1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*******************************************************************************
3  * This file contains iSCSI extentions for RDMA (iSER) Verbs
4  *
5  * (c) Copyright 2013 Datera, Inc.
6  *
7  * Nicholas A. Bellinger <nab@linux-iscsi.org>
8  *
9  ****************************************************************************/
10 
11 #include <linux/string.h>
12 #include <linux/module.h>
13 #include <linux/scatterlist.h>
14 #include <linux/socket.h>
15 #include <linux/in.h>
16 #include <linux/in6.h>
17 #include <rdma/ib_verbs.h>
18 #include <rdma/rdma_cm.h>
19 #include <target/target_core_base.h>
20 #include <target/target_core_fabric.h>
21 #include <target/iscsi/iscsi_transport.h>
22 #include <linux/semaphore.h>
23 
24 #include "ib_isert.h"
25 
26 #define	ISERT_MAX_CONN		8
27 #define ISER_MAX_RX_CQ_LEN	(ISERT_QP_MAX_RECV_DTOS * ISERT_MAX_CONN)
28 #define ISER_MAX_TX_CQ_LEN \
29 	((ISERT_QP_MAX_REQ_DTOS + ISCSI_DEF_XMIT_CMDS_MAX) * ISERT_MAX_CONN)
30 #define ISER_MAX_CQ_LEN		(ISER_MAX_RX_CQ_LEN + ISER_MAX_TX_CQ_LEN + \
31 				 ISERT_MAX_CONN)
32 
33 static int isert_debug_level;
34 module_param_named(debug_level, isert_debug_level, int, 0644);
35 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0 (default:0)");
36 
37 static DEFINE_MUTEX(device_list_mutex);
38 static LIST_HEAD(device_list);
39 static struct workqueue_struct *isert_comp_wq;
40 static struct workqueue_struct *isert_release_wq;
41 
42 static int
43 isert_put_response(struct iscsi_conn *conn, struct iscsi_cmd *cmd);
44 static int
45 isert_login_post_recv(struct isert_conn *isert_conn);
46 static int
47 isert_rdma_accept(struct isert_conn *isert_conn);
48 struct rdma_cm_id *isert_setup_id(struct isert_np *isert_np);
49 
50 static void isert_release_work(struct work_struct *work);
51 static void isert_recv_done(struct ib_cq *cq, struct ib_wc *wc);
52 static void isert_send_done(struct ib_cq *cq, struct ib_wc *wc);
53 static void isert_login_recv_done(struct ib_cq *cq, struct ib_wc *wc);
54 static void isert_login_send_done(struct ib_cq *cq, struct ib_wc *wc);
55 
56 static inline bool
57 isert_prot_cmd(struct isert_conn *conn, struct se_cmd *cmd)
58 {
59 	return (conn->pi_support &&
60 		cmd->prot_op != TARGET_PROT_NORMAL);
61 }
62 
63 
64 static void
65 isert_qp_event_callback(struct ib_event *e, void *context)
66 {
67 	struct isert_conn *isert_conn = context;
68 
69 	isert_err("%s (%d): conn %p\n",
70 		  ib_event_msg(e->event), e->event, isert_conn);
71 
72 	switch (e->event) {
73 	case IB_EVENT_COMM_EST:
74 		rdma_notify(isert_conn->cm_id, IB_EVENT_COMM_EST);
75 		break;
76 	case IB_EVENT_QP_LAST_WQE_REACHED:
77 		isert_warn("Reached TX IB_EVENT_QP_LAST_WQE_REACHED\n");
78 		break;
79 	default:
80 		break;
81 	}
82 }
83 
84 static struct isert_comp *
85 isert_comp_get(struct isert_conn *isert_conn)
86 {
87 	struct isert_device *device = isert_conn->device;
88 	struct isert_comp *comp;
89 	int i, min = 0;
90 
91 	mutex_lock(&device_list_mutex);
92 	for (i = 0; i < device->comps_used; i++)
93 		if (device->comps[i].active_qps <
94 		    device->comps[min].active_qps)
95 			min = i;
96 	comp = &device->comps[min];
97 	comp->active_qps++;
98 	mutex_unlock(&device_list_mutex);
99 
100 	isert_info("conn %p, using comp %p min_index: %d\n",
101 		   isert_conn, comp, min);
102 
103 	return comp;
104 }
105 
106 static void
107 isert_comp_put(struct isert_comp *comp)
108 {
109 	mutex_lock(&device_list_mutex);
110 	comp->active_qps--;
111 	mutex_unlock(&device_list_mutex);
112 }
113 
114 static struct ib_qp *
115 isert_create_qp(struct isert_conn *isert_conn,
116 		struct isert_comp *comp,
117 		struct rdma_cm_id *cma_id)
118 {
119 	struct isert_device *device = isert_conn->device;
120 	struct ib_qp_init_attr attr;
121 	int ret;
122 
123 	memset(&attr, 0, sizeof(struct ib_qp_init_attr));
124 	attr.event_handler = isert_qp_event_callback;
125 	attr.qp_context = isert_conn;
126 	attr.send_cq = comp->cq;
127 	attr.recv_cq = comp->cq;
128 	attr.cap.max_send_wr = ISERT_QP_MAX_REQ_DTOS + 1;
129 	attr.cap.max_recv_wr = ISERT_QP_MAX_RECV_DTOS + 1;
130 	attr.cap.max_rdma_ctxs = ISCSI_DEF_XMIT_CMDS_MAX;
131 	attr.cap.max_send_sge = device->ib_device->attrs.max_send_sge;
132 	attr.cap.max_recv_sge = 1;
133 	attr.sq_sig_type = IB_SIGNAL_REQ_WR;
134 	attr.qp_type = IB_QPT_RC;
135 	if (device->pi_capable)
136 		attr.create_flags |= IB_QP_CREATE_INTEGRITY_EN;
137 
138 	ret = rdma_create_qp(cma_id, device->pd, &attr);
139 	if (ret) {
140 		isert_err("rdma_create_qp failed for cma_id %d\n", ret);
141 		return ERR_PTR(ret);
142 	}
143 
144 	return cma_id->qp;
145 }
146 
147 static int
148 isert_conn_setup_qp(struct isert_conn *isert_conn, struct rdma_cm_id *cma_id)
149 {
150 	struct isert_comp *comp;
151 	int ret;
152 
153 	comp = isert_comp_get(isert_conn);
154 	isert_conn->qp = isert_create_qp(isert_conn, comp, cma_id);
155 	if (IS_ERR(isert_conn->qp)) {
156 		ret = PTR_ERR(isert_conn->qp);
157 		goto err;
158 	}
159 
160 	return 0;
161 err:
162 	isert_comp_put(comp);
163 	return ret;
164 }
165 
166 static int
167 isert_alloc_rx_descriptors(struct isert_conn *isert_conn)
168 {
169 	struct isert_device *device = isert_conn->device;
170 	struct ib_device *ib_dev = device->ib_device;
171 	struct iser_rx_desc *rx_desc;
172 	struct ib_sge *rx_sg;
173 	u64 dma_addr;
174 	int i, j;
175 
176 	isert_conn->rx_descs = kcalloc(ISERT_QP_MAX_RECV_DTOS,
177 				       sizeof(struct iser_rx_desc),
178 				       GFP_KERNEL);
179 	if (!isert_conn->rx_descs)
180 		return -ENOMEM;
181 
182 	rx_desc = isert_conn->rx_descs;
183 
184 	for (i = 0; i < ISERT_QP_MAX_RECV_DTOS; i++, rx_desc++)  {
185 		dma_addr = ib_dma_map_single(ib_dev, (void *)rx_desc,
186 					ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
187 		if (ib_dma_mapping_error(ib_dev, dma_addr))
188 			goto dma_map_fail;
189 
190 		rx_desc->dma_addr = dma_addr;
191 
192 		rx_sg = &rx_desc->rx_sg;
193 		rx_sg->addr = rx_desc->dma_addr;
194 		rx_sg->length = ISER_RX_PAYLOAD_SIZE;
195 		rx_sg->lkey = device->pd->local_dma_lkey;
196 		rx_desc->rx_cqe.done = isert_recv_done;
197 	}
198 
199 	return 0;
200 
201 dma_map_fail:
202 	rx_desc = isert_conn->rx_descs;
203 	for (j = 0; j < i; j++, rx_desc++) {
204 		ib_dma_unmap_single(ib_dev, rx_desc->dma_addr,
205 				    ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
206 	}
207 	kfree(isert_conn->rx_descs);
208 	isert_conn->rx_descs = NULL;
209 	isert_err("conn %p failed to allocate rx descriptors\n", isert_conn);
210 	return -ENOMEM;
211 }
212 
213 static void
214 isert_free_rx_descriptors(struct isert_conn *isert_conn)
215 {
216 	struct ib_device *ib_dev = isert_conn->device->ib_device;
217 	struct iser_rx_desc *rx_desc;
218 	int i;
219 
220 	if (!isert_conn->rx_descs)
221 		return;
222 
223 	rx_desc = isert_conn->rx_descs;
224 	for (i = 0; i < ISERT_QP_MAX_RECV_DTOS; i++, rx_desc++)  {
225 		ib_dma_unmap_single(ib_dev, rx_desc->dma_addr,
226 				    ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
227 	}
228 
229 	kfree(isert_conn->rx_descs);
230 	isert_conn->rx_descs = NULL;
231 }
232 
233 static void
234 isert_free_comps(struct isert_device *device)
235 {
236 	int i;
237 
238 	for (i = 0; i < device->comps_used; i++) {
239 		struct isert_comp *comp = &device->comps[i];
240 
241 		if (comp->cq)
242 			ib_free_cq(comp->cq);
243 	}
244 	kfree(device->comps);
245 }
246 
247 static int
248 isert_alloc_comps(struct isert_device *device)
249 {
250 	int i, max_cqe, ret = 0;
251 
252 	device->comps_used = min(ISERT_MAX_CQ, min_t(int, num_online_cpus(),
253 				 device->ib_device->num_comp_vectors));
254 
255 	isert_info("Using %d CQs, %s supports %d vectors support "
256 		   "pi_capable %d\n",
257 		   device->comps_used, dev_name(&device->ib_device->dev),
258 		   device->ib_device->num_comp_vectors,
259 		   device->pi_capable);
260 
261 	device->comps = kcalloc(device->comps_used, sizeof(struct isert_comp),
262 				GFP_KERNEL);
263 	if (!device->comps)
264 		return -ENOMEM;
265 
266 	max_cqe = min(ISER_MAX_CQ_LEN, device->ib_device->attrs.max_cqe);
267 
268 	for (i = 0; i < device->comps_used; i++) {
269 		struct isert_comp *comp = &device->comps[i];
270 
271 		comp->device = device;
272 		comp->cq = ib_alloc_cq(device->ib_device, comp, max_cqe, i,
273 				IB_POLL_WORKQUEUE);
274 		if (IS_ERR(comp->cq)) {
275 			isert_err("Unable to allocate cq\n");
276 			ret = PTR_ERR(comp->cq);
277 			comp->cq = NULL;
278 			goto out_cq;
279 		}
280 	}
281 
282 	return 0;
283 out_cq:
284 	isert_free_comps(device);
285 	return ret;
286 }
287 
288 static int
289 isert_create_device_ib_res(struct isert_device *device)
290 {
291 	struct ib_device *ib_dev = device->ib_device;
292 	int ret;
293 
294 	isert_dbg("devattr->max_send_sge: %d devattr->max_recv_sge %d\n",
295 		  ib_dev->attrs.max_send_sge, ib_dev->attrs.max_recv_sge);
296 	isert_dbg("devattr->max_sge_rd: %d\n", ib_dev->attrs.max_sge_rd);
297 
298 	ret = isert_alloc_comps(device);
299 	if (ret)
300 		goto out;
301 
302 	device->pd = ib_alloc_pd(ib_dev, 0);
303 	if (IS_ERR(device->pd)) {
304 		ret = PTR_ERR(device->pd);
305 		isert_err("failed to allocate pd, device %p, ret=%d\n",
306 			  device, ret);
307 		goto out_cq;
308 	}
309 
310 	/* Check signature cap */
311 	device->pi_capable = ib_dev->attrs.device_cap_flags &
312 			     IB_DEVICE_INTEGRITY_HANDOVER ? true : false;
313 
314 	return 0;
315 
316 out_cq:
317 	isert_free_comps(device);
318 out:
319 	if (ret > 0)
320 		ret = -EINVAL;
321 	return ret;
322 }
323 
324 static void
325 isert_free_device_ib_res(struct isert_device *device)
326 {
327 	isert_info("device %p\n", device);
328 
329 	ib_dealloc_pd(device->pd);
330 	isert_free_comps(device);
331 }
332 
333 static void
334 isert_device_put(struct isert_device *device)
335 {
336 	mutex_lock(&device_list_mutex);
337 	device->refcount--;
338 	isert_info("device %p refcount %d\n", device, device->refcount);
339 	if (!device->refcount) {
340 		isert_free_device_ib_res(device);
341 		list_del(&device->dev_node);
342 		kfree(device);
343 	}
344 	mutex_unlock(&device_list_mutex);
345 }
346 
347 static struct isert_device *
348 isert_device_get(struct rdma_cm_id *cma_id)
349 {
350 	struct isert_device *device;
351 	int ret;
352 
353 	mutex_lock(&device_list_mutex);
354 	list_for_each_entry(device, &device_list, dev_node) {
355 		if (device->ib_device->node_guid == cma_id->device->node_guid) {
356 			device->refcount++;
357 			isert_info("Found iser device %p refcount %d\n",
358 				   device, device->refcount);
359 			mutex_unlock(&device_list_mutex);
360 			return device;
361 		}
362 	}
363 
364 	device = kzalloc(sizeof(struct isert_device), GFP_KERNEL);
365 	if (!device) {
366 		mutex_unlock(&device_list_mutex);
367 		return ERR_PTR(-ENOMEM);
368 	}
369 
370 	INIT_LIST_HEAD(&device->dev_node);
371 
372 	device->ib_device = cma_id->device;
373 	ret = isert_create_device_ib_res(device);
374 	if (ret) {
375 		kfree(device);
376 		mutex_unlock(&device_list_mutex);
377 		return ERR_PTR(ret);
378 	}
379 
380 	device->refcount++;
381 	list_add_tail(&device->dev_node, &device_list);
382 	isert_info("Created a new iser device %p refcount %d\n",
383 		   device, device->refcount);
384 	mutex_unlock(&device_list_mutex);
385 
386 	return device;
387 }
388 
389 static void
390 isert_init_conn(struct isert_conn *isert_conn)
391 {
392 	isert_conn->state = ISER_CONN_INIT;
393 	INIT_LIST_HEAD(&isert_conn->node);
394 	init_completion(&isert_conn->login_comp);
395 	init_completion(&isert_conn->login_req_comp);
396 	init_waitqueue_head(&isert_conn->rem_wait);
397 	kref_init(&isert_conn->kref);
398 	mutex_init(&isert_conn->mutex);
399 	INIT_WORK(&isert_conn->release_work, isert_release_work);
400 }
401 
402 static void
403 isert_free_login_buf(struct isert_conn *isert_conn)
404 {
405 	struct ib_device *ib_dev = isert_conn->device->ib_device;
406 
407 	ib_dma_unmap_single(ib_dev, isert_conn->login_rsp_dma,
408 			    ISER_RX_PAYLOAD_SIZE, DMA_TO_DEVICE);
409 	kfree(isert_conn->login_rsp_buf);
410 
411 	ib_dma_unmap_single(ib_dev, isert_conn->login_req_dma,
412 			    ISER_RX_PAYLOAD_SIZE,
413 			    DMA_FROM_DEVICE);
414 	kfree(isert_conn->login_req_buf);
415 }
416 
417 static int
418 isert_alloc_login_buf(struct isert_conn *isert_conn,
419 		      struct ib_device *ib_dev)
420 {
421 	int ret;
422 
423 	isert_conn->login_req_buf = kzalloc(sizeof(*isert_conn->login_req_buf),
424 			GFP_KERNEL);
425 	if (!isert_conn->login_req_buf)
426 		return -ENOMEM;
427 
428 	isert_conn->login_req_dma = ib_dma_map_single(ib_dev,
429 				isert_conn->login_req_buf,
430 				ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
431 	ret = ib_dma_mapping_error(ib_dev, isert_conn->login_req_dma);
432 	if (ret) {
433 		isert_err("login_req_dma mapping error: %d\n", ret);
434 		isert_conn->login_req_dma = 0;
435 		goto out_free_login_req_buf;
436 	}
437 
438 	isert_conn->login_rsp_buf = kzalloc(ISER_RX_PAYLOAD_SIZE, GFP_KERNEL);
439 	if (!isert_conn->login_rsp_buf) {
440 		ret = -ENOMEM;
441 		goto out_unmap_login_req_buf;
442 	}
443 
444 	isert_conn->login_rsp_dma = ib_dma_map_single(ib_dev,
445 					isert_conn->login_rsp_buf,
446 					ISER_RX_PAYLOAD_SIZE, DMA_TO_DEVICE);
447 	ret = ib_dma_mapping_error(ib_dev, isert_conn->login_rsp_dma);
448 	if (ret) {
449 		isert_err("login_rsp_dma mapping error: %d\n", ret);
450 		isert_conn->login_rsp_dma = 0;
451 		goto out_free_login_rsp_buf;
452 	}
453 
454 	return 0;
455 
456 out_free_login_rsp_buf:
457 	kfree(isert_conn->login_rsp_buf);
458 out_unmap_login_req_buf:
459 	ib_dma_unmap_single(ib_dev, isert_conn->login_req_dma,
460 			    ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
461 out_free_login_req_buf:
462 	kfree(isert_conn->login_req_buf);
463 	return ret;
464 }
465 
466 static void
467 isert_set_nego_params(struct isert_conn *isert_conn,
468 		      struct rdma_conn_param *param)
469 {
470 	struct ib_device_attr *attr = &isert_conn->device->ib_device->attrs;
471 
472 	/* Set max inflight RDMA READ requests */
473 	isert_conn->initiator_depth = min_t(u8, param->initiator_depth,
474 				attr->max_qp_init_rd_atom);
475 	isert_dbg("Using initiator_depth: %u\n", isert_conn->initiator_depth);
476 
477 	if (param->private_data) {
478 		u8 flags = *(u8 *)param->private_data;
479 
480 		/*
481 		 * use remote invalidation if the both initiator
482 		 * and the HCA support it
483 		 */
484 		isert_conn->snd_w_inv = !(flags & ISER_SEND_W_INV_NOT_SUP) &&
485 					  (attr->device_cap_flags &
486 					   IB_DEVICE_MEM_MGT_EXTENSIONS);
487 		if (isert_conn->snd_w_inv)
488 			isert_info("Using remote invalidation\n");
489 	}
490 }
491 
492 static int
493 isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
494 {
495 	struct isert_np *isert_np = cma_id->context;
496 	struct iscsi_np *np = isert_np->np;
497 	struct isert_conn *isert_conn;
498 	struct isert_device *device;
499 	int ret = 0;
500 
501 	spin_lock_bh(&np->np_thread_lock);
502 	if (!np->enabled) {
503 		spin_unlock_bh(&np->np_thread_lock);
504 		isert_dbg("iscsi_np is not enabled, reject connect request\n");
505 		return rdma_reject(cma_id, NULL, 0);
506 	}
507 	spin_unlock_bh(&np->np_thread_lock);
508 
509 	isert_dbg("cma_id: %p, portal: %p\n",
510 		 cma_id, cma_id->context);
511 
512 	isert_conn = kzalloc(sizeof(struct isert_conn), GFP_KERNEL);
513 	if (!isert_conn)
514 		return -ENOMEM;
515 
516 	isert_init_conn(isert_conn);
517 	isert_conn->cm_id = cma_id;
518 
519 	ret = isert_alloc_login_buf(isert_conn, cma_id->device);
520 	if (ret)
521 		goto out;
522 
523 	device = isert_device_get(cma_id);
524 	if (IS_ERR(device)) {
525 		ret = PTR_ERR(device);
526 		goto out_rsp_dma_map;
527 	}
528 	isert_conn->device = device;
529 
530 	isert_set_nego_params(isert_conn, &event->param.conn);
531 
532 	ret = isert_conn_setup_qp(isert_conn, cma_id);
533 	if (ret)
534 		goto out_conn_dev;
535 
536 	ret = isert_login_post_recv(isert_conn);
537 	if (ret)
538 		goto out_conn_dev;
539 
540 	ret = isert_rdma_accept(isert_conn);
541 	if (ret)
542 		goto out_conn_dev;
543 
544 	mutex_lock(&isert_np->mutex);
545 	list_add_tail(&isert_conn->node, &isert_np->accepted);
546 	mutex_unlock(&isert_np->mutex);
547 
548 	return 0;
549 
550 out_conn_dev:
551 	isert_device_put(device);
552 out_rsp_dma_map:
553 	isert_free_login_buf(isert_conn);
554 out:
555 	kfree(isert_conn);
556 	rdma_reject(cma_id, NULL, 0);
557 	return ret;
558 }
559 
560 static void
561 isert_connect_release(struct isert_conn *isert_conn)
562 {
563 	struct isert_device *device = isert_conn->device;
564 
565 	isert_dbg("conn %p\n", isert_conn);
566 
567 	BUG_ON(!device);
568 
569 	isert_free_rx_descriptors(isert_conn);
570 	if (isert_conn->cm_id &&
571 	    !isert_conn->dev_removed)
572 		rdma_destroy_id(isert_conn->cm_id);
573 
574 	if (isert_conn->qp) {
575 		struct isert_comp *comp = isert_conn->qp->recv_cq->cq_context;
576 
577 		isert_comp_put(comp);
578 		ib_destroy_qp(isert_conn->qp);
579 	}
580 
581 	if (isert_conn->login_req_buf)
582 		isert_free_login_buf(isert_conn);
583 
584 	isert_device_put(device);
585 
586 	if (isert_conn->dev_removed)
587 		wake_up_interruptible(&isert_conn->rem_wait);
588 	else
589 		kfree(isert_conn);
590 }
591 
592 static void
593 isert_connected_handler(struct rdma_cm_id *cma_id)
594 {
595 	struct isert_conn *isert_conn = cma_id->qp->qp_context;
596 	struct isert_np *isert_np = cma_id->context;
597 
598 	isert_info("conn %p\n", isert_conn);
599 
600 	mutex_lock(&isert_conn->mutex);
601 	isert_conn->state = ISER_CONN_UP;
602 	kref_get(&isert_conn->kref);
603 	mutex_unlock(&isert_conn->mutex);
604 
605 	mutex_lock(&isert_np->mutex);
606 	list_move_tail(&isert_conn->node, &isert_np->pending);
607 	mutex_unlock(&isert_np->mutex);
608 
609 	isert_info("np %p: Allow accept_np to continue\n", isert_np);
610 	up(&isert_np->sem);
611 }
612 
613 static void
614 isert_release_kref(struct kref *kref)
615 {
616 	struct isert_conn *isert_conn = container_of(kref,
617 				struct isert_conn, kref);
618 
619 	isert_info("conn %p final kref %s/%d\n", isert_conn, current->comm,
620 		   current->pid);
621 
622 	isert_connect_release(isert_conn);
623 }
624 
625 static void
626 isert_put_conn(struct isert_conn *isert_conn)
627 {
628 	kref_put(&isert_conn->kref, isert_release_kref);
629 }
630 
631 static void
632 isert_handle_unbound_conn(struct isert_conn *isert_conn)
633 {
634 	struct isert_np *isert_np = isert_conn->cm_id->context;
635 
636 	mutex_lock(&isert_np->mutex);
637 	if (!list_empty(&isert_conn->node)) {
638 		/*
639 		 * This means iscsi doesn't know this connection
640 		 * so schedule a cleanup ourselves
641 		 */
642 		list_del_init(&isert_conn->node);
643 		isert_put_conn(isert_conn);
644 		queue_work(isert_release_wq, &isert_conn->release_work);
645 	}
646 	mutex_unlock(&isert_np->mutex);
647 }
648 
649 /**
650  * isert_conn_terminate() - Initiate connection termination
651  * @isert_conn: isert connection struct
652  *
653  * Notes:
654  * In case the connection state is BOUND, move state
655  * to TEMINATING and start teardown sequence (rdma_disconnect).
656  * In case the connection state is UP, complete flush as well.
657  *
658  * This routine must be called with mutex held. Thus it is
659  * safe to call multiple times.
660  */
661 static void
662 isert_conn_terminate(struct isert_conn *isert_conn)
663 {
664 	int err;
665 
666 	if (isert_conn->state >= ISER_CONN_TERMINATING)
667 		return;
668 
669 	isert_info("Terminating conn %p state %d\n",
670 		   isert_conn, isert_conn->state);
671 	isert_conn->state = ISER_CONN_TERMINATING;
672 	err = rdma_disconnect(isert_conn->cm_id);
673 	if (err)
674 		isert_warn("Failed rdma_disconnect isert_conn %p\n",
675 			   isert_conn);
676 }
677 
678 static int
679 isert_np_cma_handler(struct isert_np *isert_np,
680 		     enum rdma_cm_event_type event)
681 {
682 	isert_dbg("%s (%d): isert np %p\n",
683 		  rdma_event_msg(event), event, isert_np);
684 
685 	switch (event) {
686 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
687 		isert_np->cm_id = NULL;
688 		break;
689 	case RDMA_CM_EVENT_ADDR_CHANGE:
690 		isert_np->cm_id = isert_setup_id(isert_np);
691 		if (IS_ERR(isert_np->cm_id)) {
692 			isert_err("isert np %p setup id failed: %ld\n",
693 				  isert_np, PTR_ERR(isert_np->cm_id));
694 			isert_np->cm_id = NULL;
695 		}
696 		break;
697 	default:
698 		isert_err("isert np %p Unexpected event %d\n",
699 			  isert_np, event);
700 	}
701 
702 	return -1;
703 }
704 
705 static int
706 isert_disconnected_handler(struct rdma_cm_id *cma_id,
707 			   enum rdma_cm_event_type event)
708 {
709 	struct isert_conn *isert_conn = cma_id->qp->qp_context;
710 
711 	mutex_lock(&isert_conn->mutex);
712 	switch (isert_conn->state) {
713 	case ISER_CONN_TERMINATING:
714 		break;
715 	case ISER_CONN_UP:
716 		isert_conn_terminate(isert_conn);
717 		ib_drain_qp(isert_conn->qp);
718 		isert_handle_unbound_conn(isert_conn);
719 		break;
720 	case ISER_CONN_BOUND:
721 	case ISER_CONN_FULL_FEATURE: /* FALLTHRU */
722 		iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
723 		break;
724 	default:
725 		isert_warn("conn %p terminating in state %d\n",
726 			   isert_conn, isert_conn->state);
727 	}
728 	mutex_unlock(&isert_conn->mutex);
729 
730 	return 0;
731 }
732 
733 static int
734 isert_connect_error(struct rdma_cm_id *cma_id)
735 {
736 	struct isert_conn *isert_conn = cma_id->qp->qp_context;
737 
738 	ib_drain_qp(isert_conn->qp);
739 	list_del_init(&isert_conn->node);
740 	isert_conn->cm_id = NULL;
741 	isert_put_conn(isert_conn);
742 
743 	return -1;
744 }
745 
746 static int
747 isert_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
748 {
749 	struct isert_np *isert_np = cma_id->context;
750 	struct isert_conn *isert_conn;
751 	int ret = 0;
752 
753 	isert_info("%s (%d): status %d id %p np %p\n",
754 		   rdma_event_msg(event->event), event->event,
755 		   event->status, cma_id, cma_id->context);
756 
757 	if (isert_np->cm_id == cma_id)
758 		return isert_np_cma_handler(cma_id->context, event->event);
759 
760 	switch (event->event) {
761 	case RDMA_CM_EVENT_CONNECT_REQUEST:
762 		ret = isert_connect_request(cma_id, event);
763 		if (ret)
764 			isert_err("failed handle connect request %d\n", ret);
765 		break;
766 	case RDMA_CM_EVENT_ESTABLISHED:
767 		isert_connected_handler(cma_id);
768 		break;
769 	case RDMA_CM_EVENT_ADDR_CHANGE:    /* FALLTHRU */
770 	case RDMA_CM_EVENT_DISCONNECTED:   /* FALLTHRU */
771 	case RDMA_CM_EVENT_TIMEWAIT_EXIT:  /* FALLTHRU */
772 		ret = isert_disconnected_handler(cma_id, event->event);
773 		break;
774 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
775 		isert_conn = cma_id->qp->qp_context;
776 		isert_conn->dev_removed = true;
777 		isert_disconnected_handler(cma_id, event->event);
778 		wait_event_interruptible(isert_conn->rem_wait,
779 					 isert_conn->state == ISER_CONN_DOWN);
780 		kfree(isert_conn);
781 		/*
782 		 * return non-zero from the callback to destroy
783 		 * the rdma cm id
784 		 */
785 		return 1;
786 	case RDMA_CM_EVENT_REJECTED:
787 		isert_info("Connection rejected: %s\n",
788 			   rdma_reject_msg(cma_id, event->status));
789 		/* fall through */
790 	case RDMA_CM_EVENT_UNREACHABLE:
791 	case RDMA_CM_EVENT_CONNECT_ERROR:
792 		ret = isert_connect_error(cma_id);
793 		break;
794 	default:
795 		isert_err("Unhandled RDMA CMA event: %d\n", event->event);
796 		break;
797 	}
798 
799 	return ret;
800 }
801 
802 static int
803 isert_post_recvm(struct isert_conn *isert_conn, u32 count)
804 {
805 	struct ib_recv_wr *rx_wr;
806 	int i, ret;
807 	struct iser_rx_desc *rx_desc;
808 
809 	for (rx_wr = isert_conn->rx_wr, i = 0; i < count; i++, rx_wr++) {
810 		rx_desc = &isert_conn->rx_descs[i];
811 
812 		rx_wr->wr_cqe = &rx_desc->rx_cqe;
813 		rx_wr->sg_list = &rx_desc->rx_sg;
814 		rx_wr->num_sge = 1;
815 		rx_wr->next = rx_wr + 1;
816 		rx_desc->in_use = false;
817 	}
818 	rx_wr--;
819 	rx_wr->next = NULL; /* mark end of work requests list */
820 
821 	ret = ib_post_recv(isert_conn->qp, isert_conn->rx_wr, NULL);
822 	if (ret)
823 		isert_err("ib_post_recv() failed with ret: %d\n", ret);
824 
825 	return ret;
826 }
827 
828 static int
829 isert_post_recv(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc)
830 {
831 	struct ib_recv_wr rx_wr;
832 	int ret;
833 
834 	if (!rx_desc->in_use) {
835 		/*
836 		 * if the descriptor is not in-use we already reposted it
837 		 * for recv, so just silently return
838 		 */
839 		return 0;
840 	}
841 
842 	rx_desc->in_use = false;
843 	rx_wr.wr_cqe = &rx_desc->rx_cqe;
844 	rx_wr.sg_list = &rx_desc->rx_sg;
845 	rx_wr.num_sge = 1;
846 	rx_wr.next = NULL;
847 
848 	ret = ib_post_recv(isert_conn->qp, &rx_wr, NULL);
849 	if (ret)
850 		isert_err("ib_post_recv() failed with ret: %d\n", ret);
851 
852 	return ret;
853 }
854 
855 static int
856 isert_login_post_send(struct isert_conn *isert_conn, struct iser_tx_desc *tx_desc)
857 {
858 	struct ib_device *ib_dev = isert_conn->cm_id->device;
859 	struct ib_send_wr send_wr;
860 	int ret;
861 
862 	ib_dma_sync_single_for_device(ib_dev, tx_desc->dma_addr,
863 				      ISER_HEADERS_LEN, DMA_TO_DEVICE);
864 
865 	tx_desc->tx_cqe.done = isert_login_send_done;
866 
867 	send_wr.next	= NULL;
868 	send_wr.wr_cqe	= &tx_desc->tx_cqe;
869 	send_wr.sg_list	= tx_desc->tx_sg;
870 	send_wr.num_sge	= tx_desc->num_sge;
871 	send_wr.opcode	= IB_WR_SEND;
872 	send_wr.send_flags = IB_SEND_SIGNALED;
873 
874 	ret = ib_post_send(isert_conn->qp, &send_wr, NULL);
875 	if (ret)
876 		isert_err("ib_post_send() failed, ret: %d\n", ret);
877 
878 	return ret;
879 }
880 
881 static void
882 __isert_create_send_desc(struct isert_device *device,
883 			 struct iser_tx_desc *tx_desc)
884 {
885 
886 	memset(&tx_desc->iser_header, 0, sizeof(struct iser_ctrl));
887 	tx_desc->iser_header.flags = ISCSI_CTRL;
888 
889 	tx_desc->num_sge = 1;
890 
891 	if (tx_desc->tx_sg[0].lkey != device->pd->local_dma_lkey) {
892 		tx_desc->tx_sg[0].lkey = device->pd->local_dma_lkey;
893 		isert_dbg("tx_desc %p lkey mismatch, fixing\n", tx_desc);
894 	}
895 }
896 
897 static void
898 isert_create_send_desc(struct isert_conn *isert_conn,
899 		       struct isert_cmd *isert_cmd,
900 		       struct iser_tx_desc *tx_desc)
901 {
902 	struct isert_device *device = isert_conn->device;
903 	struct ib_device *ib_dev = device->ib_device;
904 
905 	ib_dma_sync_single_for_cpu(ib_dev, tx_desc->dma_addr,
906 				   ISER_HEADERS_LEN, DMA_TO_DEVICE);
907 
908 	__isert_create_send_desc(device, tx_desc);
909 }
910 
911 static int
912 isert_init_tx_hdrs(struct isert_conn *isert_conn,
913 		   struct iser_tx_desc *tx_desc)
914 {
915 	struct isert_device *device = isert_conn->device;
916 	struct ib_device *ib_dev = device->ib_device;
917 	u64 dma_addr;
918 
919 	dma_addr = ib_dma_map_single(ib_dev, (void *)tx_desc,
920 			ISER_HEADERS_LEN, DMA_TO_DEVICE);
921 	if (ib_dma_mapping_error(ib_dev, dma_addr)) {
922 		isert_err("ib_dma_mapping_error() failed\n");
923 		return -ENOMEM;
924 	}
925 
926 	tx_desc->dma_addr = dma_addr;
927 	tx_desc->tx_sg[0].addr	= tx_desc->dma_addr;
928 	tx_desc->tx_sg[0].length = ISER_HEADERS_LEN;
929 	tx_desc->tx_sg[0].lkey = device->pd->local_dma_lkey;
930 
931 	isert_dbg("Setup tx_sg[0].addr: 0x%llx length: %u lkey: 0x%x\n",
932 		  tx_desc->tx_sg[0].addr, tx_desc->tx_sg[0].length,
933 		  tx_desc->tx_sg[0].lkey);
934 
935 	return 0;
936 }
937 
938 static void
939 isert_init_send_wr(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
940 		   struct ib_send_wr *send_wr)
941 {
942 	struct iser_tx_desc *tx_desc = &isert_cmd->tx_desc;
943 
944 	tx_desc->tx_cqe.done = isert_send_done;
945 	send_wr->wr_cqe = &tx_desc->tx_cqe;
946 
947 	if (isert_conn->snd_w_inv && isert_cmd->inv_rkey) {
948 		send_wr->opcode  = IB_WR_SEND_WITH_INV;
949 		send_wr->ex.invalidate_rkey = isert_cmd->inv_rkey;
950 	} else {
951 		send_wr->opcode = IB_WR_SEND;
952 	}
953 
954 	send_wr->sg_list = &tx_desc->tx_sg[0];
955 	send_wr->num_sge = isert_cmd->tx_desc.num_sge;
956 	send_wr->send_flags = IB_SEND_SIGNALED;
957 }
958 
959 static int
960 isert_login_post_recv(struct isert_conn *isert_conn)
961 {
962 	struct ib_recv_wr rx_wr;
963 	struct ib_sge sge;
964 	int ret;
965 
966 	memset(&sge, 0, sizeof(struct ib_sge));
967 	sge.addr = isert_conn->login_req_dma;
968 	sge.length = ISER_RX_PAYLOAD_SIZE;
969 	sge.lkey = isert_conn->device->pd->local_dma_lkey;
970 
971 	isert_dbg("Setup sge: addr: %llx length: %d 0x%08x\n",
972 		sge.addr, sge.length, sge.lkey);
973 
974 	isert_conn->login_req_buf->rx_cqe.done = isert_login_recv_done;
975 
976 	memset(&rx_wr, 0, sizeof(struct ib_recv_wr));
977 	rx_wr.wr_cqe = &isert_conn->login_req_buf->rx_cqe;
978 	rx_wr.sg_list = &sge;
979 	rx_wr.num_sge = 1;
980 
981 	ret = ib_post_recv(isert_conn->qp, &rx_wr, NULL);
982 	if (ret)
983 		isert_err("ib_post_recv() failed: %d\n", ret);
984 
985 	return ret;
986 }
987 
988 static int
989 isert_put_login_tx(struct iscsi_conn *conn, struct iscsi_login *login,
990 		   u32 length)
991 {
992 	struct isert_conn *isert_conn = conn->context;
993 	struct isert_device *device = isert_conn->device;
994 	struct ib_device *ib_dev = device->ib_device;
995 	struct iser_tx_desc *tx_desc = &isert_conn->login_tx_desc;
996 	int ret;
997 
998 	__isert_create_send_desc(device, tx_desc);
999 
1000 	memcpy(&tx_desc->iscsi_header, &login->rsp[0],
1001 	       sizeof(struct iscsi_hdr));
1002 
1003 	isert_init_tx_hdrs(isert_conn, tx_desc);
1004 
1005 	if (length > 0) {
1006 		struct ib_sge *tx_dsg = &tx_desc->tx_sg[1];
1007 
1008 		ib_dma_sync_single_for_cpu(ib_dev, isert_conn->login_rsp_dma,
1009 					   length, DMA_TO_DEVICE);
1010 
1011 		memcpy(isert_conn->login_rsp_buf, login->rsp_buf, length);
1012 
1013 		ib_dma_sync_single_for_device(ib_dev, isert_conn->login_rsp_dma,
1014 					      length, DMA_TO_DEVICE);
1015 
1016 		tx_dsg->addr	= isert_conn->login_rsp_dma;
1017 		tx_dsg->length	= length;
1018 		tx_dsg->lkey	= isert_conn->device->pd->local_dma_lkey;
1019 		tx_desc->num_sge = 2;
1020 	}
1021 	if (!login->login_failed) {
1022 		if (login->login_complete) {
1023 			ret = isert_alloc_rx_descriptors(isert_conn);
1024 			if (ret)
1025 				return ret;
1026 
1027 			ret = isert_post_recvm(isert_conn,
1028 					       ISERT_QP_MAX_RECV_DTOS);
1029 			if (ret)
1030 				return ret;
1031 
1032 			/* Now we are in FULL_FEATURE phase */
1033 			mutex_lock(&isert_conn->mutex);
1034 			isert_conn->state = ISER_CONN_FULL_FEATURE;
1035 			mutex_unlock(&isert_conn->mutex);
1036 			goto post_send;
1037 		}
1038 
1039 		ret = isert_login_post_recv(isert_conn);
1040 		if (ret)
1041 			return ret;
1042 	}
1043 post_send:
1044 	ret = isert_login_post_send(isert_conn, tx_desc);
1045 	if (ret)
1046 		return ret;
1047 
1048 	return 0;
1049 }
1050 
1051 static void
1052 isert_rx_login_req(struct isert_conn *isert_conn)
1053 {
1054 	struct iser_rx_desc *rx_desc = isert_conn->login_req_buf;
1055 	int rx_buflen = isert_conn->login_req_len;
1056 	struct iscsi_conn *conn = isert_conn->conn;
1057 	struct iscsi_login *login = conn->conn_login;
1058 	int size;
1059 
1060 	isert_info("conn %p\n", isert_conn);
1061 
1062 	WARN_ON_ONCE(!login);
1063 
1064 	if (login->first_request) {
1065 		struct iscsi_login_req *login_req =
1066 			(struct iscsi_login_req *)&rx_desc->iscsi_header;
1067 		/*
1068 		 * Setup the initial iscsi_login values from the leading
1069 		 * login request PDU.
1070 		 */
1071 		login->leading_connection = (!login_req->tsih) ? 1 : 0;
1072 		login->current_stage =
1073 			(login_req->flags & ISCSI_FLAG_LOGIN_CURRENT_STAGE_MASK)
1074 			 >> 2;
1075 		login->version_min	= login_req->min_version;
1076 		login->version_max	= login_req->max_version;
1077 		memcpy(login->isid, login_req->isid, 6);
1078 		login->cmd_sn		= be32_to_cpu(login_req->cmdsn);
1079 		login->init_task_tag	= login_req->itt;
1080 		login->initial_exp_statsn = be32_to_cpu(login_req->exp_statsn);
1081 		login->cid		= be16_to_cpu(login_req->cid);
1082 		login->tsih		= be16_to_cpu(login_req->tsih);
1083 	}
1084 
1085 	memcpy(&login->req[0], (void *)&rx_desc->iscsi_header, ISCSI_HDR_LEN);
1086 
1087 	size = min(rx_buflen, MAX_KEY_VALUE_PAIRS);
1088 	isert_dbg("Using login payload size: %d, rx_buflen: %d "
1089 		  "MAX_KEY_VALUE_PAIRS: %d\n", size, rx_buflen,
1090 		  MAX_KEY_VALUE_PAIRS);
1091 	memcpy(login->req_buf, &rx_desc->data[0], size);
1092 
1093 	if (login->first_request) {
1094 		complete(&isert_conn->login_comp);
1095 		return;
1096 	}
1097 	schedule_delayed_work(&conn->login_work, 0);
1098 }
1099 
1100 static struct iscsi_cmd
1101 *isert_allocate_cmd(struct iscsi_conn *conn, struct iser_rx_desc *rx_desc)
1102 {
1103 	struct isert_conn *isert_conn = conn->context;
1104 	struct isert_cmd *isert_cmd;
1105 	struct iscsi_cmd *cmd;
1106 
1107 	cmd = iscsit_allocate_cmd(conn, TASK_INTERRUPTIBLE);
1108 	if (!cmd) {
1109 		isert_err("Unable to allocate iscsi_cmd + isert_cmd\n");
1110 		return NULL;
1111 	}
1112 	isert_cmd = iscsit_priv_cmd(cmd);
1113 	isert_cmd->conn = isert_conn;
1114 	isert_cmd->iscsi_cmd = cmd;
1115 	isert_cmd->rx_desc = rx_desc;
1116 
1117 	return cmd;
1118 }
1119 
1120 static int
1121 isert_handle_scsi_cmd(struct isert_conn *isert_conn,
1122 		      struct isert_cmd *isert_cmd, struct iscsi_cmd *cmd,
1123 		      struct iser_rx_desc *rx_desc, unsigned char *buf)
1124 {
1125 	struct iscsi_conn *conn = isert_conn->conn;
1126 	struct iscsi_scsi_req *hdr = (struct iscsi_scsi_req *)buf;
1127 	int imm_data, imm_data_len, unsol_data, sg_nents, rc;
1128 	bool dump_payload = false;
1129 	unsigned int data_len;
1130 
1131 	rc = iscsit_setup_scsi_cmd(conn, cmd, buf);
1132 	if (rc < 0)
1133 		return rc;
1134 
1135 	imm_data = cmd->immediate_data;
1136 	imm_data_len = cmd->first_burst_len;
1137 	unsol_data = cmd->unsolicited_data;
1138 	data_len = cmd->se_cmd.data_length;
1139 
1140 	if (imm_data && imm_data_len == data_len)
1141 		cmd->se_cmd.se_cmd_flags |= SCF_PASSTHROUGH_SG_TO_MEM_NOALLOC;
1142 	rc = iscsit_process_scsi_cmd(conn, cmd, hdr);
1143 	if (rc < 0) {
1144 		return 0;
1145 	} else if (rc > 0) {
1146 		dump_payload = true;
1147 		goto sequence_cmd;
1148 	}
1149 
1150 	if (!imm_data)
1151 		return 0;
1152 
1153 	if (imm_data_len != data_len) {
1154 		sg_nents = max(1UL, DIV_ROUND_UP(imm_data_len, PAGE_SIZE));
1155 		sg_copy_from_buffer(cmd->se_cmd.t_data_sg, sg_nents,
1156 				    &rx_desc->data[0], imm_data_len);
1157 		isert_dbg("Copy Immediate sg_nents: %u imm_data_len: %d\n",
1158 			  sg_nents, imm_data_len);
1159 	} else {
1160 		sg_init_table(&isert_cmd->sg, 1);
1161 		cmd->se_cmd.t_data_sg = &isert_cmd->sg;
1162 		cmd->se_cmd.t_data_nents = 1;
1163 		sg_set_buf(&isert_cmd->sg, &rx_desc->data[0], imm_data_len);
1164 		isert_dbg("Transfer Immediate imm_data_len: %d\n",
1165 			  imm_data_len);
1166 	}
1167 
1168 	cmd->write_data_done += imm_data_len;
1169 
1170 	if (cmd->write_data_done == cmd->se_cmd.data_length) {
1171 		spin_lock_bh(&cmd->istate_lock);
1172 		cmd->cmd_flags |= ICF_GOT_LAST_DATAOUT;
1173 		cmd->i_state = ISTATE_RECEIVED_LAST_DATAOUT;
1174 		spin_unlock_bh(&cmd->istate_lock);
1175 	}
1176 
1177 sequence_cmd:
1178 	rc = iscsit_sequence_cmd(conn, cmd, buf, hdr->cmdsn);
1179 
1180 	if (!rc && dump_payload == false && unsol_data)
1181 		iscsit_set_unsolicited_dataout(cmd);
1182 	else if (dump_payload && imm_data)
1183 		target_put_sess_cmd(&cmd->se_cmd);
1184 
1185 	return 0;
1186 }
1187 
1188 static int
1189 isert_handle_iscsi_dataout(struct isert_conn *isert_conn,
1190 			   struct iser_rx_desc *rx_desc, unsigned char *buf)
1191 {
1192 	struct scatterlist *sg_start;
1193 	struct iscsi_conn *conn = isert_conn->conn;
1194 	struct iscsi_cmd *cmd = NULL;
1195 	struct iscsi_data *hdr = (struct iscsi_data *)buf;
1196 	u32 unsol_data_len = ntoh24(hdr->dlength);
1197 	int rc, sg_nents, sg_off, page_off;
1198 
1199 	rc = iscsit_check_dataout_hdr(conn, buf, &cmd);
1200 	if (rc < 0)
1201 		return rc;
1202 	else if (!cmd)
1203 		return 0;
1204 	/*
1205 	 * FIXME: Unexpected unsolicited_data out
1206 	 */
1207 	if (!cmd->unsolicited_data) {
1208 		isert_err("Received unexpected solicited data payload\n");
1209 		dump_stack();
1210 		return -1;
1211 	}
1212 
1213 	isert_dbg("Unsolicited DataOut unsol_data_len: %u, "
1214 		  "write_data_done: %u, data_length: %u\n",
1215 		  unsol_data_len,  cmd->write_data_done,
1216 		  cmd->se_cmd.data_length);
1217 
1218 	sg_off = cmd->write_data_done / PAGE_SIZE;
1219 	sg_start = &cmd->se_cmd.t_data_sg[sg_off];
1220 	sg_nents = max(1UL, DIV_ROUND_UP(unsol_data_len, PAGE_SIZE));
1221 	page_off = cmd->write_data_done % PAGE_SIZE;
1222 	/*
1223 	 * FIXME: Non page-aligned unsolicited_data out
1224 	 */
1225 	if (page_off) {
1226 		isert_err("unexpected non-page aligned data payload\n");
1227 		dump_stack();
1228 		return -1;
1229 	}
1230 	isert_dbg("Copying DataOut: sg_start: %p, sg_off: %u "
1231 		  "sg_nents: %u from %p %u\n", sg_start, sg_off,
1232 		  sg_nents, &rx_desc->data[0], unsol_data_len);
1233 
1234 	sg_copy_from_buffer(sg_start, sg_nents, &rx_desc->data[0],
1235 			    unsol_data_len);
1236 
1237 	rc = iscsit_check_dataout_payload(cmd, hdr, false);
1238 	if (rc < 0)
1239 		return rc;
1240 
1241 	/*
1242 	 * multiple data-outs on the same command can arrive -
1243 	 * so post the buffer before hand
1244 	 */
1245 	rc = isert_post_recv(isert_conn, rx_desc);
1246 	if (rc) {
1247 		isert_err("ib_post_recv failed with %d\n", rc);
1248 		return rc;
1249 	}
1250 	return 0;
1251 }
1252 
1253 static int
1254 isert_handle_nop_out(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
1255 		     struct iscsi_cmd *cmd, struct iser_rx_desc *rx_desc,
1256 		     unsigned char *buf)
1257 {
1258 	struct iscsi_conn *conn = isert_conn->conn;
1259 	struct iscsi_nopout *hdr = (struct iscsi_nopout *)buf;
1260 	int rc;
1261 
1262 	rc = iscsit_setup_nop_out(conn, cmd, hdr);
1263 	if (rc < 0)
1264 		return rc;
1265 	/*
1266 	 * FIXME: Add support for NOPOUT payload using unsolicited RDMA payload
1267 	 */
1268 
1269 	return iscsit_process_nop_out(conn, cmd, hdr);
1270 }
1271 
1272 static int
1273 isert_handle_text_cmd(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
1274 		      struct iscsi_cmd *cmd, struct iser_rx_desc *rx_desc,
1275 		      struct iscsi_text *hdr)
1276 {
1277 	struct iscsi_conn *conn = isert_conn->conn;
1278 	u32 payload_length = ntoh24(hdr->dlength);
1279 	int rc;
1280 	unsigned char *text_in = NULL;
1281 
1282 	rc = iscsit_setup_text_cmd(conn, cmd, hdr);
1283 	if (rc < 0)
1284 		return rc;
1285 
1286 	if (payload_length) {
1287 		text_in = kzalloc(payload_length, GFP_KERNEL);
1288 		if (!text_in)
1289 			return -ENOMEM;
1290 	}
1291 	cmd->text_in_ptr = text_in;
1292 
1293 	memcpy(cmd->text_in_ptr, &rx_desc->data[0], payload_length);
1294 
1295 	return iscsit_process_text_cmd(conn, cmd, hdr);
1296 }
1297 
1298 static int
1299 isert_rx_opcode(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc,
1300 		uint32_t read_stag, uint64_t read_va,
1301 		uint32_t write_stag, uint64_t write_va)
1302 {
1303 	struct iscsi_hdr *hdr = &rx_desc->iscsi_header;
1304 	struct iscsi_conn *conn = isert_conn->conn;
1305 	struct iscsi_cmd *cmd;
1306 	struct isert_cmd *isert_cmd;
1307 	int ret = -EINVAL;
1308 	u8 opcode = (hdr->opcode & ISCSI_OPCODE_MASK);
1309 
1310 	if (conn->sess->sess_ops->SessionType &&
1311 	   (!(opcode & ISCSI_OP_TEXT) || !(opcode & ISCSI_OP_LOGOUT))) {
1312 		isert_err("Got illegal opcode: 0x%02x in SessionType=Discovery,"
1313 			  " ignoring\n", opcode);
1314 		return 0;
1315 	}
1316 
1317 	switch (opcode) {
1318 	case ISCSI_OP_SCSI_CMD:
1319 		cmd = isert_allocate_cmd(conn, rx_desc);
1320 		if (!cmd)
1321 			break;
1322 
1323 		isert_cmd = iscsit_priv_cmd(cmd);
1324 		isert_cmd->read_stag = read_stag;
1325 		isert_cmd->read_va = read_va;
1326 		isert_cmd->write_stag = write_stag;
1327 		isert_cmd->write_va = write_va;
1328 		isert_cmd->inv_rkey = read_stag ? read_stag : write_stag;
1329 
1330 		ret = isert_handle_scsi_cmd(isert_conn, isert_cmd, cmd,
1331 					rx_desc, (unsigned char *)hdr);
1332 		break;
1333 	case ISCSI_OP_NOOP_OUT:
1334 		cmd = isert_allocate_cmd(conn, rx_desc);
1335 		if (!cmd)
1336 			break;
1337 
1338 		isert_cmd = iscsit_priv_cmd(cmd);
1339 		ret = isert_handle_nop_out(isert_conn, isert_cmd, cmd,
1340 					   rx_desc, (unsigned char *)hdr);
1341 		break;
1342 	case ISCSI_OP_SCSI_DATA_OUT:
1343 		ret = isert_handle_iscsi_dataout(isert_conn, rx_desc,
1344 						(unsigned char *)hdr);
1345 		break;
1346 	case ISCSI_OP_SCSI_TMFUNC:
1347 		cmd = isert_allocate_cmd(conn, rx_desc);
1348 		if (!cmd)
1349 			break;
1350 
1351 		ret = iscsit_handle_task_mgt_cmd(conn, cmd,
1352 						(unsigned char *)hdr);
1353 		break;
1354 	case ISCSI_OP_LOGOUT:
1355 		cmd = isert_allocate_cmd(conn, rx_desc);
1356 		if (!cmd)
1357 			break;
1358 
1359 		ret = iscsit_handle_logout_cmd(conn, cmd, (unsigned char *)hdr);
1360 		break;
1361 	case ISCSI_OP_TEXT:
1362 		if (be32_to_cpu(hdr->ttt) != 0xFFFFFFFF)
1363 			cmd = iscsit_find_cmd_from_itt(conn, hdr->itt);
1364 		else
1365 			cmd = isert_allocate_cmd(conn, rx_desc);
1366 
1367 		if (!cmd)
1368 			break;
1369 
1370 		isert_cmd = iscsit_priv_cmd(cmd);
1371 		ret = isert_handle_text_cmd(isert_conn, isert_cmd, cmd,
1372 					    rx_desc, (struct iscsi_text *)hdr);
1373 		break;
1374 	default:
1375 		isert_err("Got unknown iSCSI OpCode: 0x%02x\n", opcode);
1376 		dump_stack();
1377 		break;
1378 	}
1379 
1380 	return ret;
1381 }
1382 
1383 static void
1384 isert_print_wc(struct ib_wc *wc, const char *type)
1385 {
1386 	if (wc->status != IB_WC_WR_FLUSH_ERR)
1387 		isert_err("%s failure: %s (%d) vend_err %x\n", type,
1388 			  ib_wc_status_msg(wc->status), wc->status,
1389 			  wc->vendor_err);
1390 	else
1391 		isert_dbg("%s failure: %s (%d)\n", type,
1392 			  ib_wc_status_msg(wc->status), wc->status);
1393 }
1394 
1395 static void
1396 isert_recv_done(struct ib_cq *cq, struct ib_wc *wc)
1397 {
1398 	struct isert_conn *isert_conn = wc->qp->qp_context;
1399 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1400 	struct iser_rx_desc *rx_desc = cqe_to_rx_desc(wc->wr_cqe);
1401 	struct iscsi_hdr *hdr = &rx_desc->iscsi_header;
1402 	struct iser_ctrl *iser_ctrl = &rx_desc->iser_header;
1403 	uint64_t read_va = 0, write_va = 0;
1404 	uint32_t read_stag = 0, write_stag = 0;
1405 
1406 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1407 		isert_print_wc(wc, "recv");
1408 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1409 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1410 		return;
1411 	}
1412 
1413 	rx_desc->in_use = true;
1414 
1415 	ib_dma_sync_single_for_cpu(ib_dev, rx_desc->dma_addr,
1416 			ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
1417 
1418 	isert_dbg("DMA: 0x%llx, iSCSI opcode: 0x%02x, ITT: 0x%08x, flags: 0x%02x dlen: %d\n",
1419 		 rx_desc->dma_addr, hdr->opcode, hdr->itt, hdr->flags,
1420 		 (int)(wc->byte_len - ISER_HEADERS_LEN));
1421 
1422 	switch (iser_ctrl->flags & 0xF0) {
1423 	case ISCSI_CTRL:
1424 		if (iser_ctrl->flags & ISER_RSV) {
1425 			read_stag = be32_to_cpu(iser_ctrl->read_stag);
1426 			read_va = be64_to_cpu(iser_ctrl->read_va);
1427 			isert_dbg("ISER_RSV: read_stag: 0x%x read_va: 0x%llx\n",
1428 				  read_stag, (unsigned long long)read_va);
1429 		}
1430 		if (iser_ctrl->flags & ISER_WSV) {
1431 			write_stag = be32_to_cpu(iser_ctrl->write_stag);
1432 			write_va = be64_to_cpu(iser_ctrl->write_va);
1433 			isert_dbg("ISER_WSV: write_stag: 0x%x write_va: 0x%llx\n",
1434 				  write_stag, (unsigned long long)write_va);
1435 		}
1436 
1437 		isert_dbg("ISER ISCSI_CTRL PDU\n");
1438 		break;
1439 	case ISER_HELLO:
1440 		isert_err("iSER Hello message\n");
1441 		break;
1442 	default:
1443 		isert_warn("Unknown iSER hdr flags: 0x%02x\n", iser_ctrl->flags);
1444 		break;
1445 	}
1446 
1447 	isert_rx_opcode(isert_conn, rx_desc,
1448 			read_stag, read_va, write_stag, write_va);
1449 
1450 	ib_dma_sync_single_for_device(ib_dev, rx_desc->dma_addr,
1451 			ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
1452 }
1453 
1454 static void
1455 isert_login_recv_done(struct ib_cq *cq, struct ib_wc *wc)
1456 {
1457 	struct isert_conn *isert_conn = wc->qp->qp_context;
1458 	struct ib_device *ib_dev = isert_conn->device->ib_device;
1459 
1460 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1461 		isert_print_wc(wc, "login recv");
1462 		return;
1463 	}
1464 
1465 	ib_dma_sync_single_for_cpu(ib_dev, isert_conn->login_req_dma,
1466 			ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
1467 
1468 	isert_conn->login_req_len = wc->byte_len - ISER_HEADERS_LEN;
1469 
1470 	if (isert_conn->conn) {
1471 		struct iscsi_login *login = isert_conn->conn->conn_login;
1472 
1473 		if (login && !login->first_request)
1474 			isert_rx_login_req(isert_conn);
1475 	}
1476 
1477 	mutex_lock(&isert_conn->mutex);
1478 	complete(&isert_conn->login_req_comp);
1479 	mutex_unlock(&isert_conn->mutex);
1480 
1481 	ib_dma_sync_single_for_device(ib_dev, isert_conn->login_req_dma,
1482 				ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
1483 }
1484 
1485 static void
1486 isert_rdma_rw_ctx_destroy(struct isert_cmd *cmd, struct isert_conn *conn)
1487 {
1488 	struct se_cmd *se_cmd = &cmd->iscsi_cmd->se_cmd;
1489 	enum dma_data_direction dir = target_reverse_dma_direction(se_cmd);
1490 
1491 	if (!cmd->rw.nr_ops)
1492 		return;
1493 
1494 	if (isert_prot_cmd(conn, se_cmd)) {
1495 		rdma_rw_ctx_destroy_signature(&cmd->rw, conn->qp,
1496 				conn->cm_id->port_num, se_cmd->t_data_sg,
1497 				se_cmd->t_data_nents, se_cmd->t_prot_sg,
1498 				se_cmd->t_prot_nents, dir);
1499 	} else {
1500 		rdma_rw_ctx_destroy(&cmd->rw, conn->qp, conn->cm_id->port_num,
1501 				se_cmd->t_data_sg, se_cmd->t_data_nents, dir);
1502 	}
1503 
1504 	cmd->rw.nr_ops = 0;
1505 }
1506 
1507 static void
1508 isert_put_cmd(struct isert_cmd *isert_cmd, bool comp_err)
1509 {
1510 	struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd;
1511 	struct isert_conn *isert_conn = isert_cmd->conn;
1512 	struct iscsi_conn *conn = isert_conn->conn;
1513 	struct iscsi_text_rsp *hdr;
1514 
1515 	isert_dbg("Cmd %p\n", isert_cmd);
1516 
1517 	switch (cmd->iscsi_opcode) {
1518 	case ISCSI_OP_SCSI_CMD:
1519 		spin_lock_bh(&conn->cmd_lock);
1520 		if (!list_empty(&cmd->i_conn_node))
1521 			list_del_init(&cmd->i_conn_node);
1522 		spin_unlock_bh(&conn->cmd_lock);
1523 
1524 		if (cmd->data_direction == DMA_TO_DEVICE) {
1525 			iscsit_stop_dataout_timer(cmd);
1526 			/*
1527 			 * Check for special case during comp_err where
1528 			 * WRITE_PENDING has been handed off from core,
1529 			 * but requires an extra target_put_sess_cmd()
1530 			 * before transport_generic_free_cmd() below.
1531 			 */
1532 			if (comp_err &&
1533 			    cmd->se_cmd.t_state == TRANSPORT_WRITE_PENDING) {
1534 				struct se_cmd *se_cmd = &cmd->se_cmd;
1535 
1536 				target_put_sess_cmd(se_cmd);
1537 			}
1538 		}
1539 
1540 		isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1541 		transport_generic_free_cmd(&cmd->se_cmd, 0);
1542 		break;
1543 	case ISCSI_OP_SCSI_TMFUNC:
1544 		spin_lock_bh(&conn->cmd_lock);
1545 		if (!list_empty(&cmd->i_conn_node))
1546 			list_del_init(&cmd->i_conn_node);
1547 		spin_unlock_bh(&conn->cmd_lock);
1548 
1549 		transport_generic_free_cmd(&cmd->se_cmd, 0);
1550 		break;
1551 	case ISCSI_OP_REJECT:
1552 	case ISCSI_OP_NOOP_OUT:
1553 	case ISCSI_OP_TEXT:
1554 		hdr = (struct iscsi_text_rsp *)&isert_cmd->tx_desc.iscsi_header;
1555 		/* If the continue bit is on, keep the command alive */
1556 		if (hdr->flags & ISCSI_FLAG_TEXT_CONTINUE)
1557 			break;
1558 
1559 		spin_lock_bh(&conn->cmd_lock);
1560 		if (!list_empty(&cmd->i_conn_node))
1561 			list_del_init(&cmd->i_conn_node);
1562 		spin_unlock_bh(&conn->cmd_lock);
1563 
1564 		/*
1565 		 * Handle special case for REJECT when iscsi_add_reject*() has
1566 		 * overwritten the original iscsi_opcode assignment, and the
1567 		 * associated cmd->se_cmd needs to be released.
1568 		 */
1569 		if (cmd->se_cmd.se_tfo != NULL) {
1570 			isert_dbg("Calling transport_generic_free_cmd for 0x%02x\n",
1571 				 cmd->iscsi_opcode);
1572 			transport_generic_free_cmd(&cmd->se_cmd, 0);
1573 			break;
1574 		}
1575 		/* fall through */
1576 	default:
1577 		iscsit_release_cmd(cmd);
1578 		break;
1579 	}
1580 }
1581 
1582 static void
1583 isert_unmap_tx_desc(struct iser_tx_desc *tx_desc, struct ib_device *ib_dev)
1584 {
1585 	if (tx_desc->dma_addr != 0) {
1586 		isert_dbg("unmap single for tx_desc->dma_addr\n");
1587 		ib_dma_unmap_single(ib_dev, tx_desc->dma_addr,
1588 				    ISER_HEADERS_LEN, DMA_TO_DEVICE);
1589 		tx_desc->dma_addr = 0;
1590 	}
1591 }
1592 
1593 static void
1594 isert_completion_put(struct iser_tx_desc *tx_desc, struct isert_cmd *isert_cmd,
1595 		     struct ib_device *ib_dev, bool comp_err)
1596 {
1597 	if (isert_cmd->pdu_buf_dma != 0) {
1598 		isert_dbg("unmap single for isert_cmd->pdu_buf_dma\n");
1599 		ib_dma_unmap_single(ib_dev, isert_cmd->pdu_buf_dma,
1600 				    isert_cmd->pdu_buf_len, DMA_TO_DEVICE);
1601 		isert_cmd->pdu_buf_dma = 0;
1602 	}
1603 
1604 	isert_unmap_tx_desc(tx_desc, ib_dev);
1605 	isert_put_cmd(isert_cmd, comp_err);
1606 }
1607 
1608 static int
1609 isert_check_pi_status(struct se_cmd *se_cmd, struct ib_mr *sig_mr)
1610 {
1611 	struct ib_mr_status mr_status;
1612 	int ret;
1613 
1614 	ret = ib_check_mr_status(sig_mr, IB_MR_CHECK_SIG_STATUS, &mr_status);
1615 	if (ret) {
1616 		isert_err("ib_check_mr_status failed, ret %d\n", ret);
1617 		goto fail_mr_status;
1618 	}
1619 
1620 	if (mr_status.fail_status & IB_MR_CHECK_SIG_STATUS) {
1621 		u64 sec_offset_err;
1622 		u32 block_size = se_cmd->se_dev->dev_attrib.block_size + 8;
1623 
1624 		switch (mr_status.sig_err.err_type) {
1625 		case IB_SIG_BAD_GUARD:
1626 			se_cmd->pi_err = TCM_LOGICAL_BLOCK_GUARD_CHECK_FAILED;
1627 			break;
1628 		case IB_SIG_BAD_REFTAG:
1629 			se_cmd->pi_err = TCM_LOGICAL_BLOCK_REF_TAG_CHECK_FAILED;
1630 			break;
1631 		case IB_SIG_BAD_APPTAG:
1632 			se_cmd->pi_err = TCM_LOGICAL_BLOCK_APP_TAG_CHECK_FAILED;
1633 			break;
1634 		}
1635 		sec_offset_err = mr_status.sig_err.sig_err_offset;
1636 		do_div(sec_offset_err, block_size);
1637 		se_cmd->bad_sector = sec_offset_err + se_cmd->t_task_lba;
1638 
1639 		isert_err("PI error found type %d at sector 0x%llx "
1640 			  "expected 0x%x vs actual 0x%x\n",
1641 			  mr_status.sig_err.err_type,
1642 			  (unsigned long long)se_cmd->bad_sector,
1643 			  mr_status.sig_err.expected,
1644 			  mr_status.sig_err.actual);
1645 		ret = 1;
1646 	}
1647 
1648 fail_mr_status:
1649 	return ret;
1650 }
1651 
1652 static void
1653 isert_rdma_write_done(struct ib_cq *cq, struct ib_wc *wc)
1654 {
1655 	struct isert_conn *isert_conn = wc->qp->qp_context;
1656 	struct isert_device *device = isert_conn->device;
1657 	struct iser_tx_desc *desc = cqe_to_tx_desc(wc->wr_cqe);
1658 	struct isert_cmd *isert_cmd = tx_desc_to_cmd(desc);
1659 	struct se_cmd *cmd = &isert_cmd->iscsi_cmd->se_cmd;
1660 	int ret = 0;
1661 
1662 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1663 		isert_print_wc(wc, "rdma write");
1664 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1665 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1666 		isert_completion_put(desc, isert_cmd, device->ib_device, true);
1667 		return;
1668 	}
1669 
1670 	isert_dbg("Cmd %p\n", isert_cmd);
1671 
1672 	ret = isert_check_pi_status(cmd, isert_cmd->rw.reg->mr);
1673 	isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1674 
1675 	if (ret) {
1676 		/*
1677 		 * transport_generic_request_failure() expects to have
1678 		 * plus two references to handle queue-full, so re-add
1679 		 * one here as target-core will have already dropped
1680 		 * it after the first isert_put_datain() callback.
1681 		 */
1682 		kref_get(&cmd->cmd_kref);
1683 		transport_generic_request_failure(cmd, cmd->pi_err);
1684 	} else {
1685 		/*
1686 		 * XXX: isert_put_response() failure is not retried.
1687 		 */
1688 		ret = isert_put_response(isert_conn->conn, isert_cmd->iscsi_cmd);
1689 		if (ret)
1690 			pr_warn_ratelimited("isert_put_response() ret: %d\n", ret);
1691 	}
1692 }
1693 
1694 static void
1695 isert_rdma_read_done(struct ib_cq *cq, struct ib_wc *wc)
1696 {
1697 	struct isert_conn *isert_conn = wc->qp->qp_context;
1698 	struct isert_device *device = isert_conn->device;
1699 	struct iser_tx_desc *desc = cqe_to_tx_desc(wc->wr_cqe);
1700 	struct isert_cmd *isert_cmd = tx_desc_to_cmd(desc);
1701 	struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd;
1702 	struct se_cmd *se_cmd = &cmd->se_cmd;
1703 	int ret = 0;
1704 
1705 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1706 		isert_print_wc(wc, "rdma read");
1707 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1708 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1709 		isert_completion_put(desc, isert_cmd, device->ib_device, true);
1710 		return;
1711 	}
1712 
1713 	isert_dbg("Cmd %p\n", isert_cmd);
1714 
1715 	iscsit_stop_dataout_timer(cmd);
1716 
1717 	if (isert_prot_cmd(isert_conn, se_cmd))
1718 		ret = isert_check_pi_status(se_cmd, isert_cmd->rw.reg->mr);
1719 	isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1720 	cmd->write_data_done = 0;
1721 
1722 	isert_dbg("Cmd: %p RDMA_READ comp calling execute_cmd\n", isert_cmd);
1723 	spin_lock_bh(&cmd->istate_lock);
1724 	cmd->cmd_flags |= ICF_GOT_LAST_DATAOUT;
1725 	cmd->i_state = ISTATE_RECEIVED_LAST_DATAOUT;
1726 	spin_unlock_bh(&cmd->istate_lock);
1727 
1728 	/*
1729 	 * transport_generic_request_failure() will drop the extra
1730 	 * se_cmd->cmd_kref reference after T10-PI error, and handle
1731 	 * any non-zero ->queue_status() callback error retries.
1732 	 */
1733 	if (ret)
1734 		transport_generic_request_failure(se_cmd, se_cmd->pi_err);
1735 	else
1736 		target_execute_cmd(se_cmd);
1737 }
1738 
1739 static void
1740 isert_do_control_comp(struct work_struct *work)
1741 {
1742 	struct isert_cmd *isert_cmd = container_of(work,
1743 			struct isert_cmd, comp_work);
1744 	struct isert_conn *isert_conn = isert_cmd->conn;
1745 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1746 	struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd;
1747 
1748 	isert_dbg("Cmd %p i_state %d\n", isert_cmd, cmd->i_state);
1749 
1750 	switch (cmd->i_state) {
1751 	case ISTATE_SEND_TASKMGTRSP:
1752 		iscsit_tmr_post_handler(cmd, cmd->conn);
1753 		/* fall through */
1754 	case ISTATE_SEND_REJECT:
1755 	case ISTATE_SEND_TEXTRSP:
1756 		cmd->i_state = ISTATE_SENT_STATUS;
1757 		isert_completion_put(&isert_cmd->tx_desc, isert_cmd,
1758 				     ib_dev, false);
1759 		break;
1760 	case ISTATE_SEND_LOGOUTRSP:
1761 		iscsit_logout_post_handler(cmd, cmd->conn);
1762 		break;
1763 	default:
1764 		isert_err("Unknown i_state %d\n", cmd->i_state);
1765 		dump_stack();
1766 		break;
1767 	}
1768 }
1769 
1770 static void
1771 isert_login_send_done(struct ib_cq *cq, struct ib_wc *wc)
1772 {
1773 	struct isert_conn *isert_conn = wc->qp->qp_context;
1774 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1775 	struct iser_tx_desc *tx_desc = cqe_to_tx_desc(wc->wr_cqe);
1776 
1777 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1778 		isert_print_wc(wc, "login send");
1779 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1780 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1781 	}
1782 
1783 	isert_unmap_tx_desc(tx_desc, ib_dev);
1784 }
1785 
1786 static void
1787 isert_send_done(struct ib_cq *cq, struct ib_wc *wc)
1788 {
1789 	struct isert_conn *isert_conn = wc->qp->qp_context;
1790 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1791 	struct iser_tx_desc *tx_desc = cqe_to_tx_desc(wc->wr_cqe);
1792 	struct isert_cmd *isert_cmd = tx_desc_to_cmd(tx_desc);
1793 
1794 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1795 		isert_print_wc(wc, "send");
1796 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1797 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1798 		isert_completion_put(tx_desc, isert_cmd, ib_dev, true);
1799 		return;
1800 	}
1801 
1802 	isert_dbg("Cmd %p\n", isert_cmd);
1803 
1804 	switch (isert_cmd->iscsi_cmd->i_state) {
1805 	case ISTATE_SEND_TASKMGTRSP:
1806 	case ISTATE_SEND_LOGOUTRSP:
1807 	case ISTATE_SEND_REJECT:
1808 	case ISTATE_SEND_TEXTRSP:
1809 		isert_unmap_tx_desc(tx_desc, ib_dev);
1810 
1811 		INIT_WORK(&isert_cmd->comp_work, isert_do_control_comp);
1812 		queue_work(isert_comp_wq, &isert_cmd->comp_work);
1813 		return;
1814 	default:
1815 		isert_cmd->iscsi_cmd->i_state = ISTATE_SENT_STATUS;
1816 		isert_completion_put(tx_desc, isert_cmd, ib_dev, false);
1817 		break;
1818 	}
1819 }
1820 
1821 static int
1822 isert_post_response(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd)
1823 {
1824 	int ret;
1825 
1826 	ret = isert_post_recv(isert_conn, isert_cmd->rx_desc);
1827 	if (ret) {
1828 		isert_err("ib_post_recv failed with %d\n", ret);
1829 		return ret;
1830 	}
1831 
1832 	ret = ib_post_send(isert_conn->qp, &isert_cmd->tx_desc.send_wr, NULL);
1833 	if (ret) {
1834 		isert_err("ib_post_send failed with %d\n", ret);
1835 		return ret;
1836 	}
1837 	return ret;
1838 }
1839 
1840 static int
1841 isert_put_response(struct iscsi_conn *conn, struct iscsi_cmd *cmd)
1842 {
1843 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1844 	struct isert_conn *isert_conn = conn->context;
1845 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1846 	struct iscsi_scsi_rsp *hdr = (struct iscsi_scsi_rsp *)
1847 				&isert_cmd->tx_desc.iscsi_header;
1848 
1849 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1850 	iscsit_build_rsp_pdu(cmd, conn, true, hdr);
1851 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1852 	/*
1853 	 * Attach SENSE DATA payload to iSCSI Response PDU
1854 	 */
1855 	if (cmd->se_cmd.sense_buffer &&
1856 	    ((cmd->se_cmd.se_cmd_flags & SCF_TRANSPORT_TASK_SENSE) ||
1857 	    (cmd->se_cmd.se_cmd_flags & SCF_EMULATED_TASK_SENSE))) {
1858 		struct isert_device *device = isert_conn->device;
1859 		struct ib_device *ib_dev = device->ib_device;
1860 		struct ib_sge *tx_dsg = &isert_cmd->tx_desc.tx_sg[1];
1861 		u32 padding, pdu_len;
1862 
1863 		put_unaligned_be16(cmd->se_cmd.scsi_sense_length,
1864 				   cmd->sense_buffer);
1865 		cmd->se_cmd.scsi_sense_length += sizeof(__be16);
1866 
1867 		padding = -(cmd->se_cmd.scsi_sense_length) & 3;
1868 		hton24(hdr->dlength, (u32)cmd->se_cmd.scsi_sense_length);
1869 		pdu_len = cmd->se_cmd.scsi_sense_length + padding;
1870 
1871 		isert_cmd->pdu_buf_dma = ib_dma_map_single(ib_dev,
1872 				(void *)cmd->sense_buffer, pdu_len,
1873 				DMA_TO_DEVICE);
1874 		if (ib_dma_mapping_error(ib_dev, isert_cmd->pdu_buf_dma))
1875 			return -ENOMEM;
1876 
1877 		isert_cmd->pdu_buf_len = pdu_len;
1878 		tx_dsg->addr	= isert_cmd->pdu_buf_dma;
1879 		tx_dsg->length	= pdu_len;
1880 		tx_dsg->lkey	= device->pd->local_dma_lkey;
1881 		isert_cmd->tx_desc.num_sge = 2;
1882 	}
1883 
1884 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1885 
1886 	isert_dbg("Posting SCSI Response\n");
1887 
1888 	return isert_post_response(isert_conn, isert_cmd);
1889 }
1890 
1891 static void
1892 isert_aborted_task(struct iscsi_conn *conn, struct iscsi_cmd *cmd)
1893 {
1894 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1895 	struct isert_conn *isert_conn = conn->context;
1896 
1897 	spin_lock_bh(&conn->cmd_lock);
1898 	if (!list_empty(&cmd->i_conn_node))
1899 		list_del_init(&cmd->i_conn_node);
1900 	spin_unlock_bh(&conn->cmd_lock);
1901 
1902 	if (cmd->data_direction == DMA_TO_DEVICE)
1903 		iscsit_stop_dataout_timer(cmd);
1904 	isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1905 }
1906 
1907 static enum target_prot_op
1908 isert_get_sup_prot_ops(struct iscsi_conn *conn)
1909 {
1910 	struct isert_conn *isert_conn = conn->context;
1911 	struct isert_device *device = isert_conn->device;
1912 
1913 	if (conn->tpg->tpg_attrib.t10_pi) {
1914 		if (device->pi_capable) {
1915 			isert_info("conn %p PI offload enabled\n", isert_conn);
1916 			isert_conn->pi_support = true;
1917 			return TARGET_PROT_ALL;
1918 		}
1919 	}
1920 
1921 	isert_info("conn %p PI offload disabled\n", isert_conn);
1922 	isert_conn->pi_support = false;
1923 
1924 	return TARGET_PROT_NORMAL;
1925 }
1926 
1927 static int
1928 isert_put_nopin(struct iscsi_cmd *cmd, struct iscsi_conn *conn,
1929 		bool nopout_response)
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_nopin_rsp(cmd, conn, (struct iscsi_nopin *)
1937 			       &isert_cmd->tx_desc.iscsi_header,
1938 			       nopout_response);
1939 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1940 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1941 
1942 	isert_dbg("conn %p Posting NOPIN Response\n", isert_conn);
1943 
1944 	return isert_post_response(isert_conn, isert_cmd);
1945 }
1946 
1947 static int
1948 isert_put_logout_rsp(struct iscsi_cmd *cmd, struct iscsi_conn *conn)
1949 {
1950 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1951 	struct isert_conn *isert_conn = conn->context;
1952 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1953 
1954 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1955 	iscsit_build_logout_rsp(cmd, conn, (struct iscsi_logout_rsp *)
1956 				&isert_cmd->tx_desc.iscsi_header);
1957 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1958 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1959 
1960 	isert_dbg("conn %p Posting Logout Response\n", isert_conn);
1961 
1962 	return isert_post_response(isert_conn, isert_cmd);
1963 }
1964 
1965 static int
1966 isert_put_tm_rsp(struct iscsi_cmd *cmd, struct iscsi_conn *conn)
1967 {
1968 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1969 	struct isert_conn *isert_conn = conn->context;
1970 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1971 
1972 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1973 	iscsit_build_task_mgt_rsp(cmd, conn, (struct iscsi_tm_rsp *)
1974 				  &isert_cmd->tx_desc.iscsi_header);
1975 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1976 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1977 
1978 	isert_dbg("conn %p Posting Task Management Response\n", isert_conn);
1979 
1980 	return isert_post_response(isert_conn, isert_cmd);
1981 }
1982 
1983 static int
1984 isert_put_reject(struct iscsi_cmd *cmd, struct iscsi_conn *conn)
1985 {
1986 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1987 	struct isert_conn *isert_conn = conn->context;
1988 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1989 	struct isert_device *device = isert_conn->device;
1990 	struct ib_device *ib_dev = device->ib_device;
1991 	struct ib_sge *tx_dsg = &isert_cmd->tx_desc.tx_sg[1];
1992 	struct iscsi_reject *hdr =
1993 		(struct iscsi_reject *)&isert_cmd->tx_desc.iscsi_header;
1994 
1995 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1996 	iscsit_build_reject(cmd, conn, hdr);
1997 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1998 
1999 	hton24(hdr->dlength, ISCSI_HDR_LEN);
2000 	isert_cmd->pdu_buf_dma = ib_dma_map_single(ib_dev,
2001 			(void *)cmd->buf_ptr, ISCSI_HDR_LEN,
2002 			DMA_TO_DEVICE);
2003 	if (ib_dma_mapping_error(ib_dev, isert_cmd->pdu_buf_dma))
2004 		return -ENOMEM;
2005 	isert_cmd->pdu_buf_len = ISCSI_HDR_LEN;
2006 	tx_dsg->addr	= isert_cmd->pdu_buf_dma;
2007 	tx_dsg->length	= ISCSI_HDR_LEN;
2008 	tx_dsg->lkey	= device->pd->local_dma_lkey;
2009 	isert_cmd->tx_desc.num_sge = 2;
2010 
2011 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
2012 
2013 	isert_dbg("conn %p Posting Reject\n", isert_conn);
2014 
2015 	return isert_post_response(isert_conn, isert_cmd);
2016 }
2017 
2018 static int
2019 isert_put_text_rsp(struct iscsi_cmd *cmd, struct iscsi_conn *conn)
2020 {
2021 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2022 	struct isert_conn *isert_conn = conn->context;
2023 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
2024 	struct iscsi_text_rsp *hdr =
2025 		(struct iscsi_text_rsp *)&isert_cmd->tx_desc.iscsi_header;
2026 	u32 txt_rsp_len;
2027 	int rc;
2028 
2029 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
2030 	rc = iscsit_build_text_rsp(cmd, conn, hdr, ISCSI_INFINIBAND);
2031 	if (rc < 0)
2032 		return rc;
2033 
2034 	txt_rsp_len = rc;
2035 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
2036 
2037 	if (txt_rsp_len) {
2038 		struct isert_device *device = isert_conn->device;
2039 		struct ib_device *ib_dev = device->ib_device;
2040 		struct ib_sge *tx_dsg = &isert_cmd->tx_desc.tx_sg[1];
2041 		void *txt_rsp_buf = cmd->buf_ptr;
2042 
2043 		isert_cmd->pdu_buf_dma = ib_dma_map_single(ib_dev,
2044 				txt_rsp_buf, txt_rsp_len, DMA_TO_DEVICE);
2045 		if (ib_dma_mapping_error(ib_dev, isert_cmd->pdu_buf_dma))
2046 			return -ENOMEM;
2047 
2048 		isert_cmd->pdu_buf_len = txt_rsp_len;
2049 		tx_dsg->addr	= isert_cmd->pdu_buf_dma;
2050 		tx_dsg->length	= txt_rsp_len;
2051 		tx_dsg->lkey	= device->pd->local_dma_lkey;
2052 		isert_cmd->tx_desc.num_sge = 2;
2053 	}
2054 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
2055 
2056 	isert_dbg("conn %p Text Response\n", isert_conn);
2057 
2058 	return isert_post_response(isert_conn, isert_cmd);
2059 }
2060 
2061 static inline void
2062 isert_set_dif_domain(struct se_cmd *se_cmd, struct ib_sig_domain *domain)
2063 {
2064 	domain->sig_type = IB_SIG_TYPE_T10_DIF;
2065 	domain->sig.dif.bg_type = IB_T10DIF_CRC;
2066 	domain->sig.dif.pi_interval = se_cmd->se_dev->dev_attrib.block_size;
2067 	domain->sig.dif.ref_tag = se_cmd->reftag_seed;
2068 	/*
2069 	 * At the moment we hard code those, but if in the future
2070 	 * the target core would like to use it, we will take it
2071 	 * from se_cmd.
2072 	 */
2073 	domain->sig.dif.apptag_check_mask = 0xffff;
2074 	domain->sig.dif.app_escape = true;
2075 	domain->sig.dif.ref_escape = true;
2076 	if (se_cmd->prot_type == TARGET_DIF_TYPE1_PROT ||
2077 	    se_cmd->prot_type == TARGET_DIF_TYPE2_PROT)
2078 		domain->sig.dif.ref_remap = true;
2079 };
2080 
2081 static int
2082 isert_set_sig_attrs(struct se_cmd *se_cmd, struct ib_sig_attrs *sig_attrs)
2083 {
2084 	memset(sig_attrs, 0, sizeof(*sig_attrs));
2085 
2086 	switch (se_cmd->prot_op) {
2087 	case TARGET_PROT_DIN_INSERT:
2088 	case TARGET_PROT_DOUT_STRIP:
2089 		sig_attrs->mem.sig_type = IB_SIG_TYPE_NONE;
2090 		isert_set_dif_domain(se_cmd, &sig_attrs->wire);
2091 		break;
2092 	case TARGET_PROT_DOUT_INSERT:
2093 	case TARGET_PROT_DIN_STRIP:
2094 		sig_attrs->wire.sig_type = IB_SIG_TYPE_NONE;
2095 		isert_set_dif_domain(se_cmd, &sig_attrs->mem);
2096 		break;
2097 	case TARGET_PROT_DIN_PASS:
2098 	case TARGET_PROT_DOUT_PASS:
2099 		isert_set_dif_domain(se_cmd, &sig_attrs->wire);
2100 		isert_set_dif_domain(se_cmd, &sig_attrs->mem);
2101 		break;
2102 	default:
2103 		isert_err("Unsupported PI operation %d\n", se_cmd->prot_op);
2104 		return -EINVAL;
2105 	}
2106 
2107 	if (se_cmd->prot_checks & TARGET_DIF_CHECK_GUARD)
2108 		sig_attrs->check_mask |= IB_SIG_CHECK_GUARD;
2109 	if (se_cmd->prot_checks & TARGET_DIF_CHECK_APPTAG)
2110 		sig_attrs->check_mask |= IB_SIG_CHECK_APPTAG;
2111 	if (se_cmd->prot_checks & TARGET_DIF_CHECK_REFTAG)
2112 		sig_attrs->check_mask |= IB_SIG_CHECK_REFTAG;
2113 
2114 	return 0;
2115 }
2116 
2117 static int
2118 isert_rdma_rw_ctx_post(struct isert_cmd *cmd, struct isert_conn *conn,
2119 		struct ib_cqe *cqe, struct ib_send_wr *chain_wr)
2120 {
2121 	struct se_cmd *se_cmd = &cmd->iscsi_cmd->se_cmd;
2122 	enum dma_data_direction dir = target_reverse_dma_direction(se_cmd);
2123 	u8 port_num = conn->cm_id->port_num;
2124 	u64 addr;
2125 	u32 rkey, offset;
2126 	int ret;
2127 
2128 	if (cmd->ctx_init_done)
2129 		goto rdma_ctx_post;
2130 
2131 	if (dir == DMA_FROM_DEVICE) {
2132 		addr = cmd->write_va;
2133 		rkey = cmd->write_stag;
2134 		offset = cmd->iscsi_cmd->write_data_done;
2135 	} else {
2136 		addr = cmd->read_va;
2137 		rkey = cmd->read_stag;
2138 		offset = 0;
2139 	}
2140 
2141 	if (isert_prot_cmd(conn, se_cmd)) {
2142 		struct ib_sig_attrs sig_attrs;
2143 
2144 		ret = isert_set_sig_attrs(se_cmd, &sig_attrs);
2145 		if (ret)
2146 			return ret;
2147 
2148 		WARN_ON_ONCE(offset);
2149 		ret = rdma_rw_ctx_signature_init(&cmd->rw, conn->qp, port_num,
2150 				se_cmd->t_data_sg, se_cmd->t_data_nents,
2151 				se_cmd->t_prot_sg, se_cmd->t_prot_nents,
2152 				&sig_attrs, addr, rkey, dir);
2153 	} else {
2154 		ret = rdma_rw_ctx_init(&cmd->rw, conn->qp, port_num,
2155 				se_cmd->t_data_sg, se_cmd->t_data_nents,
2156 				offset, addr, rkey, dir);
2157 	}
2158 
2159 	if (ret < 0) {
2160 		isert_err("Cmd: %p failed to prepare RDMA res\n", cmd);
2161 		return ret;
2162 	}
2163 
2164 	cmd->ctx_init_done = true;
2165 
2166 rdma_ctx_post:
2167 	ret = rdma_rw_ctx_post(&cmd->rw, conn->qp, port_num, cqe, chain_wr);
2168 	if (ret < 0)
2169 		isert_err("Cmd: %p failed to post RDMA res\n", cmd);
2170 	return ret;
2171 }
2172 
2173 static int
2174 isert_put_datain(struct iscsi_conn *conn, struct iscsi_cmd *cmd)
2175 {
2176 	struct se_cmd *se_cmd = &cmd->se_cmd;
2177 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2178 	struct isert_conn *isert_conn = conn->context;
2179 	struct ib_cqe *cqe = NULL;
2180 	struct ib_send_wr *chain_wr = NULL;
2181 	int rc;
2182 
2183 	isert_dbg("Cmd: %p RDMA_WRITE data_length: %u\n",
2184 		 isert_cmd, se_cmd->data_length);
2185 
2186 	if (isert_prot_cmd(isert_conn, se_cmd)) {
2187 		isert_cmd->tx_desc.tx_cqe.done = isert_rdma_write_done;
2188 		cqe = &isert_cmd->tx_desc.tx_cqe;
2189 	} else {
2190 		/*
2191 		 * Build isert_conn->tx_desc for iSCSI response PDU and attach
2192 		 */
2193 		isert_create_send_desc(isert_conn, isert_cmd,
2194 				       &isert_cmd->tx_desc);
2195 		iscsit_build_rsp_pdu(cmd, conn, true, (struct iscsi_scsi_rsp *)
2196 				     &isert_cmd->tx_desc.iscsi_header);
2197 		isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
2198 		isert_init_send_wr(isert_conn, isert_cmd,
2199 				   &isert_cmd->tx_desc.send_wr);
2200 
2201 		rc = isert_post_recv(isert_conn, isert_cmd->rx_desc);
2202 		if (rc) {
2203 			isert_err("ib_post_recv failed with %d\n", rc);
2204 			return rc;
2205 		}
2206 
2207 		chain_wr = &isert_cmd->tx_desc.send_wr;
2208 	}
2209 
2210 	rc = isert_rdma_rw_ctx_post(isert_cmd, isert_conn, cqe, chain_wr);
2211 	isert_dbg("Cmd: %p posted RDMA_WRITE for iSER Data READ rc: %d\n",
2212 		  isert_cmd, rc);
2213 	return rc;
2214 }
2215 
2216 static int
2217 isert_get_dataout(struct iscsi_conn *conn, struct iscsi_cmd *cmd, bool recovery)
2218 {
2219 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2220 	int ret;
2221 
2222 	isert_dbg("Cmd: %p RDMA_READ data_length: %u write_data_done: %u\n",
2223 		 isert_cmd, cmd->se_cmd.data_length, cmd->write_data_done);
2224 
2225 	isert_cmd->tx_desc.tx_cqe.done = isert_rdma_read_done;
2226 	ret = isert_rdma_rw_ctx_post(isert_cmd, conn->context,
2227 				     &isert_cmd->tx_desc.tx_cqe, NULL);
2228 
2229 	isert_dbg("Cmd: %p posted RDMA_READ memory for ISER Data WRITE rc: %d\n",
2230 		 isert_cmd, ret);
2231 	return ret;
2232 }
2233 
2234 static int
2235 isert_immediate_queue(struct iscsi_conn *conn, struct iscsi_cmd *cmd, int state)
2236 {
2237 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2238 	int ret = 0;
2239 
2240 	switch (state) {
2241 	case ISTATE_REMOVE:
2242 		spin_lock_bh(&conn->cmd_lock);
2243 		list_del_init(&cmd->i_conn_node);
2244 		spin_unlock_bh(&conn->cmd_lock);
2245 		isert_put_cmd(isert_cmd, true);
2246 		break;
2247 	case ISTATE_SEND_NOPIN_WANT_RESPONSE:
2248 		ret = isert_put_nopin(cmd, conn, false);
2249 		break;
2250 	default:
2251 		isert_err("Unknown immediate state: 0x%02x\n", state);
2252 		ret = -EINVAL;
2253 		break;
2254 	}
2255 
2256 	return ret;
2257 }
2258 
2259 static int
2260 isert_response_queue(struct iscsi_conn *conn, struct iscsi_cmd *cmd, int state)
2261 {
2262 	struct isert_conn *isert_conn = conn->context;
2263 	int ret;
2264 
2265 	switch (state) {
2266 	case ISTATE_SEND_LOGOUTRSP:
2267 		ret = isert_put_logout_rsp(cmd, conn);
2268 		if (!ret)
2269 			isert_conn->logout_posted = true;
2270 		break;
2271 	case ISTATE_SEND_NOPIN:
2272 		ret = isert_put_nopin(cmd, conn, true);
2273 		break;
2274 	case ISTATE_SEND_TASKMGTRSP:
2275 		ret = isert_put_tm_rsp(cmd, conn);
2276 		break;
2277 	case ISTATE_SEND_REJECT:
2278 		ret = isert_put_reject(cmd, conn);
2279 		break;
2280 	case ISTATE_SEND_TEXTRSP:
2281 		ret = isert_put_text_rsp(cmd, conn);
2282 		break;
2283 	case ISTATE_SEND_STATUS:
2284 		/*
2285 		 * Special case for sending non GOOD SCSI status from TX thread
2286 		 * context during pre se_cmd excecution failure.
2287 		 */
2288 		ret = isert_put_response(conn, cmd);
2289 		break;
2290 	default:
2291 		isert_err("Unknown response state: 0x%02x\n", state);
2292 		ret = -EINVAL;
2293 		break;
2294 	}
2295 
2296 	return ret;
2297 }
2298 
2299 struct rdma_cm_id *
2300 isert_setup_id(struct isert_np *isert_np)
2301 {
2302 	struct iscsi_np *np = isert_np->np;
2303 	struct rdma_cm_id *id;
2304 	struct sockaddr *sa;
2305 	int ret;
2306 
2307 	sa = (struct sockaddr *)&np->np_sockaddr;
2308 	isert_dbg("ksockaddr: %p, sa: %p\n", &np->np_sockaddr, sa);
2309 
2310 	id = rdma_create_id(&init_net, isert_cma_handler, isert_np,
2311 			    RDMA_PS_TCP, IB_QPT_RC);
2312 	if (IS_ERR(id)) {
2313 		isert_err("rdma_create_id() failed: %ld\n", PTR_ERR(id));
2314 		ret = PTR_ERR(id);
2315 		goto out;
2316 	}
2317 	isert_dbg("id %p context %p\n", id, id->context);
2318 
2319 	ret = rdma_bind_addr(id, sa);
2320 	if (ret) {
2321 		isert_err("rdma_bind_addr() failed: %d\n", ret);
2322 		goto out_id;
2323 	}
2324 
2325 	ret = rdma_listen(id, 0);
2326 	if (ret) {
2327 		isert_err("rdma_listen() failed: %d\n", ret);
2328 		goto out_id;
2329 	}
2330 
2331 	return id;
2332 out_id:
2333 	rdma_destroy_id(id);
2334 out:
2335 	return ERR_PTR(ret);
2336 }
2337 
2338 static int
2339 isert_setup_np(struct iscsi_np *np,
2340 	       struct sockaddr_storage *ksockaddr)
2341 {
2342 	struct isert_np *isert_np;
2343 	struct rdma_cm_id *isert_lid;
2344 	int ret;
2345 
2346 	isert_np = kzalloc(sizeof(struct isert_np), GFP_KERNEL);
2347 	if (!isert_np)
2348 		return -ENOMEM;
2349 
2350 	sema_init(&isert_np->sem, 0);
2351 	mutex_init(&isert_np->mutex);
2352 	INIT_LIST_HEAD(&isert_np->accepted);
2353 	INIT_LIST_HEAD(&isert_np->pending);
2354 	isert_np->np = np;
2355 
2356 	/*
2357 	 * Setup the np->np_sockaddr from the passed sockaddr setup
2358 	 * in iscsi_target_configfs.c code..
2359 	 */
2360 	memcpy(&np->np_sockaddr, ksockaddr,
2361 	       sizeof(struct sockaddr_storage));
2362 
2363 	isert_lid = isert_setup_id(isert_np);
2364 	if (IS_ERR(isert_lid)) {
2365 		ret = PTR_ERR(isert_lid);
2366 		goto out;
2367 	}
2368 
2369 	isert_np->cm_id = isert_lid;
2370 	np->np_context = isert_np;
2371 
2372 	return 0;
2373 
2374 out:
2375 	kfree(isert_np);
2376 
2377 	return ret;
2378 }
2379 
2380 static int
2381 isert_rdma_accept(struct isert_conn *isert_conn)
2382 {
2383 	struct rdma_cm_id *cm_id = isert_conn->cm_id;
2384 	struct rdma_conn_param cp;
2385 	int ret;
2386 	struct iser_cm_hdr rsp_hdr;
2387 
2388 	memset(&cp, 0, sizeof(struct rdma_conn_param));
2389 	cp.initiator_depth = isert_conn->initiator_depth;
2390 	cp.retry_count = 7;
2391 	cp.rnr_retry_count = 7;
2392 
2393 	memset(&rsp_hdr, 0, sizeof(rsp_hdr));
2394 	rsp_hdr.flags = ISERT_ZBVA_NOT_USED;
2395 	if (!isert_conn->snd_w_inv)
2396 		rsp_hdr.flags = rsp_hdr.flags | ISERT_SEND_W_INV_NOT_USED;
2397 	cp.private_data = (void *)&rsp_hdr;
2398 	cp.private_data_len = sizeof(rsp_hdr);
2399 
2400 	ret = rdma_accept(cm_id, &cp);
2401 	if (ret) {
2402 		isert_err("rdma_accept() failed with: %d\n", ret);
2403 		return ret;
2404 	}
2405 
2406 	return 0;
2407 }
2408 
2409 static int
2410 isert_get_login_rx(struct iscsi_conn *conn, struct iscsi_login *login)
2411 {
2412 	struct isert_conn *isert_conn = conn->context;
2413 	int ret;
2414 
2415 	isert_info("before login_req comp conn: %p\n", isert_conn);
2416 	ret = wait_for_completion_interruptible(&isert_conn->login_req_comp);
2417 	if (ret) {
2418 		isert_err("isert_conn %p interrupted before got login req\n",
2419 			  isert_conn);
2420 		return ret;
2421 	}
2422 	reinit_completion(&isert_conn->login_req_comp);
2423 
2424 	/*
2425 	 * For login requests after the first PDU, isert_rx_login_req() will
2426 	 * kick schedule_delayed_work(&conn->login_work) as the packet is
2427 	 * received, which turns this callback from iscsi_target_do_login_rx()
2428 	 * into a NOP.
2429 	 */
2430 	if (!login->first_request)
2431 		return 0;
2432 
2433 	isert_rx_login_req(isert_conn);
2434 
2435 	isert_info("before login_comp conn: %p\n", conn);
2436 	ret = wait_for_completion_interruptible(&isert_conn->login_comp);
2437 	if (ret)
2438 		return ret;
2439 
2440 	isert_info("processing login->req: %p\n", login->req);
2441 
2442 	return 0;
2443 }
2444 
2445 static void
2446 isert_set_conn_info(struct iscsi_np *np, struct iscsi_conn *conn,
2447 		    struct isert_conn *isert_conn)
2448 {
2449 	struct rdma_cm_id *cm_id = isert_conn->cm_id;
2450 	struct rdma_route *cm_route = &cm_id->route;
2451 
2452 	conn->login_family = np->np_sockaddr.ss_family;
2453 
2454 	conn->login_sockaddr = cm_route->addr.dst_addr;
2455 	conn->local_sockaddr = cm_route->addr.src_addr;
2456 }
2457 
2458 static int
2459 isert_accept_np(struct iscsi_np *np, struct iscsi_conn *conn)
2460 {
2461 	struct isert_np *isert_np = np->np_context;
2462 	struct isert_conn *isert_conn;
2463 	int ret;
2464 
2465 accept_wait:
2466 	ret = down_interruptible(&isert_np->sem);
2467 	if (ret)
2468 		return -ENODEV;
2469 
2470 	spin_lock_bh(&np->np_thread_lock);
2471 	if (np->np_thread_state >= ISCSI_NP_THREAD_RESET) {
2472 		spin_unlock_bh(&np->np_thread_lock);
2473 		isert_dbg("np_thread_state %d\n",
2474 			 np->np_thread_state);
2475 		/**
2476 		 * No point in stalling here when np_thread
2477 		 * is in state RESET/SHUTDOWN/EXIT - bail
2478 		 **/
2479 		return -ENODEV;
2480 	}
2481 	spin_unlock_bh(&np->np_thread_lock);
2482 
2483 	mutex_lock(&isert_np->mutex);
2484 	if (list_empty(&isert_np->pending)) {
2485 		mutex_unlock(&isert_np->mutex);
2486 		goto accept_wait;
2487 	}
2488 	isert_conn = list_first_entry(&isert_np->pending,
2489 			struct isert_conn, node);
2490 	list_del_init(&isert_conn->node);
2491 	mutex_unlock(&isert_np->mutex);
2492 
2493 	conn->context = isert_conn;
2494 	isert_conn->conn = conn;
2495 	isert_conn->state = ISER_CONN_BOUND;
2496 
2497 	isert_set_conn_info(np, conn, isert_conn);
2498 
2499 	isert_dbg("Processing isert_conn: %p\n", isert_conn);
2500 
2501 	return 0;
2502 }
2503 
2504 static void
2505 isert_free_np(struct iscsi_np *np)
2506 {
2507 	struct isert_np *isert_np = np->np_context;
2508 	struct isert_conn *isert_conn, *n;
2509 
2510 	if (isert_np->cm_id)
2511 		rdma_destroy_id(isert_np->cm_id);
2512 
2513 	/*
2514 	 * FIXME: At this point we don't have a good way to insure
2515 	 * that at this point we don't have hanging connections that
2516 	 * completed RDMA establishment but didn't start iscsi login
2517 	 * process. So work-around this by cleaning up what ever piled
2518 	 * up in accepted and pending lists.
2519 	 */
2520 	mutex_lock(&isert_np->mutex);
2521 	if (!list_empty(&isert_np->pending)) {
2522 		isert_info("Still have isert pending connections\n");
2523 		list_for_each_entry_safe(isert_conn, n,
2524 					 &isert_np->pending,
2525 					 node) {
2526 			isert_info("cleaning isert_conn %p state (%d)\n",
2527 				   isert_conn, isert_conn->state);
2528 			isert_connect_release(isert_conn);
2529 		}
2530 	}
2531 
2532 	if (!list_empty(&isert_np->accepted)) {
2533 		isert_info("Still have isert accepted connections\n");
2534 		list_for_each_entry_safe(isert_conn, n,
2535 					 &isert_np->accepted,
2536 					 node) {
2537 			isert_info("cleaning isert_conn %p state (%d)\n",
2538 				   isert_conn, isert_conn->state);
2539 			isert_connect_release(isert_conn);
2540 		}
2541 	}
2542 	mutex_unlock(&isert_np->mutex);
2543 
2544 	np->np_context = NULL;
2545 	kfree(isert_np);
2546 }
2547 
2548 static void isert_release_work(struct work_struct *work)
2549 {
2550 	struct isert_conn *isert_conn = container_of(work,
2551 						     struct isert_conn,
2552 						     release_work);
2553 
2554 	isert_info("Starting release conn %p\n", isert_conn);
2555 
2556 	mutex_lock(&isert_conn->mutex);
2557 	isert_conn->state = ISER_CONN_DOWN;
2558 	mutex_unlock(&isert_conn->mutex);
2559 
2560 	isert_info("Destroying conn %p\n", isert_conn);
2561 	isert_put_conn(isert_conn);
2562 }
2563 
2564 static void
2565 isert_wait4logout(struct isert_conn *isert_conn)
2566 {
2567 	struct iscsi_conn *conn = isert_conn->conn;
2568 
2569 	isert_info("conn %p\n", isert_conn);
2570 
2571 	if (isert_conn->logout_posted) {
2572 		isert_info("conn %p wait for conn_logout_comp\n", isert_conn);
2573 		wait_for_completion_timeout(&conn->conn_logout_comp,
2574 					    SECONDS_FOR_LOGOUT_COMP * HZ);
2575 	}
2576 }
2577 
2578 static void
2579 isert_wait4cmds(struct iscsi_conn *conn)
2580 {
2581 	isert_info("iscsi_conn %p\n", conn);
2582 
2583 	if (conn->sess) {
2584 		target_sess_cmd_list_set_waiting(conn->sess->se_sess);
2585 		target_wait_for_sess_cmds(conn->sess->se_sess);
2586 	}
2587 }
2588 
2589 /**
2590  * isert_put_unsol_pending_cmds() - Drop commands waiting for
2591  *     unsolicitate dataout
2592  * @conn:    iscsi connection
2593  *
2594  * We might still have commands that are waiting for unsolicited
2595  * dataouts messages. We must put the extra reference on those
2596  * before blocking on the target_wait_for_session_cmds
2597  */
2598 static void
2599 isert_put_unsol_pending_cmds(struct iscsi_conn *conn)
2600 {
2601 	struct iscsi_cmd *cmd, *tmp;
2602 	static LIST_HEAD(drop_cmd_list);
2603 
2604 	spin_lock_bh(&conn->cmd_lock);
2605 	list_for_each_entry_safe(cmd, tmp, &conn->conn_cmd_list, i_conn_node) {
2606 		if ((cmd->cmd_flags & ICF_NON_IMMEDIATE_UNSOLICITED_DATA) &&
2607 		    (cmd->write_data_done < conn->sess->sess_ops->FirstBurstLength) &&
2608 		    (cmd->write_data_done < cmd->se_cmd.data_length))
2609 			list_move_tail(&cmd->i_conn_node, &drop_cmd_list);
2610 	}
2611 	spin_unlock_bh(&conn->cmd_lock);
2612 
2613 	list_for_each_entry_safe(cmd, tmp, &drop_cmd_list, i_conn_node) {
2614 		list_del_init(&cmd->i_conn_node);
2615 		if (cmd->i_state != ISTATE_REMOVE) {
2616 			struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2617 
2618 			isert_info("conn %p dropping cmd %p\n", conn, cmd);
2619 			isert_put_cmd(isert_cmd, true);
2620 		}
2621 	}
2622 }
2623 
2624 static void isert_wait_conn(struct iscsi_conn *conn)
2625 {
2626 	struct isert_conn *isert_conn = conn->context;
2627 
2628 	isert_info("Starting conn %p\n", isert_conn);
2629 
2630 	mutex_lock(&isert_conn->mutex);
2631 	isert_conn_terminate(isert_conn);
2632 	mutex_unlock(&isert_conn->mutex);
2633 
2634 	ib_drain_qp(isert_conn->qp);
2635 	isert_put_unsol_pending_cmds(conn);
2636 	isert_wait4cmds(conn);
2637 	isert_wait4logout(isert_conn);
2638 
2639 	queue_work(isert_release_wq, &isert_conn->release_work);
2640 }
2641 
2642 static void isert_free_conn(struct iscsi_conn *conn)
2643 {
2644 	struct isert_conn *isert_conn = conn->context;
2645 
2646 	ib_drain_qp(isert_conn->qp);
2647 	isert_put_conn(isert_conn);
2648 }
2649 
2650 static void isert_get_rx_pdu(struct iscsi_conn *conn)
2651 {
2652 	struct completion comp;
2653 
2654 	init_completion(&comp);
2655 
2656 	wait_for_completion_interruptible(&comp);
2657 }
2658 
2659 static struct iscsit_transport iser_target_transport = {
2660 	.name			= "IB/iSER",
2661 	.transport_type		= ISCSI_INFINIBAND,
2662 	.rdma_shutdown		= true,
2663 	.priv_size		= sizeof(struct isert_cmd),
2664 	.owner			= THIS_MODULE,
2665 	.iscsit_setup_np	= isert_setup_np,
2666 	.iscsit_accept_np	= isert_accept_np,
2667 	.iscsit_free_np		= isert_free_np,
2668 	.iscsit_wait_conn	= isert_wait_conn,
2669 	.iscsit_free_conn	= isert_free_conn,
2670 	.iscsit_get_login_rx	= isert_get_login_rx,
2671 	.iscsit_put_login_tx	= isert_put_login_tx,
2672 	.iscsit_immediate_queue	= isert_immediate_queue,
2673 	.iscsit_response_queue	= isert_response_queue,
2674 	.iscsit_get_dataout	= isert_get_dataout,
2675 	.iscsit_queue_data_in	= isert_put_datain,
2676 	.iscsit_queue_status	= isert_put_response,
2677 	.iscsit_aborted_task	= isert_aborted_task,
2678 	.iscsit_get_rx_pdu	= isert_get_rx_pdu,
2679 	.iscsit_get_sup_prot_ops = isert_get_sup_prot_ops,
2680 };
2681 
2682 static int __init isert_init(void)
2683 {
2684 	int ret;
2685 
2686 	isert_comp_wq = alloc_workqueue("isert_comp_wq",
2687 					WQ_UNBOUND | WQ_HIGHPRI, 0);
2688 	if (!isert_comp_wq) {
2689 		isert_err("Unable to allocate isert_comp_wq\n");
2690 		return -ENOMEM;
2691 	}
2692 
2693 	isert_release_wq = alloc_workqueue("isert_release_wq", WQ_UNBOUND,
2694 					WQ_UNBOUND_MAX_ACTIVE);
2695 	if (!isert_release_wq) {
2696 		isert_err("Unable to allocate isert_release_wq\n");
2697 		ret = -ENOMEM;
2698 		goto destroy_comp_wq;
2699 	}
2700 
2701 	iscsit_register_transport(&iser_target_transport);
2702 	isert_info("iSER_TARGET[0] - Loaded iser_target_transport\n");
2703 
2704 	return 0;
2705 
2706 destroy_comp_wq:
2707 	destroy_workqueue(isert_comp_wq);
2708 
2709 	return ret;
2710 }
2711 
2712 static void __exit isert_exit(void)
2713 {
2714 	flush_scheduled_work();
2715 	destroy_workqueue(isert_release_wq);
2716 	destroy_workqueue(isert_comp_wq);
2717 	iscsit_unregister_transport(&iser_target_transport);
2718 	isert_info("iSER_TARGET[0] - Released iser_target_transport\n");
2719 }
2720 
2721 MODULE_DESCRIPTION("iSER-Target for mainline target infrastructure");
2722 MODULE_AUTHOR("nab@Linux-iSCSI.org");
2723 MODULE_LICENSE("GPL");
2724 
2725 module_init(isert_init);
2726 module_exit(isert_exit);
2727