 			if (ret)
1006 				return ret;
1007 
1008 			/* Now we are in FULL_FEATURE phase */
1009 			mutex_lock(&isert_conn->mutex);
1010 			isert_conn->state = ISER_CONN_FULL_FEATURE;
1011 			mutex_unlock(&isert_conn->mutex);
1012 			goto post_send;
1013 		}
1014 
1015 		ret = isert_login_post_recv(isert_conn);
1016 		if (ret)
1017 			return ret;
1018 	}
1019 post_send:
1020 	ret = isert_login_post_send(isert_conn, tx_desc);
1021 	if (ret)
1022 		return ret;
1023 
1024 	return 0;
1025 }
1026 
1027 static void
1028 isert_rx_login_req(struct isert_conn *isert_conn)
1029 {
1030 	struct iser_rx_desc *rx_desc = isert_conn->login_req_buf;
1031 	int rx_buflen = isert_conn->login_req_len;
1032 	struct iscsi_conn *conn = isert_conn->conn;
1033 	struct iscsi_login *login = conn->conn_login;
1034 	int size;
1035 
1036 	isert_info("conn %p\n", isert_conn);
1037 
1038 	WARN_ON_ONCE(!login);
1039 
1040 	if (login->first_request) {
1041 		struct iscsi_login_req *login_req =
1042 			(struct iscsi_login_req *)&rx_desc->iscsi_header;
1043 		/*
1044 		 * Setup the initial iscsi_login values from the leading
1045 		 * login request PDU.
1046 		 */
1047 		login->leading_connection = (!login_req->tsih) ? 1 : 0;
1048 		login->current_stage =
1049 			(login_req->flags & ISCSI_FLAG_LOGIN_CURRENT_STAGE_MASK)
1050 			 >> 2;
1051 		login->version_min	= login_req->min_version;
1052 		login->version_max	= login_req->max_version;
1053 		memcpy(login->isid, login_req->isid, 6);
1054 		login->cmd_sn		= be32_to_cpu(login_req->cmdsn);
1055 		login->init_task_tag	= login_req->itt;
1056 		login->initial_exp_statsn = be32_to_cpu(login_req->exp_statsn);
1057 		login->cid		= be16_to_cpu(login_req->cid);
1058 		login->tsih		= be16_to_cpu(login_req->tsih);
1059 	}
1060 
1061 	memcpy(&login->req[0], (void *)&rx_desc->iscsi_header, ISCSI_HDR_LEN);
1062 
1063 	size = min(rx_buflen, MAX_KEY_VALUE_PAIRS);
1064 	isert_dbg("Using login payload size: %d, rx_buflen: %d "
1065 		  "MAX_KEY_VALUE_PAIRS: %d\n", size, rx_buflen,
1066 		  MAX_KEY_VALUE_PAIRS);
1067 	memcpy(login->req_buf, &rx_desc->data[0], size);
1068 
1069 	if (login->first_request) {
1070 		complete(&isert_conn->login_comp);
1071 		return;
1072 	}
1073 	schedule_delayed_work(&conn->login_work, 0);
1074 }
1075 
1076 static struct iscsi_cmd
1077 *isert_allocate_cmd(struct iscsi_conn *conn, struct iser_rx_desc *rx_desc)
1078 {
1079 	struct isert_conn *isert_conn = conn->context;
1080 	struct isert_cmd *isert_cmd;
1081 	struct iscsi_cmd *cmd;
1082 
1083 	cmd = iscsit_allocate_cmd(conn, TASK_INTERRUPTIBLE);
1084 	if (!cmd) {
1085 		isert_err("Unable to allocate iscsi_cmd + isert_cmd\n");
1086 		return NULL;
1087 	}
1088 	isert_cmd = iscsit_priv_cmd(cmd);
1089 	isert_cmd->conn = isert_conn;
1090 	isert_cmd->iscsi_cmd = cmd;
1091 	isert_cmd->rx_desc = rx_desc;
1092 
1093 	return cmd;
1094 }
1095 
1096 static int
1097 isert_handle_scsi_cmd(struct isert_conn *isert_conn,
1098 		      struct isert_cmd *isert_cmd, struct iscsi_cmd *cmd,
1099 		      struct iser_rx_desc *rx_desc, unsigned char *buf)
1100 {
1101 	struct iscsi_conn *conn = isert_conn->conn;
1102 	struct iscsi_scsi_req *hdr = (struct iscsi_scsi_req *)buf;
1103 	int imm_data, imm_data_len, unsol_data, sg_nents, rc;
1104 	bool dump_payload = false;
1105 	unsigned int data_len;
1106 
1107 	rc = iscsit_setup_scsi_cmd(conn, cmd, buf);
1108 	if (rc < 0)
1109 		return rc;
1110 
1111 	imm_data = cmd->immediate_data;
1112 	imm_data_len = cmd->first_burst_len;
1113 	unsol_data = cmd->unsolicited_data;
1114 	data_len = cmd->se_cmd.data_length;
1115 
1116 	if (imm_data && imm_data_len == data_len)
1117 		cmd->se_cmd.se_cmd_flags |= SCF_PASSTHROUGH_SG_TO_MEM_NOALLOC;
1118 	rc = iscsit_process_scsi_cmd(conn, cmd, hdr);
1119 	if (rc < 0) {
1120 		return 0;
1121 	} else if (rc > 0) {
1122 		dump_payload = true;
1123 		goto sequence_cmd;
1124 	}
1125 
1126 	if (!imm_data)
1127 		return 0;
1128 
1129 	if (imm_data_len != data_len) {
1130 		sg_nents = max(1UL, DIV_ROUND_UP(imm_data_len, PAGE_SIZE));
1131 		sg_copy_from_buffer(cmd->se_cmd.t_data_sg, sg_nents,
1132 				    &rx_desc->data[0], imm_data_len);
1133 		isert_dbg("Copy Immediate sg_nents: %u imm_data_len: %d\n",
1134 			  sg_nents, imm_data_len);
1135 	} else {
1136 		sg_init_table(&isert_cmd->sg, 1);
1137 		cmd->se_cmd.t_data_sg = &isert_cmd->sg;
1138 		cmd->se_cmd.t_data_nents = 1;
1139 		sg_set_buf(&isert_cmd->sg, &rx_desc->data[0], imm_data_len);
1140 		isert_dbg("Transfer Immediate imm_data_len: %d\n",
1141 			  imm_data_len);
1142 	}
1143 
1144 	cmd->write_data_done += imm_data_len;
1145 
1146 	if (cmd->write_data_done == cmd->se_cmd.data_length) {
1147 		spin_lock_bh(&cmd->istate_lock);
1148 		cmd->cmd_flags |= ICF_GOT_LAST_DATAOUT;
1149 		cmd->i_state = ISTATE_RECEIVED_LAST_DATAOUT;
1150 		spin_unlock_bh(&cmd->istate_lock);
1151 	}
1152 
1153 sequence_cmd:
1154 	rc = iscsit_sequence_cmd(conn, cmd, buf, hdr->cmdsn);
1155 
1156 	if (!rc && dump_payload == false && unsol_data)
1157 		iscsit_set_unsoliticed_dataout(cmd);
1158 	else if (dump_payload && imm_data)
1159 		target_put_sess_cmd(&cmd->se_cmd);
1160 
1161 	return 0;
1162 }
1163 
1164 static int
1165 isert_handle_iscsi_dataout(struct isert_conn *isert_conn,
1166 			   struct iser_rx_desc *rx_desc, unsigned char *buf)
1167 {
1168 	struct scatterlist *sg_start;
1169 	struct iscsi_conn *conn = isert_conn->conn;
1170 	struct iscsi_cmd *cmd = NULL;
1171 	struct iscsi_data *hdr = (struct iscsi_data *)buf;
1172 	u32 unsol_data_len = ntoh24(hdr->dlength);
1173 	int rc, sg_nents, sg_off, page_off;
1174 
1175 	rc = iscsit_check_dataout_hdr(conn, buf, &cmd);
1176 	if (rc < 0)
1177 		return rc;
1178 	else if (!cmd)
1179 		return 0;
1180 	/*
1181 	 * FIXME: Unexpected unsolicited_data out
1182 	 */
1183 	if (!cmd->unsolicited_data) {
1184 		isert_err("Received unexpected solicited data payload\n");
1185 		dump_stack();
1186 		return -1;
1187 	}
1188 
1189 	isert_dbg("Unsolicited DataOut unsol_data_len: %u, "
1190 		  "write_data_done: %u, data_length: %u\n",
1191 		  unsol_data_len,  cmd->write_data_done,
1192 		  cmd->se_cmd.data_length);
1193 
1194 	sg_off = cmd->write_data_done / PAGE_SIZE;
1195 	sg_start = &cmd->se_cmd.t_data_sg[sg_off];
1196 	sg_nents = max(1UL, DIV_ROUND_UP(unsol_data_len, PAGE_SIZE));
1197 	page_off = cmd->write_data_done % PAGE_SIZE;
1198 	/*
1199 	 * FIXME: Non page-aligned unsolicited_data out
1200 	 */
1201 	if (page_off) {
1202 		isert_err("unexpected non-page aligned data payload\n");
1203 		dump_stack();
1204 		return -1;
1205 	}
1206 	isert_dbg("Copying DataOut: sg_start: %p, sg_off: %u "
1207 		  "sg_nents: %u from %p %u\n", sg_start, sg_off,
1208 		  sg_nents, &rx_desc->data[0], unsol_data_len);
1209 
1210 	sg_copy_from_buffer(sg_start, sg_nents, &rx_desc->data[0],
1211 			    unsol_data_len);
1212 
1213 	rc = iscsit_check_dataout_payload(cmd, hdr, false);
1214 	if (rc < 0)
1215 		return rc;
1216 
1217 	/*
1218 	 * multiple data-outs on the same command can arrive -
1219 	 * so post the buffer before hand
1220 	 */
1221 	rc = isert_post_recv(isert_conn, rx_desc);
1222 	if (rc) {
1223 		isert_err("ib_post_recv failed with %d\n", rc);
1224 		return rc;
1225 	}
1226 	return 0;
1227 }
1228 
1229 static int
1230 isert_handle_nop_out(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
1231 		     struct iscsi_cmd *cmd, struct iser_rx_desc *rx_desc,
1232 		     unsigned char *buf)
1233 {
1234 	struct iscsi_conn *conn = isert_conn->conn;
1235 	struct iscsi_nopout *hdr = (struct iscsi_nopout *)buf;
1236 	int rc;
1237 
1238 	rc = iscsit_setup_nop_out(conn, cmd, hdr);
1239 	if (rc < 0)
1240 		return rc;
1241 	/*
1242 	 * FIXME: Add support for NOPOUT payload using unsolicited RDMA payload
1243 	 */
1244 
1245 	return iscsit_process_nop_out(conn, cmd, hdr);
1246 }
1247 
1248 static int
1249 isert_handle_text_cmd(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
1250 		      struct iscsi_cmd *cmd, struct iser_rx_desc *rx_desc,
1251 		      struct iscsi_text *hdr)
1252 {
1253 	struct iscsi_conn *conn = isert_conn->conn;
1254 	u32 payload_length = ntoh24(hdr->dlength);
1255 	int rc;
1256 	unsigned char *text_in = NULL;
1257 
1258 	rc = iscsit_setup_text_cmd(conn, cmd, hdr);
1259 	if (rc < 0)
1260 		return rc;
1261 
1262 	if (payload_length) {
1263 		text_in = kzalloc(payload_length, GFP_KERNEL);
1264 		if (!text_in) {
1265 			isert_err("Unable to allocate text_in of payload_length: %u\n",
1266 				  payload_length);
1267 			return -ENOMEM;
1268 		}
1269 	}
1270 	cmd->text_in_ptr = text_in;
1271 
1272 	memcpy(cmd->text_in_ptr, &rx_desc->data[0], payload_length);
1273 
1274 	return iscsit_process_text_cmd(conn, cmd, hdr);
1275 }
1276 
1277 static int
1278 isert_rx_opcode(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc,
1279 		uint32_t read_stag, uint64_t read_va,
1280 		uint32_t write_stag, uint64_t write_va)
1281 {
1282 	struct iscsi_hdr *hdr = &rx_desc->iscsi_header;
1283 	struct iscsi_conn *conn = isert_conn->conn;
1284 	struct iscsi_cmd *cmd;
1285 	struct isert_cmd *isert_cmd;
1286 	int ret = -EINVAL;
1287 	u8 opcode = (hdr->opcode & ISCSI_OPCODE_MASK);
1288 
1289 	if (conn->sess->sess_ops->SessionType &&
1290 	   (!(opcode & ISCSI_OP_TEXT) || !(opcode & ISCSI_OP_LOGOUT))) {
1291 		isert_err("Got illegal opcode: 0x%02x in SessionType=Discovery,"
1292 			  " ignoring\n", opcode);
1293 		return 0;
1294 	}
1295 
1296 	switch (opcode) {
1297 	case ISCSI_OP_SCSI_CMD:
1298 		cmd = isert_allocate_cmd(conn, rx_desc);
1299 		if (!cmd)
1300 			break;
1301 
1302 		isert_cmd = iscsit_priv_cmd(cmd);
1303 		isert_cmd->read_stag = read_stag;
1304 		isert_cmd->read_va = read_va;
1305 		isert_cmd->write_stag = write_stag;
1306 		isert_cmd->write_va = write_va;
1307 		isert_cmd->inv_rkey = read_stag ? read_stag : write_stag;
1308 
1309 		ret = isert_handle_scsi_cmd(isert_conn, isert_cmd, cmd,
1310 					rx_desc, (unsigned char *)hdr);
1311 		break;
1312 	case ISCSI_OP_NOOP_OUT:
1313 		cmd = isert_allocate_cmd(conn, rx_desc);
1314 		if (!cmd)
1315 			break;
1316 
1317 		isert_cmd = iscsit_priv_cmd(cmd);
1318 		ret = isert_handle_nop_out(isert_conn, isert_cmd, cmd,
1319 					   rx_desc, (unsigned char *)hdr);
1320 		break;
1321 	case ISCSI_OP_SCSI_DATA_OUT:
1322 		ret = isert_handle_iscsi_dataout(isert_conn, rx_desc,
1323 						(unsigned char *)hdr);
1324 		break;
1325 	case ISCSI_OP_SCSI_TMFUNC:
1326 		cmd = isert_allocate_cmd(conn, rx_desc);
1327 		if (!cmd)
1328 			break;
1329 
1330 		ret = iscsit_handle_task_mgt_cmd(conn, cmd,
1331 						(unsigned char *)hdr);
1332 		break;
1333 	case ISCSI_OP_LOGOUT:
1334 		cmd = isert_allocate_cmd(conn, rx_desc);
1335 		if (!cmd)
1336 			break;
1337 
1338 		ret = iscsit_handle_logout_cmd(conn, cmd, (unsigned char *)hdr);
1339 		break;
1340 	case ISCSI_OP_TEXT:
1341 		if (be32_to_cpu(hdr->ttt) != 0xFFFFFFFF)
1342 			cmd = iscsit_find_cmd_from_itt(conn, hdr->itt);
1343 		else
1344 			cmd = isert_allocate_cmd(conn, rx_desc);
1345 
1346 		if (!cmd)
1347 			break;
1348 
1349 		isert_cmd = iscsit_priv_cmd(cmd);
1350 		ret = isert_handle_text_cmd(isert_conn, isert_cmd, cmd,
1351 					    rx_desc, (struct iscsi_text *)hdr);
1352 		break;
1353 	default:
1354 		isert_err("Got unknown iSCSI OpCode: 0x%02x\n", opcode);
1355 		dump_stack();
1356 		break;
1357 	}
1358 
1359 	return ret;
1360 }
1361 
1362 static void
1363 isert_print_wc(struct ib_wc *wc, const char *type)
1364 {
1365 	if (wc->status != IB_WC_WR_FLUSH_ERR)
1366 		isert_err("%s failure: %s (%d) vend_err %x\n", type,
1367 			  ib_wc_status_msg(wc->status), wc->status,
1368 			  wc->vendor_err);
1369 	else
1370 		isert_dbg("%s failure: %s (%d)\n", type,
1371 			  ib_wc_status_msg(wc->status), wc->status);
1372 }
1373 
1374 static void
1375 isert_recv_done(struct ib_cq *cq, struct ib_wc *wc)
1376 {
1377 	struct isert_conn *isert_conn = wc->qp->qp_context;
1378 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1379 	struct iser_rx_desc *rx_desc = cqe_to_rx_desc(wc->wr_cqe);
1380 	struct iscsi_hdr *hdr = &rx_desc->iscsi_header;
1381 	struct iser_ctrl *iser_ctrl = &rx_desc->iser_header;
1382 	uint64_t read_va = 0, write_va = 0;
1383 	uint32_t read_stag = 0, write_stag = 0;
1384 
1385 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1386 		isert_print_wc(wc, "recv");
1387 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1388 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1389 		return;
1390 	}
1391 
1392 	ib_dma_sync_single_for_cpu(ib_dev, rx_desc->dma_addr,
1393 			ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
1394 
1395 	isert_dbg("DMA: 0x%llx, iSCSI opcode: 0x%02x, ITT: 0x%08x, flags: 0x%02x dlen: %d\n",
1396 		 rx_desc->dma_addr, hdr->opcode, hdr->itt, hdr->flags,
1397 		 (int)(wc->byte_len - ISER_HEADERS_LEN));
1398 
1399 	switch (iser_ctrl->flags & 0xF0) {
1400 	case ISCSI_CTRL:
1401 		if (iser_ctrl->flags & ISER_RSV) {
1402 			read_stag = be32_to_cpu(iser_ctrl->read_stag);
1403 			read_va = be64_to_cpu(iser_ctrl->read_va);
1404 			isert_dbg("ISER_RSV: read_stag: 0x%x read_va: 0x%llx\n",
1405 				  read_stag, (unsigned long long)read_va);
1406 		}
1407 		if (iser_ctrl->flags & ISER_WSV) {
1408 			write_stag = be32_to_cpu(iser_ctrl->write_stag);
1409 			write_va = be64_to_cpu(iser_ctrl->write_va);
1410 			isert_dbg("ISER_WSV: write_stag: 0x%x write_va: 0x%llx\n",
1411 				  write_stag, (unsigned long long)write_va);
1412 		}
1413 
1414 		isert_dbg("ISER ISCSI_CTRL PDU\n");
1415 		break;
1416 	case ISER_HELLO:
1417 		isert_err("iSER Hello message\n");
1418 		break;
1419 	default:
1420 		isert_warn("Unknown iSER hdr flags: 0x%02x\n", iser_ctrl->flags);
1421 		break;
1422 	}
1423 
1424 	isert_rx_opcode(isert_conn, rx_desc,
1425 			read_stag, read_va, write_stag, write_va);
1426 
1427 	ib_dma_sync_single_for_device(ib_dev, rx_desc->dma_addr,
1428 			ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
1429 }
1430 
1431 static void
1432 isert_login_recv_done(struct ib_cq *cq, struct ib_wc *wc)
1433 {
1434 	struct isert_conn *isert_conn = wc->qp->qp_context;
1435 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1436 
1437 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1438 		isert_print_wc(wc, "login recv");
1439 		return;
1440 	}
1441 
1442 	ib_dma_sync_single_for_cpu(ib_dev, isert_conn->login_req_dma,
1443 			ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
1444 
1445 	isert_conn->login_req_len = wc->byte_len - ISER_HEADERS_LEN;
1446 
1447 	if (isert_conn->conn) {
1448 		struct iscsi_login *login = isert_conn->conn->conn_login;
1449 
1450 		if (login && !login->first_request)
1451 			isert_rx_login_req(isert_conn);
1452 	}
1453 
1454 	mutex_lock(&isert_conn->mutex);
1455 	complete(&isert_conn->login_req_comp);
1456 	mutex_unlock(&isert_conn->mutex);
1457 
1458 	ib_dma_sync_single_for_device(ib_dev, isert_conn->login_req_dma,
1459 				ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
1460 }
1461 
1462 static void
1463 isert_rdma_rw_ctx_destroy(struct isert_cmd *cmd, struct isert_conn *conn)
1464 {
1465 	struct se_cmd *se_cmd = &cmd->iscsi_cmd->se_cmd;
1466 	enum dma_data_direction dir = target_reverse_dma_direction(se_cmd);
1467 
1468 	if (!cmd->rw.nr_ops)
1469 		return;
1470 
1471 	if (isert_prot_cmd(conn, se_cmd)) {
1472 		rdma_rw_ctx_destroy_signature(&cmd->rw, conn->qp,
1473 				conn->cm_id->port_num, se_cmd->t_data_sg,
1474 				se_cmd->t_data_nents, se_cmd->t_prot_sg,
1475 				se_cmd->t_prot_nents, dir);
1476 	} else {
1477 		rdma_rw_ctx_destroy(&cmd->rw, conn->qp, conn->cm_id->port_num,
1478 				se_cmd->t_data_sg, se_cmd->t_data_nents, dir);
1479 	}
1480 
1481 	cmd->rw.nr_ops = 0;
1482 }
1483 
1484 static void
1485 isert_put_cmd(struct isert_cmd *isert_cmd, bool comp_err)
1486 {
1487 	struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd;
1488 	struct isert_conn *isert_conn = isert_cmd->conn;
1489 	struct iscsi_conn *conn = isert_conn->conn;
1490 	struct iscsi_text_rsp *hdr;
1491 
1492 	isert_dbg("Cmd %p\n", isert_cmd);
1493 
1494 	switch (cmd->iscsi_opcode) {
1495 	case ISCSI_OP_SCSI_CMD:
1496 		spin_lock_bh(&conn->cmd_lock);
1497 		if (!list_empty(&cmd->i_conn_node))
1498 			list_del_init(&cmd->i_conn_node);
1499 		spin_unlock_bh(&conn->cmd_lock);
1500 
1501 		if (cmd->data_direction == DMA_TO_DEVICE) {
1502 			iscsit_stop_dataout_timer(cmd);
1503 			/*
1504 			 * Check for special case during comp_err where
1505 			 * WRITE_PENDING has been handed off from core,
1506 			 * but requires an extra target_put_sess_cmd()
1507 			 * before transport_generic_free_cmd() below.
1508 			 */
1509 			if (comp_err &&
1510 			    cmd->se_cmd.t_state == TRANSPORT_WRITE_PENDING) {
1511 				struct se_cmd *se_cmd = &cmd->se_cmd;
1512 
1513 				target_put_sess_cmd(se_cmd);
1514 			}
1515 		}
1516 
1517 		isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1518 		transport_generic_free_cmd(&cmd->se_cmd, 0);
1519 		break;
1520 	case ISCSI_OP_SCSI_TMFUNC:
1521 		spin_lock_bh(&conn->cmd_lock);
1522 		if (!list_empty(&cmd->i_conn_node))
1523 			list_del_init(&cmd->i_conn_node);
1524 		spin_unlock_bh(&conn->cmd_lock);
1525 
1526 		transport_generic_free_cmd(&cmd->se_cmd, 0);
1527 		break;
1528 	case ISCSI_OP_REJECT:
1529 	case ISCSI_OP_NOOP_OUT:
1530 	case ISCSI_OP_TEXT:
1531 		hdr = (struct iscsi_text_rsp *)&isert_cmd->tx_desc.iscsi_header;
1532 		/* If the continue bit is on, keep the command alive */
1533 		if (hdr->flags & ISCSI_FLAG_TEXT_CONTINUE)
1534 			break;
1535 
1536 		spin_lock_bh(&conn->cmd_lock);
1537 		if (!list_empty(&cmd->i_conn_node))
1538 			list_del_init(&cmd->i_conn_node);
1539 		spin_unlock_bh(&conn->cmd_lock);
1540 
1541 		/*
1542 		 * Handle special case for REJECT when iscsi_add_reject*() has
1543 		 * overwritten the original iscsi_opcode assignment, and the
1544 		 * associated cmd->se_cmd needs to be released.
1545 		 */
1546 		if (cmd->se_cmd.se_tfo != NULL) {
1547 			isert_dbg("Calling transport_generic_free_cmd for 0x%02x\n",
1548 				 cmd->iscsi_opcode);
1549 			transport_generic_free_cmd(&cmd->se_cmd, 0);
1550 			break;
1551 		}
1552 		/*
1553 		 * Fall-through
1554 		 */
1555 	default:
1556 		iscsit_release_cmd(cmd);
1557 		break;
1558 	}
1559 }
1560 
1561 static void
1562 isert_unmap_tx_desc(struct iser_tx_desc *tx_desc, struct ib_device *ib_dev)
1563 {
1564 	if (tx_desc->dma_addr != 0) {
1565 		isert_dbg("unmap single for tx_desc->dma_addr\n");
1566 		ib_dma_unmap_single(ib_dev, tx_desc->dma_addr,
1567 				    ISER_HEADERS_LEN, DMA_TO_DEVICE);
1568 		tx_desc->dma_addr = 0;
1569 	}
1570 }
1571 
1572 static void
1573 isert_completion_put(struct iser_tx_desc *tx_desc, struct isert_cmd *isert_cmd,
1574 		     struct ib_device *ib_dev, bool comp_err)
1575 {
1576 	if (isert_cmd->pdu_buf_dma != 0) {
1577 		isert_dbg("unmap single for isert_cmd->pdu_buf_dma\n");
1578 		ib_dma_unmap_single(ib_dev, isert_cmd->pdu_buf_dma,
1579 				    isert_cmd->pdu_buf_len, DMA_TO_DEVICE);
1580 		isert_cmd->pdu_buf_dma = 0;
1581 	}
1582 
1583 	isert_unmap_tx_desc(tx_desc, ib_dev);
1584 	isert_put_cmd(isert_cmd, comp_err);
1585 }
1586 
1587 static int
1588 isert_check_pi_status(struct se_cmd *se_cmd, struct ib_mr *sig_mr)
1589 {
1590 	struct ib_mr_status mr_status;
1591 	int ret;
1592 
1593 	ret = ib_check_mr_status(sig_mr, IB_MR_CHECK_SIG_STATUS, &mr_status);
1594 	if (ret) {
1595 		isert_err("ib_check_mr_status failed, ret %d\n", ret);
1596 		goto fail_mr_status;
1597 	}
1598 
1599 	if (mr_status.fail_status & IB_MR_CHECK_SIG_STATUS) {
1600 		u64 sec_offset_err;
1601 		u32 block_size = se_cmd->se_dev->dev_attrib.block_size + 8;
1602 
1603 		switch (mr_status.sig_err.err_type) {
1604 		case IB_SIG_BAD_GUARD:
1605 			se_cmd->pi_err = TCM_LOGICAL_BLOCK_GUARD_CHECK_FAILED;
1606 			break;
1607 		case IB_SIG_BAD_REFTAG:
1608 			se_cmd->pi_err = TCM_LOGICAL_BLOCK_REF_TAG_CHECK_FAILED;
1609 			break;
1610 		case IB_SIG_BAD_APPTAG:
1611 			se_cmd->pi_err = TCM_LOGICAL_BLOCK_APP_TAG_CHECK_FAILED;
1612 			break;
1613 		}
1614 		sec_offset_err = mr_status.sig_err.sig_err_offset;
1615 		do_div(sec_offset_err, block_size);
1616 		se_cmd->bad_sector = sec_offset_err + se_cmd->t_task_lba;
1617 
1618 		isert_err("PI error found type %d at sector 0x%llx "
1619 			  "expected 0x%x vs actual 0x%x\n",
1620 			  mr_status.sig_err.err_type,
1621 			  (unsigned long long)se_cmd->bad_sector,
1622 			  mr_status.sig_err.expected,
1623 			  mr_status.sig_err.actual);
1624 		ret = 1;
1625 	}
1626 
1627 fail_mr_status:
1628 	return ret;
1629 }
1630 
1631 static void
1632 isert_rdma_write_done(struct ib_cq *cq, struct ib_wc *wc)
1633 {
1634 	struct isert_conn *isert_conn = wc->qp->qp_context;
1635 	struct isert_device *device = isert_conn->device;
1636 	struct iser_tx_desc *desc = cqe_to_tx_desc(wc->wr_cqe);
1637 	struct isert_cmd *isert_cmd = tx_desc_to_cmd(desc);
1638 	struct se_cmd *cmd = &isert_cmd->iscsi_cmd->se_cmd;
1639 	int ret = 0;
1640 
1641 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1642 		isert_print_wc(wc, "rdma write");
1643 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1644 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1645 		isert_completion_put(desc, isert_cmd, device->ib_device, true);
1646 		return;
1647 	}
1648 
1649 	isert_dbg("Cmd %p\n", isert_cmd);
1650 
1651 	ret = isert_check_pi_status(cmd, isert_cmd->rw.sig->sig_mr);
1652 	isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1653 
1654 	if (ret)
1655 		transport_send_check_condition_and_sense(cmd, cmd->pi_err, 0);
1656 	else
1657 		isert_put_response(isert_conn->conn, isert_cmd->iscsi_cmd);
1658 }
1659 
1660 static void
1661 isert_rdma_read_done(struct ib_cq *cq, struct ib_wc *wc)
1662 {
1663 	struct isert_conn *isert_conn = wc->qp->qp_context;
1664 	struct isert_device *device = isert_conn->device;
1665 	struct iser_tx_desc *desc = cqe_to_tx_desc(wc->wr_cqe);
1666 	struct isert_cmd *isert_cmd = tx_desc_to_cmd(desc);
1667 	struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd;
1668 	struct se_cmd *se_cmd = &cmd->se_cmd;
1669 	int ret = 0;
1670 
1671 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1672 		isert_print_wc(wc, "rdma read");
1673 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1674 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1675 		isert_completion_put(desc, isert_cmd, device->ib_device, true);
1676 		return;
1677 	}
1678 
1679 	isert_dbg("Cmd %p\n", isert_cmd);
1680 
1681 	iscsit_stop_dataout_timer(cmd);
1682 
1683 	if (isert_prot_cmd(isert_conn, se_cmd))
1684 		ret = isert_check_pi_status(se_cmd, isert_cmd->rw.sig->sig_mr);
1685 	isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1686 	cmd->write_data_done = 0;
1687 
1688 	isert_dbg("Cmd: %p RDMA_READ comp calling execute_cmd\n", isert_cmd);
1689 	spin_lock_bh(&cmd->istate_lock);
1690 	cmd->cmd_flags |= ICF_GOT_LAST_DATAOUT;
1691 	cmd->i_state = ISTATE_RECEIVED_LAST_DATAOUT;
1692 	spin_unlock_bh(&cmd->istate_lock);
1693 
1694 	if (ret) {
1695 		target_put_sess_cmd(se_cmd);
1696 		transport_send_check_condition_and_sense(se_cmd,
1697 							 se_cmd->pi_err, 0);
1698 	} else {
1699 		target_execute_cmd(se_cmd);
1700 	}
1701 }
1702 
1703 static void
1704 isert_do_control_comp(struct work_struct *work)
1705 {
1706 	struct isert_cmd *isert_cmd = container_of(work,
1707 			struct isert_cmd, comp_work);
1708 	struct isert_conn *isert_conn = isert_cmd->conn;
1709 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1710 	struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd;
1711 
1712 	isert_dbg("Cmd %p i_state %d\n", isert_cmd, cmd->i_state);
1713 
1714 	switch (cmd->i_state) {
1715 	case ISTATE_SEND_TASKMGTRSP:
1716 		iscsit_tmr_post_handler(cmd, cmd->conn);
1717 	case ISTATE_SEND_REJECT:   /* FALLTHRU */
1718 	case ISTATE_SEND_TEXTRSP:  /* FALLTHRU */
1719 		cmd->i_state = ISTATE_SENT_STATUS;
1720 		isert_completion_put(&isert_cmd->tx_desc, isert_cmd,
1721 				     ib_dev, false);
1722 		break;
1723 	case ISTATE_SEND_LOGOUTRSP:
1724 		iscsit_logout_post_handler(cmd, cmd->conn);
1725 		break;
1726 	default:
1727 		isert_err("Unknown i_state %d\n", cmd->i_state);
1728 		dump_stack();
1729 		break;
1730 	}
1731 }
1732 
1733 static void
1734 isert_login_send_done(struct ib_cq *cq, struct ib_wc *wc)
1735 {
1736 	struct isert_conn *isert_conn = wc->qp->qp_context;
1737 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1738 	struct iser_tx_desc *tx_desc = cqe_to_tx_desc(wc->wr_cqe);
1739 
1740 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1741 		isert_print_wc(wc, "login send");
1742 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1743 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1744 	}
1745 
1746 	isert_unmap_tx_desc(tx_desc, ib_dev);
1747 }
1748 
1749 static void
1750 isert_send_done(struct ib_cq *cq, struct ib_wc *wc)
1751 {
1752 	struct isert_conn *isert_conn = wc->qp->qp_context;
1753 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1754 	struct iser_tx_desc *tx_desc = cqe_to_tx_desc(wc->wr_cqe);
1755 	struct isert_cmd *isert_cmd = tx_desc_to_cmd(tx_desc);
1756 
1757 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1758 		isert_print_wc(wc, "send");
1759 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1760 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1761 		isert_completion_put(tx_desc, isert_cmd, ib_dev, true);
1762 		return;
1763 	}
1764 
1765 	isert_dbg("Cmd %p\n", isert_cmd);
1766 
1767 	switch (isert_cmd->iscsi_cmd->i_state) {
1768 	case ISTATE_SEND_TASKMGTRSP:
1769 	case ISTATE_SEND_LOGOUTRSP:
1770 	case ISTATE_SEND_REJECT:
1771 	case ISTATE_SEND_TEXTRSP:
1772 		isert_unmap_tx_desc(tx_desc, ib_dev);
1773 
1774 		INIT_WORK(&isert_cmd->comp_work, isert_do_control_comp);
1775 		queue_work(isert_comp_wq, &isert_cmd->comp_work);
1776 		return;
1777 	default:
1778 		isert_cmd->iscsi_cmd->i_state = ISTATE_SENT_STATUS;
1779 		isert_completion_put(tx_desc, isert_cmd, ib_dev, false);
1780 		break;
1781 	}
1782 }
1783 
1784 static int
1785 isert_post_response(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd)
1786 {
1787 	struct ib_send_wr *wr_failed;
1788 	int ret;
1789 
1790 	ret = isert_post_recv(isert_conn, isert_cmd->rx_desc);
1791 	if (ret) {
1792 		isert_err("ib_post_recv failed with %d\n", ret);
1793 		return ret;
1794 	}
1795 
1796 	ret = ib_post_send(isert_conn->qp, &isert_cmd->tx_desc.send_wr,
1797 			   &wr_failed);
1798 	if (ret) {
1799 		isert_err("ib_post_send failed with %d\n", ret);
1800 		return ret;
1801 	}
1802 	return ret;
1803 }
1804 
1805 static int
1806 isert_put_response(struct iscsi_conn *conn, struct iscsi_cmd *cmd)
1807 {
1808 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1809 	struct isert_conn *isert_conn = conn->context;
1810 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1811 	struct iscsi_scsi_rsp *hdr = (struct iscsi_scsi_rsp *)
1812 				&isert_cmd->tx_desc.iscsi_header;
1813 
1814 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1815 	iscsit_build_rsp_pdu(cmd, conn, true, hdr);
1816 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1817 	/*
1818 	 * Attach SENSE DATA payload to iSCSI Response PDU
1819 	 */
1820 	if (cmd->se_cmd.sense_buffer &&
1821 	    ((cmd->se_cmd.se_cmd_flags & SCF_TRANSPORT_TASK_SENSE) ||
1822 	    (cmd->se_cmd.se_cmd_flags & SCF_EMULATED_TASK_SENSE))) {
1823 		struct isert_device *device = isert_conn->device;
1824 		struct ib_device *ib_dev = device->ib_device;
1825 		struct ib_sge *tx_dsg = &isert_cmd->tx_desc.tx_sg[1];
1826 		u32 padding, pdu_len;
1827 
1828 		put_unaligned_be16(cmd->se_cmd.scsi_sense_length,
1829 				   cmd->sense_buffer);
1830 		cmd->se_cmd.scsi_sense_length += sizeof(__be16);
1831 
1832 		padding = -(cmd->se_cmd.scsi_sense_length) & 3;
1833 		hton24(hdr->dlength, (u32)cmd->se_cmd.scsi_sense_length);
1834 		pdu_len = cmd->se_cmd.scsi_sense_length + padding;
1835 
1836 		isert_cmd->pdu_buf_dma = ib_dma_map_single(ib_dev,
1837 				(void *)cmd->sense_buffer, pdu_len,
1838 				DMA_TO_DEVICE);
1839 
1840 		isert_cmd->pdu_buf_len = pdu_len;
1841 		tx_dsg->addr	= isert_cmd->pdu_buf_dma;
1842 		tx_dsg->length	= pdu_len;
1843 		tx_dsg->lkey	= device->pd->local_dma_lkey;
1844 		isert_cmd->tx_desc.num_sge = 2;
1845 	}
1846 
1847 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1848 
1849 	isert_dbg("Posting SCSI Response\n");
1850 
1851 	return isert_post_response(isert_conn, isert_cmd);
1852 }
1853 
1854 static void
1855 isert_aborted_task(struct iscsi_conn *conn, struct iscsi_cmd *cmd)
1856 {
1857 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1858 	struct isert_conn *isert_conn = conn->context;
1859 
1860 	spin_lock_bh(&conn->cmd_lock);
1861 	if (!list_empty(&cmd->i_conn_node))
1862 		list_del_init(&cmd->i_conn_node);
1863 	spin_unlock_bh(&conn->cmd_lock);
1864 
1865 	if (cmd->data_direction == DMA_TO_DEVICE)
1866 		iscsit_stop_dataout_timer(cmd);
1867 	isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1868 }
1869 
1870 static enum target_prot_op
1871 isert_get_sup_prot_ops(struct iscsi_conn *conn)
1872 {
1873 	struct isert_conn *isert_conn = conn->context;
1874 	struct isert_device *device = isert_conn->device;
1875 
1876 	if (conn->tpg->tpg_attrib.t10_pi) {
1877 		if (device->pi_capable) {
1878 			isert_info("conn %p PI offload enabled\n", isert_conn);
1879 			isert_conn->pi_support = true;
1880 			return TARGET_PROT_ALL;
1881 		}
1882 	}
1883 
1884 	isert_info("conn %p PI offload disabled\n", isert_conn);
1885 	isert_conn->pi_support = false;
1886 
1887 	return TARGET_PROT_NORMAL;
1888 }
1889 
1890 static int
1891 isert_put_nopin(struct iscsi_cmd *cmd, struct iscsi_conn *conn,
1892 		bool nopout_response)
1893 {
1894 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1895 	struct isert_conn *isert_conn = conn->context;
1896 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1897 
1898 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1899 	iscsit_build_nopin_rsp(cmd, conn, (struct iscsi_nopin *)
1900 			       &isert_cmd->tx_desc.iscsi_header,
1901 			       nopout_response);
1902 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1903 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1904 
1905 	isert_dbg("conn %p Posting NOPIN Response\n", isert_conn);
1906 
1907 	return isert_post_response(isert_conn, isert_cmd);
1908 }
1909 
1910 static int
1911 isert_put_logout_rsp(struct iscsi_cmd *cmd, struct iscsi_conn *conn)
1912 {
1913 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1914 	struct isert_conn *isert_conn = conn->context;
1915 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1916 
1917 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1918 	iscsit_build_logout_rsp(cmd, conn, (struct iscsi_logout_rsp *)
1919 				&isert_cmd->tx_desc.iscsi_header);
1920 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1921 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1922 
1923 	isert_dbg("conn %p Posting Logout Response\n", isert_conn);
1924 
1925 	return isert_post_response(isert_conn, isert_cmd);
1926 }
1927 
1928 static int
1929 isert_put_tm_rsp(struct iscsi_cmd *cmd, struct iscsi_conn *conn)
1930 {
1931 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1932 	struct isert_conn *isert_conn = conn->context;
1933 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1934 
1935 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1936 	iscsit_build_task_mgt_rsp(cmd, conn, (struct iscsi_tm_rsp *)
1937 				  &isert_cmd->tx_desc.iscsi_header);
1938 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1939 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1940 
1941 	isert_dbg("conn %p Posting Task Management Response\n", isert_conn);
1942 
1943 	return isert_post_response(isert_conn, isert_cmd);
1944 }
1945 
1946 static int
1947 isert_put_reject(struct iscsi_cmd *cmd, struct iscsi_conn *conn)
1948 {
1949 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1950 	struct isert_conn *isert_conn = conn->context;
1951 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1952 	struct isert_device *device = isert_conn->device;
1953 	struct ib_device *ib_dev = device->ib_device;
1954 	struct ib_sge *tx_dsg = &isert_cmd->tx_desc.tx_sg[1];
1955 	struct iscsi_reject *hdr =
1956 		(struct iscsi_reject *)&isert_cmd->tx_desc.iscsi_header;
1957 
1958 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1959 	iscsit_build_reject(cmd, conn, hdr);
1960 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1961 
1962 	hton24(hdr->dlength, ISCSI_HDR_LEN);
1963 	isert_cmd->pdu_buf_dma = ib_dma_map_single(ib_dev,
1964 			(void *)cmd->buf_ptr, ISCSI_HDR_LEN,
1965 			DMA_TO_DEVICE);
1966 	isert_cmd->pdu_buf_len = ISCSI_HDR_LEN;
1967 	tx_dsg->addr	= isert_cmd->pdu_buf_dma;
1968 	tx_dsg->length	= ISCSI_HDR_LEN;
1969 	tx_dsg->lkey	= device->pd->local_dma_lkey;
1970 	isert_cmd->tx_desc.num_sge = 2;
1971 
1972 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1973 
1974 	isert_dbg("conn %p Posting Reject\n", isert_conn);
1975 
1976 	return isert_post_response(isert_conn, isert_cmd);
1977 }
1978 
1979 static int
1980 isert_put_text_rsp(struct iscsi_cmd *cmd, struct iscsi_conn *conn)
1981 {
1982 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1983 	struct isert_conn *isert_conn = conn->context;
1984 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1985 	struct iscsi_text_rsp *hdr =
1986 		(struct iscsi_text_rsp *)&isert_cmd->tx_desc.iscsi_header;
1987 	u32 txt_rsp_len;
1988 	int rc;
1989 
1990 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1991 	rc = iscsit_build_text_rsp(cmd, conn, hdr, ISCSI_INFINIBAND);
1992 	if (rc < 0)
1993 		return rc;
1994 
1995 	txt_rsp_len = rc;
1996 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1997 
1998 	if (txt_rsp_len) {
1999 		struct isert_device *device = isert_conn->device;
2000 		struct ib_device *ib_dev = device->ib_device;
2001 		struct ib_sge *tx_dsg = &isert_cmd->tx_desc.tx_sg[1];
2002 		void *txt_rsp_buf = cmd->buf_ptr;
2003 
2004 		isert_cmd->pdu_buf_dma = ib_dma_map_single(ib_dev,
2005 				txt_rsp_buf, txt_rsp_len, DMA_TO_DEVICE);
2006 
2007 		isert_cmd->pdu_buf_len = txt_rsp_len;
2008 		tx_dsg->addr	= isert_cmd->pdu_buf_dma;
2009 		tx_dsg->length	= txt_rsp_len;
2010 		tx_dsg->lkey	= device->pd->local_dma_lkey;
2011 		isert_cmd->tx_desc.num_sge = 2;
2012 	}
2013 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
2014 
2015 	isert_dbg("conn %p Text Response\n", isert_conn);
2016 
2017 	return isert_post_response(isert_conn, isert_cmd);
2018 }
2019 
2020 static inline void
2021 isert_set_dif_domain(struct se_cmd *se_cmd, struct ib_sig_attrs *sig_attrs,
2022 		     struct ib_sig_domain *domain)
2023 {
2024 	domain->sig_type = IB_SIG_TYPE_T10_DIF;
2025 	domain->sig.dif.bg_type = IB_T10DIF_CRC;
2026 	domain->sig.dif.pi_interval = se_cmd->se_dev->dev_attrib.block_size;
2027 	domain->sig.dif.ref_tag = se_cmd->reftag_seed;
2028 	/*
2029 	 * At the moment we hard code those, but if in the future
2030 	 * the target core would like to use it, we will take it
2031 	 * from se_cmd.
2032 	 */
2033 	domain->sig.dif.apptag_check_mask = 0xffff;
2034 	domain->sig.dif.app_escape = true;
2035 	domain->sig.dif.ref_escape = true;
2036 	if (se_cmd->prot_type == TARGET_DIF_TYPE1_PROT ||
2037 	    se_cmd->prot_type == TARGET_DIF_TYPE2_PROT)
2038 		domain->sig.dif.ref_remap = true;
2039 };
2040 
2041 static int
2042 isert_set_sig_attrs(struct se_cmd *se_cmd, struct ib_sig_attrs *sig_attrs)
2043 {
2044 	memset(sig_attrs, 0, sizeof(*sig_attrs));
2045 
2046 	switch (se_cmd->prot_op) {
2047 	case TARGET_PROT_DIN_INSERT:
2048 	case TARGET_PROT_DOUT_STRIP:
2049 		sig_attrs->mem.sig_type = IB_SIG_TYPE_NONE;
2050 		isert_set_dif_domain(se_cmd, sig_attrs, &sig_attrs->wire);
2051 		break;
2052 	case TARGET_PROT_DOUT_INSERT:
2053 	case TARGET_PROT_DIN_STRIP:
2054 		sig_attrs->wire.sig_type = IB_SIG_TYPE_NONE;
2055 		isert_set_dif_domain(se_cmd, sig_attrs, &sig_attrs->mem);
2056 		break;
2057 	case TARGET_PROT_DIN_PASS:
2058 	case TARGET_PROT_DOUT_PASS:
2059 		isert_set_dif_domain(se_cmd, sig_attrs, &sig_attrs->wire);
2060 		isert_set_dif_domain(se_cmd, sig_attrs, &sig_attrs->mem);
2061 		break;
2062 	default:
2063 		isert_err("Unsupported PI operation %d\n", se_cmd->prot_op);
2064 		return -EINVAL;
2065 	}
2066 
2067 	sig_attrs->check_mask =
2068 	       (se_cmd->prot_checks & TARGET_DIF_CHECK_GUARD  ? 0xc0 : 0) |
2069 	       (se_cmd->prot_checks & TARGET_DIF_CHECK_REFTAG ? 0x30 : 0) |
2070 	       (se_cmd->prot_checks & TARGET_DIF_CHECK_REFTAG ? 0x0f : 0);
2071 	return 0;
2072 }
2073 
2074 static int
2075 isert_rdma_rw_ctx_post(struct isert_cmd *cmd, struct isert_conn *conn,
2076 		struct ib_cqe *cqe, struct ib_send_wr *chain_wr)
2077 {
2078 	struct se_cmd *se_cmd = &cmd->iscsi_cmd->se_cmd;
2079 	enum dma_data_direction dir = target_reverse_dma_direction(se_cmd);
2080 	u8 port_num = conn->cm_id->port_num;
2081 	u64 addr;
2082 	u32 rkey, offset;
2083 	int ret;
2084 
2085 	if (dir == DMA_FROM_DEVICE) {
2086 		addr = cmd->write_va;
2087 		rkey = cmd->write_stag;
2088 		offset = cmd->iscsi_cmd->write_data_done;
2089 	} else {
2090 		addr = cmd->read_va;
2091 		rkey = cmd->read_stag;
2092 		offset = 0;
2093 	}
2094 
2095 	if (isert_prot_cmd(conn, se_cmd)) {
2096 		struct ib_sig_attrs sig_attrs;
2097 
2098 		ret = isert_set_sig_attrs(se_cmd, &sig_attrs);
2099 		if (ret)
2100 			return ret;
2101 
2102 		WARN_ON_ONCE(offset);
2103 		ret = rdma_rw_ctx_signature_init(&cmd->rw, conn->qp, port_num,
2104 				se_cmd->t_data_sg, se_cmd->t_data_nents,
2105 				se_cmd->t_prot_sg, se_cmd->t_prot_nents,
2106 				&sig_attrs, addr, rkey, dir);
2107 	} else {
2108 		ret = rdma_rw_ctx_init(&cmd->rw, conn->qp, port_num,
2109 				se_cmd->t_data_sg, se_cmd->t_data_nents,
2110 				offset, addr, rkey, dir);
2111 	}
2112 	if (ret < 0) {
2113 		isert_err("Cmd: %p failed to prepare RDMA res\n", cmd);
2114 		return ret;
2115 	}
2116 
2117 	ret = rdma_rw_ctx_post(&cmd->rw, conn->qp, port_num, cqe, chain_wr);
2118 	if (ret < 0)
2119 		isert_err("Cmd: %p failed to post RDMA res\n", cmd);
2120 	return ret;
2121 }
2122 
2123 static int
2124 isert_put_datain(struct iscsi_conn *conn, struct iscsi_cmd *cmd)
2125 {
2126 	struct se_cmd *se_cmd = &cmd->se_cmd;
2127 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2128 	struct isert_conn *isert_conn = conn->context;
2129 	struct ib_cqe *cqe = NULL;
2130 	struct ib_send_wr *chain_wr = NULL;
2131 	int rc;
2132 
2133 	isert_dbg("Cmd: %p RDMA_WRITE data_length: %u\n",
2134 		 isert_cmd, se_cmd->data_length);
2135 
2136 	if (isert_prot_cmd(isert_conn, se_cmd)) {
2137 		isert_cmd->tx_desc.tx_cqe.done = isert_rdma_write_done;
2138 		cqe = &isert_cmd->tx_desc.tx_cqe;
2139 	} else {
2140 		/*
2141 		 * Build isert_conn->tx_desc for iSCSI response PDU and attach
2142 		 */
2143 		isert_create_send_desc(isert_conn, isert_cmd,
2144 				       &isert_cmd->tx_desc);
2145 		iscsit_build_rsp_pdu(cmd, conn, true, (struct iscsi_scsi_rsp *)
2146 				     &isert_cmd->tx_desc.iscsi_header);
2147 		isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
2148 		isert_init_send_wr(isert_conn, isert_cmd,
2149 				   &isert_cmd->tx_desc.send_wr);
2150 
2151 		rc = isert_post_recv(isert_conn, isert_cmd->rx_desc);
2152 		if (rc) {
2153 			isert_err("ib_post_recv failed with %d\n", rc);
2154 			return rc;
2155 		}
2156 
2157 		chain_wr = &isert_cmd->tx_desc.send_wr;
2158 	}
2159 
2160 	isert_rdma_rw_ctx_post(isert_cmd, isert_conn, cqe, chain_wr);
2161 	isert_dbg("Cmd: %p posted RDMA_WRITE for iSER Data READ\n", isert_cmd);
2162 	return 1;
2163 }
2164 
2165 static int
2166 isert_get_dataout(struct iscsi_conn *conn, struct iscsi_cmd *cmd, bool recovery)
2167 {
2168 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2169 
2170 	isert_dbg("Cmd: %p RDMA_READ data_length: %u write_data_done: %u\n",
2171 		 isert_cmd, cmd->se_cmd.data_length, cmd->write_data_done);
2172 
2173 	isert_cmd->tx_desc.tx_cqe.done = isert_rdma_read_done;
2174 	isert_rdma_rw_ctx_post(isert_cmd, conn->context,
2175 			&isert_cmd->tx_desc.tx_cqe, NULL);
2176 
2177 	isert_dbg("Cmd: %p posted RDMA_READ memory for ISER Data WRITE\n",
2178 		 isert_cmd);
2179 	return 0;
2180 }
2181 
2182 static int
2183 isert_immediate_queue(struct iscsi_conn *conn, struct iscsi_cmd *cmd, int state)
2184 {
2185 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2186 	int ret = 0;
2187 
2188 	switch (state) {
2189 	case ISTATE_REMOVE:
2190 		spin_lock_bh(&conn->cmd_lock);
2191 		list_del_init(&cmd->i_conn_node);
2192 		spin_unlock_bh(&conn->cmd_lock);
2193 		isert_put_cmd(isert_cmd, true);
2194 		break;
2195 	case ISTATE_SEND_NOPIN_WANT_RESPONSE:
2196 		ret = isert_put_nopin(cmd, conn, false);
2197 		break;
2198 	default:
2199 		isert_err("Unknown immediate state: 0x%02x\n", state);
2200 		ret = -EINVAL;
2201 		break;
2202 	}
2203 
2204 	return ret;
2205 }
2206 
2207 static int
2208 isert_response_queue(struct iscsi_conn *conn, struct iscsi_cmd *cmd, int state)
2209 {
2210 	struct isert_conn *isert_conn = conn->context;
2211 	int ret;
2212 
2213 	switch (state) {
2214 	case ISTATE_SEND_LOGOUTRSP:
2215 		ret = isert_put_logout_rsp(cmd, conn);
2216 		if (!ret)
2217 			isert_conn->logout_posted = true;
2218 		break;
2219 	case ISTATE_SEND_NOPIN:
2220 		ret = isert_put_nopin(cmd, conn, true);
2221 		break;
2222 	case ISTATE_SEND_TASKMGTRSP:
2223 		ret = isert_put_tm_rsp(cmd, conn);
2224 		break;
2225 	case ISTATE_SEND_REJECT:
2226 		ret = isert_put_reject(cmd, conn);
2227 		break;
2228 	case ISTATE_SEND_TEXTRSP:
2229 		ret = isert_put_text_rsp(cmd, conn);
2230 		break;
2231 	case ISTATE_SEND_STATUS:
2232 		/*
2233 		 * Special case for sending non GOOD SCSI status from TX thread
2234 		 * context during pre se_cmd excecution failure.
2235 		 */
2236 		ret = isert_put_response(conn, cmd);
2237 		break;
2238 	default:
2239 		isert_err("Unknown response state: 0x%02x\n", state);
2240 		ret = -EINVAL;
2241 		break;
2242 	}
2243 
2244 	return ret;
2245 }
2246 
2247 struct rdma_cm_id *
2248 isert_setup_id(struct isert_np *isert_np)
2249 {
2250 	struct iscsi_np *np = isert_np->np;
2251 	struct rdma_cm_id *id;
2252 	struct sockaddr *sa;
2253 	int ret;
2254 
2255 	sa = (struct sockaddr *)&np->np_sockaddr;
2256 	isert_dbg("ksockaddr: %p, sa: %p\n", &np->np_sockaddr, sa);
2257 
2258 	id = rdma_create_id(&init_net, isert_cma_handler, isert_np,
2259 			    RDMA_PS_TCP, IB_QPT_RC);
2260 	if (IS_ERR(id)) {
2261 		isert_err("rdma_create_id() failed: %ld\n", PTR_ERR(id));
2262 		ret = PTR_ERR(id);
2263 		goto out;
2264 	}
2265 	isert_dbg("id %p context %p\n", id, id->context);
2266 
2267 	ret = rdma_bind_addr(id, sa);
2268 	if (ret) {
2269 		isert_err("rdma_bind_addr() failed: %d\n", ret);
2270 		goto out_id;
2271 	}
2272 
2273 	ret = rdma_listen(id, 0);
2274 	if (ret) {
2275 		isert_err("rdma_listen() failed: %d\n", ret);
2276 		goto out_id;
2277 	}
2278 
2279 	return id;
2280 out_id:
2281 	rdma_destroy_id(id);
2282 out:
2283 	return ERR_PTR(ret);
2284 }
2285 
2286 static int
2287 isert_setup_np(struct iscsi_np *np,
2288 	       struct sockaddr_storage *ksockaddr)
2289 {
2290 	struct isert_np *isert_np;
2291 	struct rdma_cm_id *isert_lid;
2292 	int ret;
2293 
2294 	isert_np = kzalloc(sizeof(struct isert_np), GFP_KERNEL);
2295 	if (!isert_np) {
2296 		isert_err("Unable to allocate struct isert_np\n");
2297 		return -ENOMEM;
2298 	}
2299 	sema_init(&isert_np->sem, 0);
2300 	mutex_init(&isert_np->mutex);
2301 	INIT_LIST_HEAD(&isert_np->accepted);
2302 	INIT_LIST_HEAD(&isert_np->pending);
2303 	isert_np->np = np;
2304 
2305 	/*
2306 	 * Setup the np->np_sockaddr from the passed sockaddr setup
2307 	 * in iscsi_target_configfs.c code..
2308 	 */
2309 	memcpy(&np->np_sockaddr, ksockaddr,
2310 	       sizeof(struct sockaddr_storage));
2311 
2312 	isert_lid = isert_setup_id(isert_np);
2313 	if (IS_ERR(isert_lid)) {
2314 		ret = PTR_ERR(isert_lid);
2315 		goto out;
2316 	}
2317 
2318 	isert_np->cm_id = isert_lid;
2319 	np->np_context = isert_np;
2320 
2321 	return 0;
2322 
2323 out:
2324 	kfree(isert_np);
2325 
2326 	return ret;
2327 }
2328 
2329 static int
2330 isert_rdma_accept(struct isert_conn *isert_conn)
2331 {
2332 	struct rdma_cm_id *cm_id = isert_conn->cm_id;
2333 	struct rdma_conn_param cp;
2334 	int ret;
2335 	struct iser_cm_hdr rsp_hdr;
2336 
2337 	memset(&cp, 0, sizeof(struct rdma_conn_param));
2338 	cp.initiator_depth = isert_conn->initiator_depth;
2339 	cp.retry_count = 7;
2340 	cp.rnr_retry_count = 7;
2341 
2342 	memset(&rsp_hdr, 0, sizeof(rsp_hdr));
2343 	rsp_hdr.flags = ISERT_ZBVA_NOT_USED;
2344 	if (!isert_conn->snd_w_inv)
2345 		rsp_hdr.flags = rsp_hdr.flags | ISERT_SEND_W_INV_NOT_USED;
2346 	cp.private_data = (void *)&rsp_hdr;
2347 	cp.private_data_len = sizeof(rsp_hdr);
2348 
2349 	ret = rdma_accept(cm_id, &cp);
2350 	if (ret) {
2351 		isert_err("rdma_accept() failed with: %d\n", ret);
2352 		return ret;
2353 	}
2354 
2355 	return 0;
2356 }
2357 
2358 static int
2359 isert_get_login_rx(struct iscsi_conn *conn, struct iscsi_login *login)
2360 {
2361 	struct isert_conn *isert_conn = conn->context;
2362 	int ret;
2363 
2364 	isert_info("before login_req comp conn: %p\n", isert_conn);
2365 	ret = wait_for_completion_interruptible(&isert_conn->login_req_comp);
2366 	if (ret) {
2367 		isert_err("isert_conn %p interrupted before got login req\n",
2368 			  isert_conn);
2369 		return ret;
2370 	}
2371 	reinit_completion(&isert_conn->login_req_comp);
2372 
2373 	/*
2374 	 * For login requests after the first PDU, isert_rx_login_req() will
2375 	 * kick schedule_delayed_work(&conn->login_work) as the packet is
2376 	 * received, which turns this callback from iscsi_target_do_login_rx()
2377 	 * into a NOP.
2378 	 */
2379 	if (!login->first_request)
2380 		return 0;
2381 
2382 	isert_rx_login_req(isert_conn);
2383 
2384 	isert_info("before login_comp conn: %p\n", conn);
2385 	ret = wait_for_completion_interruptible(&isert_conn->login_comp);
2386 	if (ret)
2387 		return ret;
2388 
2389 	isert_info("processing login->req: %p\n", login->req);
2390 
2391 	return 0;
2392 }
2393 
2394 static void
2395 isert_set_conn_info(struct iscsi_np *np, struct iscsi_conn *conn,
2396 		    struct isert_conn *isert_conn)
2397 {
2398 	struct rdma_cm_id *cm_id = isert_conn->cm_id;
2399 	struct rdma_route *cm_route = &cm_id->route;
2400 
2401 	conn->login_family = np->np_sockaddr.ss_family;
2402 
2403 	conn->login_sockaddr = cm_route->addr.dst_addr;
2404 	conn->local_sockaddr = cm_route->addr.src_addr;
2405 }
2406 
2407 static int
2408 isert_accept_np(struct iscsi_np *np, struct iscsi_conn *conn)
2409 {
2410 	struct isert_np *isert_np = np->np_context;
2411 	struct isert_conn *isert_conn;
2412 	int ret;
2413 
2414 accept_wait:
2415 	ret = down_interruptible(&isert_np->sem);
2416 	if (ret)
2417 		return -ENODEV;
2418 
2419 	spin_lock_bh(&np->np_thread_lock);
2420 	if (np->np_thread_state >= ISCSI_NP_THREAD_RESET) {
2421 		spin_unlock_bh(&np->np_thread_lock);
2422 		isert_dbg("np_thread_state %d\n",
2423 			 np->np_thread_state);
2424 		/**
2425 		 * No point in stalling here when np_thread
2426 		 * is in state RESET/SHUTDOWN/EXIT - bail
2427 		 **/
2428 		return -ENODEV;
2429 	}
2430 	spin_unlock_bh(&np->np_thread_lock);
2431 
2432 	mutex_lock(&isert_np->mutex);
2433 	if (list_empty(&isert_np->pending)) {
2434 		mutex_unlock(&isert_np->mutex);
2435 		goto accept_wait;
2436 	}
2437 	isert_conn = list_first_entry(&isert_np->pending,
2438 			struct isert_conn, node);
2439 	list_del_init(&isert_conn->node);
2440 	mutex_unlock(&isert_np->mutex);
2441 
2442 	conn->context = isert_conn;
2443 	isert_conn->conn = conn;
2444 	isert_conn->state = ISER_CONN_BOUND;
2445 
2446 	isert_set_conn_info(np, conn, isert_conn);
2447 
2448 	isert_dbg("Processing isert_conn: %p\n", isert_conn);
2449 
2450 	return 0;
2451 }
2452 
2453 static void
2454 isert_free_np(struct iscsi_np *np)
2455 {
2456 	struct isert_np *isert_np = np->np_context;
2457 	struct isert_conn *isert_conn, *n;
2458 
2459 	if (isert_np->cm_id)
2460 		rdma_destroy_id(isert_np->cm_id);
2461 
2462 	/*
2463 	 * FIXME: At this point we don't have a good way to insure
2464 	 * that at this point we don't have hanging connections that
2465 	 * completed RDMA establishment but didn't start iscsi login
2466 	 * process. So work-around this by cleaning up what ever piled
2467 	 * up in accepted and pending lists.
2468 	 */
2469 	mutex_lock(&isert_np->mutex);
2470 	if (!list_empty(&isert_np->pending)) {
2471 		isert_info("Still have isert pending connections\n");
2472 		list_for_each_entry_safe(isert_conn, n,
2473 					 &isert_np->pending,
2474 					 node) {
2475 			isert_info("cleaning isert_conn %p state (%d)\n",
2476 				   isert_conn, isert_conn->state);
2477 			isert_connect_release(isert_conn);
2478 		}
2479 	}
2480 
2481 	if (!list_empty(&isert_np->accepted)) {
2482 		isert_info("Still have isert accepted connections\n");
2483 		list_for_each_entry_safe(isert_conn, n,
2484 					 &isert_np->accepted,
2485 					 node) {
2486 			isert_info("cleaning isert_conn %p state (%d)\n",
2487 				   isert_conn, isert_conn->state);
2488 			isert_connect_release(isert_conn);
2489 		}
2490 	}
2491 	mutex_unlock(&isert_np->mutex);
2492 
2493 	np->np_context = NULL;
2494 	kfree(isert_np);
2495 }
2496 
2497 static void isert_release_work(struct work_struct *work)
2498 {
2499 	struct isert_conn *isert_conn = container_of(work,
2500 						     struct isert_conn,
2501 						     release_work);
2502 
2503 	isert_info("Starting release conn %p\n", isert_conn);
2504 
2505 	mutex_lock(&isert_conn->mutex);
2506 	isert_conn->state = ISER_CONN_DOWN;
2507 	mutex_unlock(&isert_conn->mutex);
2508 
2509 	isert_info("Destroying conn %p\n", isert_conn);
2510 	isert_put_conn(isert_conn);
2511 }
2512 
2513 static void
2514 isert_wait4logout(struct isert_conn *isert_conn)
2515 {
2516 	struct iscsi_conn *conn = isert_conn->conn;
2517 
2518 	isert_info("conn %p\n", isert_conn);
2519 
2520 	if (isert_conn->logout_posted) {
2521 		isert_info("conn %p wait for conn_logout_comp\n", isert_conn);
2522 		wait_for_completion_timeout(&conn->conn_logout_comp,
2523 					    SECONDS_FOR_LOGOUT_COMP * HZ);
2524 	}
2525 }
2526 
2527 static void
2528 isert_wait4cmds(struct iscsi_conn *conn)
2529 {
2530 	isert_info("iscsi_conn %p\n", conn);
2531 
2532 	if (conn->sess) {
2533 		target_sess_cmd_list_set_waiting(conn->sess->se_sess);
2534 		target_wait_for_sess_cmds(conn->sess->se_sess);
2535 	}
2536 }
2537 
2538 /**
2539  * isert_put_unsol_pending_cmds() - Drop commands waiting for
2540  *     unsolicitate dataout
2541  * @conn:    iscsi connection
2542  *
2543  * We might still have commands that are waiting for unsolicited
2544  * dataouts messages. We must put the extra reference on those
2545  * before blocking on the target_wait_for_session_cmds
2546  */
2547 static void
2548 isert_put_unsol_pending_cmds(struct iscsi_conn *conn)
2549 {
2550 	struct iscsi_cmd *cmd, *tmp;
2551 	static LIST_HEAD(drop_cmd_list);
2552 
2553 	spin_lock_bh(&conn->cmd_lock);
2554 	list_for_each_entry_safe(cmd, tmp, &conn->conn_cmd_list, i_conn_node) {
2555 		if ((cmd->cmd_flags & ICF_NON_IMMEDIATE_UNSOLICITED_DATA) &&
2556 		    (cmd->write_data_done < conn->sess->sess_ops->FirstBurstLength) &&
2557 		    (cmd->write_data_done < cmd->se_cmd.data_length))
2558 			list_move_tail(&cmd->i_conn_node, &drop_cmd_list);
2559 	}
2560 	spin_unlock_bh(&conn->cmd_lock);
2561 
2562 	list_for_each_entry_safe(cmd, tmp, &drop_cmd_list, i_conn_node) {
2563 		list_del_init(&cmd->i_conn_node);
2564 		if (cmd->i_state != ISTATE_REMOVE) {
2565 			struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2566 
2567 			isert_info("conn %p dropping cmd %p\n", conn, cmd);
2568 			isert_put_cmd(isert_cmd, true);
2569 		}
2570 	}
2571 }
2572 
2573 static void isert_wait_conn(struct iscsi_conn *conn)
2574 {
2575 	struct isert_conn *isert_conn = conn->context;
2576 
2577 	isert_info("Starting conn %p\n", isert_conn);
2578 
2579 	mutex_lock(&isert_conn->mutex);
2580 	isert_conn_terminate(isert_conn);
2581 	mutex_unlock(&isert_conn->mutex);
2582 
2583 	ib_drain_qp(isert_conn->qp);
2584 	isert_put_unsol_pending_cmds(conn);
2585 	isert_wait4cmds(conn);
2586 	isert_wait4logout(isert_conn);
2587 
2588 	queue_work(isert_release_wq, &isert_conn->release_work);
2589 }
2590 
2591 static void isert_free_conn(struct iscsi_conn *conn)
2592 {
2593 	struct isert_conn *isert_conn = conn->context;
2594 
2595 	ib_drain_qp(isert_conn->qp);
2596 	isert_put_conn(isert_conn);
2597 }
2598 
2599 static void isert_get_rx_pdu(struct iscsi_conn *conn)
2600 {
2601 	struct completion comp;
2602 
2603 	init_completion(&comp);
2604 
2605 	wait_for_completion_interruptible(&comp);
2606 }
2607 
2608 static struct iscsit_transport iser_target_transport = {
2609 	.name			= "IB/iSER",
2610 	.transport_type		= ISCSI_INFINIBAND,
2611 	.rdma_shutdown		= true,
2612 	.priv_size		= sizeof(struct isert_cmd),
2613 	.owner			= THIS_MODULE,
2614 	.iscsit_setup_np	= isert_setup_np,
2615 	.iscsit_accept_np	= isert_accept_np,
2616 	.iscsit_free_np		= isert_free_np,
2617 	.iscsit_wait_conn	= isert_wait_conn,
2618 	.iscsit_free_conn	= isert_free_conn,
2619 	.iscsit_get_login_rx	= isert_get_login_rx,
2620 	.iscsit_put_login_tx	= isert_put_login_tx,
2621 	.iscsit_immediate_queue	= isert_immediate_queue,
2622 	.iscsit_response_queue	= isert_response_queue,
2623 	.iscsit_get_dataout	= isert_get_dataout,
2624 	.iscsit_queue_data_in	= isert_put_datain,
2625 	.iscsit_queue_status	= isert_put_response,
2626 	.iscsit_aborted_task	= isert_aborted_task,
2627 	.iscsit_get_rx_pdu	= isert_get_rx_pdu,
2628 	.iscsit_get_sup_prot_ops = isert_get_sup_prot_ops,
2629 };
2630 
2631 static int __init isert_init(void)
2632 {
2633 	int ret;
2634 
2635 	isert_comp_wq = alloc_workqueue("isert_comp_wq",
2636 					WQ_UNBOUND | WQ_HIGHPRI, 0);
2637 	if (!isert_comp_wq) {
2638 		isert_err("Unable to allocate isert_comp_wq\n");
2639 		ret = -ENOMEM;
2640 		return -ENOMEM;
2641 	}
2642 
2643 	isert_release_wq = alloc_workqueue("isert_release_wq", WQ_UNBOUND,
2644 					WQ_UNBOUND_MAX_ACTIVE);
2645 	if (!isert_release_wq) {
2646 		isert_err("Unable to allocate isert_release_wq\n");
2647 		ret = -ENOMEM;
2648 		goto destroy_comp_wq;
2649 	}
2650 
2651 	iscsit_register_transport(&iser_target_transport);
2652 	isert_info("iSER_TARGET[0] - Loaded iser_target_transport\n");
2653 
2654 	return 0;
2655 
2656 destroy_comp_wq:
2657 	destroy_workqueue(isert_comp_wq);
2658 
2659 	return ret;
2660 }
2661 
2662 static void __exit isert_exit(void)
2663 {
2664 	flush_scheduled_work();
2665 	destroy_workqueue(isert_release_wq);
2666 	destroy_workqueue(isert_comp_wq);
2667 	iscsit_unregister_transport(&iser_target_transport);
2668 	isert_info("iSER_TARGET[0] - Released iser_target_transport\n");
2669 }
2670 
2671 MODULE_DESCRIPTION("iSER-Target for mainline target infrastructure");
2672 MODULE_VERSION("1.0");
2673 MODULE_AUTHOR("nab@Linux-iSCSI.org");
2674 MODULE_LICENSE("GPL");
2675 
2676 module_init(isert_init);
2677 module_exit(isert_exit);
2678