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 	attr.cap.max_recv_sge = 1;
141 	attr.sq_sig_type = IB_SIGNAL_REQ_WR;
142 	attr.qp_type = IB_QPT_RC;
143 	if (device->pi_capable)
144 		attr.create_flags |= IB_QP_CREATE_SIGNATURE_EN;
145 
146 	ret = rdma_create_qp(cma_id, device->pd, &attr);
147 	if (ret) {
148 		isert_err("rdma_create_qp failed for cma_id %d\n", ret);
149 		return ERR_PTR(ret);
150 	}
151 
152 	return cma_id->qp;
153 }
154 
155 static int
156 isert_conn_setup_qp(struct isert_conn *isert_conn, struct rdma_cm_id *cma_id)
157 {
158 	struct isert_comp *comp;
159 	int ret;
160 
161 	comp = isert_comp_get(isert_conn);
162 	isert_conn->qp = isert_create_qp(isert_conn, comp, cma_id);
163 	if (IS_ERR(isert_conn->qp)) {
164 		ret = PTR_ERR(isert_conn->qp);
165 		goto err;
166 	}
167 
168 	return 0;
169 err:
170 	isert_comp_put(comp);
171 	return ret;
172 }
173 
174 static int
175 isert_alloc_rx_descriptors(struct isert_conn *isert_conn)
176 {
177 	struct isert_device *device = isert_conn->device;
178 	struct ib_device *ib_dev = device->ib_device;
179 	struct iser_rx_desc *rx_desc;
180 	struct ib_sge *rx_sg;
181 	u64 dma_addr;
182 	int i, j;
183 
184 	isert_conn->rx_descs = kzalloc(ISERT_QP_MAX_RECV_DTOS *
185 				sizeof(struct iser_rx_desc), GFP_KERNEL);
186 	if (!isert_conn->rx_descs)
187 		return -ENOMEM;
188 
189 	rx_desc = isert_conn->rx_descs;
190 
191 	for (i = 0; i < ISERT_QP_MAX_RECV_DTOS; i++, rx_desc++)  {
192 		dma_addr = ib_dma_map_single(ib_dev, (void *)rx_desc,
193 					ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
194 		if (ib_dma_mapping_error(ib_dev, dma_addr))
195 			goto dma_map_fail;
196 
197 		rx_desc->dma_addr = dma_addr;
198 
199 		rx_sg = &rx_desc->rx_sg;
200 		rx_sg->addr = rx_desc->dma_addr;
201 		rx_sg->length = ISER_RX_PAYLOAD_SIZE;
202 		rx_sg->lkey = device->pd->local_dma_lkey;
203 		rx_desc->rx_cqe.done = isert_recv_done;
204 	}
205 
206 	return 0;
207 
208 dma_map_fail:
209 	rx_desc = isert_conn->rx_descs;
210 	for (j = 0; j < i; j++, rx_desc++) {
211 		ib_dma_unmap_single(ib_dev, rx_desc->dma_addr,
212 				    ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
213 	}
214 	kfree(isert_conn->rx_descs);
215 	isert_conn->rx_descs = NULL;
216 	isert_err("conn %p failed to allocate rx descriptors\n", isert_conn);
217 	return -ENOMEM;
218 }
219 
220 static void
221 isert_free_rx_descriptors(struct isert_conn *isert_conn)
222 {
223 	struct ib_device *ib_dev = isert_conn->device->ib_device;
224 	struct iser_rx_desc *rx_desc;
225 	int i;
226 
227 	if (!isert_conn->rx_descs)
228 		return;
229 
230 	rx_desc = isert_conn->rx_descs;
231 	for (i = 0; i < ISERT_QP_MAX_RECV_DTOS; i++, rx_desc++)  {
232 		ib_dma_unmap_single(ib_dev, rx_desc->dma_addr,
233 				    ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
234 	}
235 
236 	kfree(isert_conn->rx_descs);
237 	isert_conn->rx_descs = NULL;
238 }
239 
240 static void
241 isert_free_comps(struct isert_device *device)
242 {
243 	int i;
244 
245 	for (i = 0; i < device->comps_used; i++) {
246 		struct isert_comp *comp = &device->comps[i];
247 
248 		if (comp->cq)
249 			ib_free_cq(comp->cq);
250 	}
251 	kfree(device->comps);
252 }
253 
254 static int
255 isert_alloc_comps(struct isert_device *device)
256 {
257 	int i, max_cqe, ret = 0;
258 
259 	device->comps_used = min(ISERT_MAX_CQ, min_t(int, num_online_cpus(),
260 				 device->ib_device->num_comp_vectors));
261 
262 	isert_info("Using %d CQs, %s supports %d vectors support "
263 		   "pi_capable %d\n",
264 		   device->comps_used, device->ib_device->name,
265 		   device->ib_device->num_comp_vectors,
266 		   device->pi_capable);
267 
268 	device->comps = kcalloc(device->comps_used, sizeof(struct isert_comp),
269 				GFP_KERNEL);
270 	if (!device->comps)
271 		return -ENOMEM;
272 
273 	max_cqe = min(ISER_MAX_CQ_LEN, device->ib_device->attrs.max_cqe);
274 
275 	for (i = 0; i < device->comps_used; i++) {
276 		struct isert_comp *comp = &device->comps[i];
277 
278 		comp->device = device;
279 		comp->cq = ib_alloc_cq(device->ib_device, comp, max_cqe, i,
280 				IB_POLL_WORKQUEUE);
281 		if (IS_ERR(comp->cq)) {
282 			isert_err("Unable to allocate cq\n");
283 			ret = PTR_ERR(comp->cq);
284 			comp->cq = NULL;
285 			goto out_cq;
286 		}
287 	}
288 
289 	return 0;
290 out_cq:
291 	isert_free_comps(device);
292 	return ret;
293 }
294 
295 static int
296 isert_create_device_ib_res(struct isert_device *device)
297 {
298 	struct ib_device *ib_dev = device->ib_device;
299 	int ret;
300 
301 	isert_dbg("devattr->max_sge: %d\n", ib_dev->attrs.max_sge);
302 	isert_dbg("devattr->max_sge_rd: %d\n", ib_dev->attrs.max_sge_rd);
303 
304 	ret = isert_alloc_comps(device);
305 	if (ret)
306 		goto out;
307 
308 	device->pd = ib_alloc_pd(ib_dev, 0);
309 	if (IS_ERR(device->pd)) {
310 		ret = PTR_ERR(device->pd);
311 		isert_err("failed to allocate pd, device %p, ret=%d\n",
312 			  device, ret);
313 		goto out_cq;
314 	}
315 
316 	/* Check signature cap */
317 	device->pi_capable = ib_dev->attrs.device_cap_flags &
318 			     IB_DEVICE_SIGNATURE_HANDOVER ? true : false;
319 
320 	return 0;
321 
322 out_cq:
323 	isert_free_comps(device);
324 out:
325 	if (ret > 0)
326 		ret = -EINVAL;
327 	return ret;
328 }
329 
330 static void
331 isert_free_device_ib_res(struct isert_device *device)
332 {
333 	isert_info("device %p\n", device);
334 
335 	ib_dealloc_pd(device->pd);
336 	isert_free_comps(device);
337 }
338 
339 static void
340 isert_device_put(struct isert_device *device)
341 {
342 	mutex_lock(&device_list_mutex);
343 	device->refcount--;
344 	isert_info("device %p refcount %d\n", device, device->refcount);
345 	if (!device->refcount) {
346 		isert_free_device_ib_res(device);
347 		list_del(&device->dev_node);
348 		kfree(device);
349 	}
350 	mutex_unlock(&device_list_mutex);
351 }
352 
353 static struct isert_device *
354 isert_device_get(struct rdma_cm_id *cma_id)
355 {
356 	struct isert_device *device;
357 	int ret;
358 
359 	mutex_lock(&device_list_mutex);
360 	list_for_each_entry(device, &device_list, dev_node) {
361 		if (device->ib_device->node_guid == cma_id->device->node_guid) {
362 			device->refcount++;
363 			isert_info("Found iser device %p refcount %d\n",
364 				   device, device->refcount);
365 			mutex_unlock(&device_list_mutex);
366 			return device;
367 		}
368 	}
369 
370 	device = kzalloc(sizeof(struct isert_device), GFP_KERNEL);
371 	if (!device) {
372 		mutex_unlock(&device_list_mutex);
373 		return ERR_PTR(-ENOMEM);
374 	}
375 
376 	INIT_LIST_HEAD(&device->dev_node);
377 
378 	device->ib_device = cma_id->device;
379 	ret = isert_create_device_ib_res(device);
380 	if (ret) {
381 		kfree(device);
382 		mutex_unlock(&device_list_mutex);
383 		return ERR_PTR(ret);
384 	}
385 
386 	device->refcount++;
387 	list_add_tail(&device->dev_node, &device_list);
388 	isert_info("Created a new iser device %p refcount %d\n",
389 		   device, device->refcount);
390 	mutex_unlock(&device_list_mutex);
391 
392 	return device;
393 }
394 
395 static void
396 isert_init_conn(struct isert_conn *isert_conn)
397 {
398 	isert_conn->state = ISER_CONN_INIT;
399 	INIT_LIST_HEAD(&isert_conn->node);
400 	init_completion(&isert_conn->login_comp);
401 	init_completion(&isert_conn->login_req_comp);
402 	init_waitqueue_head(&isert_conn->rem_wait);
403 	kref_init(&isert_conn->kref);
404 	mutex_init(&isert_conn->mutex);
405 	INIT_WORK(&isert_conn->release_work, isert_release_work);
406 }
407 
408 static void
409 isert_free_login_buf(struct isert_conn *isert_conn)
410 {
411 	struct ib_device *ib_dev = isert_conn->device->ib_device;
412 
413 	ib_dma_unmap_single(ib_dev, isert_conn->login_rsp_dma,
414 			    ISER_RX_PAYLOAD_SIZE, DMA_TO_DEVICE);
415 	kfree(isert_conn->login_rsp_buf);
416 
417 	ib_dma_unmap_single(ib_dev, isert_conn->login_req_dma,
418 			    ISER_RX_PAYLOAD_SIZE,
419 			    DMA_FROM_DEVICE);
420 	kfree(isert_conn->login_req_buf);
421 }
422 
423 static int
424 isert_alloc_login_buf(struct isert_conn *isert_conn,
425 		      struct ib_device *ib_dev)
426 {
427 	int ret;
428 
429 	isert_conn->login_req_buf = kzalloc(sizeof(*isert_conn->login_req_buf),
430 			GFP_KERNEL);
431 	if (!isert_conn->login_req_buf)
432 		return -ENOMEM;
433 
434 	isert_conn->login_req_dma = ib_dma_map_single(ib_dev,
435 				isert_conn->login_req_buf,
436 				ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
437 	ret = ib_dma_mapping_error(ib_dev, isert_conn->login_req_dma);
438 	if (ret) {
439 		isert_err("login_req_dma mapping error: %d\n", ret);
440 		isert_conn->login_req_dma = 0;
441 		goto out_free_login_req_buf;
442 	}
443 
444 	isert_conn->login_rsp_buf = kzalloc(ISER_RX_PAYLOAD_SIZE, GFP_KERNEL);
445 	if (!isert_conn->login_rsp_buf) {
446 		ret = -ENOMEM;
447 		goto out_unmap_login_req_buf;
448 	}
449 
450 	isert_conn->login_rsp_dma = ib_dma_map_single(ib_dev,
451 					isert_conn->login_rsp_buf,
452 					ISER_RX_PAYLOAD_SIZE, DMA_TO_DEVICE);
453 	ret = ib_dma_mapping_error(ib_dev, isert_conn->login_rsp_dma);
454 	if (ret) {
455 		isert_err("login_rsp_dma mapping error: %d\n", ret);
456 		isert_conn->login_rsp_dma = 0;
457 		goto out_free_login_rsp_buf;
458 	}
459 
460 	return 0;
461 
462 out_free_login_rsp_buf:
463 	kfree(isert_conn->login_rsp_buf);
464 out_unmap_login_req_buf:
465 	ib_dma_unmap_single(ib_dev, isert_conn->login_req_dma,
466 			    ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
467 out_free_login_req_buf:
468 	kfree(isert_conn->login_req_buf);
469 	return ret;
470 }
471 
472 static void
473 isert_set_nego_params(struct isert_conn *isert_conn,
474 		      struct rdma_conn_param *param)
475 {
476 	struct ib_device_attr *attr = &isert_conn->device->ib_device->attrs;
477 
478 	/* Set max inflight RDMA READ requests */
479 	isert_conn->initiator_depth = min_t(u8, param->initiator_depth,
480 				attr->max_qp_init_rd_atom);
481 	isert_dbg("Using initiator_depth: %u\n", isert_conn->initiator_depth);
482 
483 	if (param->private_data) {
484 		u8 flags = *(u8 *)param->private_data;
485 
486 		/*
487 		 * use remote invalidation if the both initiator
488 		 * and the HCA support it
489 		 */
490 		isert_conn->snd_w_inv = !(flags & ISER_SEND_W_INV_NOT_SUP) &&
491 					  (attr->device_cap_flags &
492 					   IB_DEVICE_MEM_MGT_EXTENSIONS);
493 		if (isert_conn->snd_w_inv)
494 			isert_info("Using remote invalidation\n");
495 	}
496 }
497 
498 static int
499 isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
500 {
501 	struct isert_np *isert_np = cma_id->context;
502 	struct iscsi_np *np = isert_np->np;
503 	struct isert_conn *isert_conn;
504 	struct isert_device *device;
505 	int ret = 0;
506 
507 	spin_lock_bh(&np->np_thread_lock);
508 	if (!np->enabled) {
509 		spin_unlock_bh(&np->np_thread_lock);
510 		isert_dbg("iscsi_np is not enabled, reject connect request\n");
511 		return rdma_reject(cma_id, NULL, 0);
512 	}
513 	spin_unlock_bh(&np->np_thread_lock);
514 
515 	isert_dbg("cma_id: %p, portal: %p\n",
516 		 cma_id, cma_id->context);
517 
518 	isert_conn = kzalloc(sizeof(struct isert_conn), GFP_KERNEL);
519 	if (!isert_conn)
520 		return -ENOMEM;
521 
522 	isert_init_conn(isert_conn);
523 	isert_conn->cm_id = cma_id;
524 
525 	ret = isert_alloc_login_buf(isert_conn, cma_id->device);
526 	if (ret)
527 		goto out;
528 
529 	device = isert_device_get(cma_id);
530 	if (IS_ERR(device)) {
531 		ret = PTR_ERR(device);
532 		goto out_rsp_dma_map;
533 	}
534 	isert_conn->device = device;
535 
536 	isert_set_nego_params(isert_conn, &event->param.conn);
537 
538 	ret = isert_conn_setup_qp(isert_conn, cma_id);
539 	if (ret)
540 		goto out_conn_dev;
541 
542 	ret = isert_login_post_recv(isert_conn);
543 	if (ret)
544 		goto out_conn_dev;
545 
546 	ret = isert_rdma_accept(isert_conn);
547 	if (ret)
548 		goto out_conn_dev;
549 
550 	mutex_lock(&isert_np->mutex);
551 	list_add_tail(&isert_conn->node, &isert_np->accepted);
552 	mutex_unlock(&isert_np->mutex);
553 
554 	return 0;
555 
556 out_conn_dev:
557 	isert_device_put(device);
558 out_rsp_dma_map:
559 	isert_free_login_buf(isert_conn);
560 out:
561 	kfree(isert_conn);
562 	rdma_reject(cma_id, NULL, 0);
563 	return ret;
564 }
565 
566 static void
567 isert_connect_release(struct isert_conn *isert_conn)
568 {
569 	struct isert_device *device = isert_conn->device;
570 
571 	isert_dbg("conn %p\n", isert_conn);
572 
573 	BUG_ON(!device);
574 
575 	isert_free_rx_descriptors(isert_conn);
576 	if (isert_conn->cm_id &&
577 	    !isert_conn->dev_removed)
578 		rdma_destroy_id(isert_conn->cm_id);
579 
580 	if (isert_conn->qp) {
581 		struct isert_comp *comp = isert_conn->qp->recv_cq->cq_context;
582 
583 		isert_comp_put(comp);
584 		ib_destroy_qp(isert_conn->qp);
585 	}
586 
587 	if (isert_conn->login_req_buf)
588 		isert_free_login_buf(isert_conn);
589 
590 	isert_device_put(device);
591 
592 	if (isert_conn->dev_removed)
593 		wake_up_interruptible(&isert_conn->rem_wait);
594 	else
595 		kfree(isert_conn);
596 }
597 
598 static void
599 isert_connected_handler(struct rdma_cm_id *cma_id)
600 {
601 	struct isert_conn *isert_conn = cma_id->qp->qp_context;
602 	struct isert_np *isert_np = cma_id->context;
603 
604 	isert_info("conn %p\n", isert_conn);
605 
606 	mutex_lock(&isert_conn->mutex);
607 	isert_conn->state = ISER_CONN_UP;
608 	kref_get(&isert_conn->kref);
609 	mutex_unlock(&isert_conn->mutex);
610 
611 	mutex_lock(&isert_np->mutex);
612 	list_move_tail(&isert_conn->node, &isert_np->pending);
613 	mutex_unlock(&isert_np->mutex);
614 
615 	isert_info("np %p: Allow accept_np to continue\n", isert_np);
616 	up(&isert_np->sem);
617 }
618 
619 static void
620 isert_release_kref(struct kref *kref)
621 {
622 	struct isert_conn *isert_conn = container_of(kref,
623 				struct isert_conn, kref);
624 
625 	isert_info("conn %p final kref %s/%d\n", isert_conn, current->comm,
626 		   current->pid);
627 
628 	isert_connect_release(isert_conn);
629 }
630 
631 static void
632 isert_put_conn(struct isert_conn *isert_conn)
633 {
634 	kref_put(&isert_conn->kref, isert_release_kref);
635 }
636 
637 static void
638 isert_handle_unbound_conn(struct isert_conn *isert_conn)
639 {
640 	struct isert_np *isert_np = isert_conn->cm_id->context;
641 
642 	mutex_lock(&isert_np->mutex);
643 	if (!list_empty(&isert_conn->node)) {
644 		/*
645 		 * This means iscsi doesn't know this connection
646 		 * so schedule a cleanup ourselves
647 		 */
648 		list_del_init(&isert_conn->node);
649 		isert_put_conn(isert_conn);
650 		queue_work(isert_release_wq, &isert_conn->release_work);
651 	}
652 	mutex_unlock(&isert_np->mutex);
653 }
654 
655 /**
656  * isert_conn_terminate() - Initiate connection termination
657  * @isert_conn: isert connection struct
658  *
659  * Notes:
660  * In case the connection state is BOUND, move state
661  * to TEMINATING and start teardown sequence (rdma_disconnect).
662  * In case the connection state is UP, complete flush as well.
663  *
664  * This routine must be called with mutex held. Thus it is
665  * safe to call multiple times.
666  */
667 static void
668 isert_conn_terminate(struct isert_conn *isert_conn)
669 {
670 	int err;
671 
672 	if (isert_conn->state >= ISER_CONN_TERMINATING)
673 		return;
674 
675 	isert_info("Terminating conn %p state %d\n",
676 		   isert_conn, isert_conn->state);
677 	isert_conn->state = ISER_CONN_TERMINATING;
678 	err = rdma_disconnect(isert_conn->cm_id);
679 	if (err)
680 		isert_warn("Failed rdma_disconnect isert_conn %p\n",
681 			   isert_conn);
682 }
683 
684 static int
685 isert_np_cma_handler(struct isert_np *isert_np,
686 		     enum rdma_cm_event_type event)
687 {
688 	isert_dbg("%s (%d): isert np %p\n",
689 		  rdma_event_msg(event), event, isert_np);
690 
691 	switch (event) {
692 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
693 		isert_np->cm_id = NULL;
694 		break;
695 	case RDMA_CM_EVENT_ADDR_CHANGE:
696 		isert_np->cm_id = isert_setup_id(isert_np);
697 		if (IS_ERR(isert_np->cm_id)) {
698 			isert_err("isert np %p setup id failed: %ld\n",
699 				  isert_np, PTR_ERR(isert_np->cm_id));
700 			isert_np->cm_id = NULL;
701 		}
702 		break;
703 	default:
704 		isert_err("isert np %p Unexpected event %d\n",
705 			  isert_np, event);
706 	}
707 
708 	return -1;
709 }
710 
711 static int
712 isert_disconnected_handler(struct rdma_cm_id *cma_id,
713 			   enum rdma_cm_event_type event)
714 {
715 	struct isert_conn *isert_conn = cma_id->qp->qp_context;
716 
717 	mutex_lock(&isert_conn->mutex);
718 	switch (isert_conn->state) {
719 	case ISER_CONN_TERMINATING:
720 		break;
721 	case ISER_CONN_UP:
722 		isert_conn_terminate(isert_conn);
723 		ib_drain_qp(isert_conn->qp);
724 		isert_handle_unbound_conn(isert_conn);
725 		break;
726 	case ISER_CONN_BOUND:
727 	case ISER_CONN_FULL_FEATURE: /* FALLTHRU */
728 		iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
729 		break;
730 	default:
731 		isert_warn("conn %p terminating in state %d\n",
732 			   isert_conn, isert_conn->state);
733 	}
734 	mutex_unlock(&isert_conn->mutex);
735 
736 	return 0;
737 }
738 
739 static int
740 isert_connect_error(struct rdma_cm_id *cma_id)
741 {
742 	struct isert_conn *isert_conn = cma_id->qp->qp_context;
743 
744 	list_del_init(&isert_conn->node);
745 	isert_conn->cm_id = NULL;
746 	isert_put_conn(isert_conn);
747 
748 	return -1;
749 }
750 
751 static int
752 isert_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
753 {
754 	struct isert_np *isert_np = cma_id->context;
755 	struct isert_conn *isert_conn;
756 	int ret = 0;
757 
758 	isert_info("%s (%d): status %d id %p np %p\n",
759 		   rdma_event_msg(event->event), event->event,
760 		   event->status, cma_id, cma_id->context);
761 
762 	if (isert_np->cm_id == cma_id)
763 		return isert_np_cma_handler(cma_id->context, event->event);
764 
765 	switch (event->event) {
766 	case RDMA_CM_EVENT_CONNECT_REQUEST:
767 		ret = isert_connect_request(cma_id, event);
768 		if (ret)
769 			isert_err("failed handle connect request %d\n", ret);
770 		break;
771 	case RDMA_CM_EVENT_ESTABLISHED:
772 		isert_connected_handler(cma_id);
773 		break;
774 	case RDMA_CM_EVENT_ADDR_CHANGE:    /* FALLTHRU */
775 	case RDMA_CM_EVENT_DISCONNECTED:   /* FALLTHRU */
776 	case RDMA_CM_EVENT_TIMEWAIT_EXIT:  /* FALLTHRU */
777 		ret = isert_disconnected_handler(cma_id, event->event);
778 		break;
779 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
780 		isert_conn = cma_id->qp->qp_context;
781 		isert_conn->dev_removed = true;
782 		isert_disconnected_handler(cma_id, event->event);
783 		wait_event_interruptible(isert_conn->rem_wait,
784 					 isert_conn->state == ISER_CONN_DOWN);
785 		kfree(isert_conn);
786 		/*
787 		 * return non-zero from the callback to destroy
788 		 * the rdma cm id
789 		 */
790 		return 1;
791 	case RDMA_CM_EVENT_REJECTED:       /* FALLTHRU */
792 		isert_info("Connection rejected: %s\n",
793 			   rdma_reject_msg(cma_id, event->status));
794 	case RDMA_CM_EVENT_UNREACHABLE:    /* FALLTHRU */
795 	case RDMA_CM_EVENT_CONNECT_ERROR:
796 		ret = isert_connect_error(cma_id);
797 		break;
798 	default:
799 		isert_err("Unhandled RDMA CMA event: %d\n", event->event);
800 		break;
801 	}
802 
803 	return ret;
804 }
805 
806 static int
807 isert_post_recvm(struct isert_conn *isert_conn, u32 count)
808 {
809 	struct ib_recv_wr *rx_wr, *rx_wr_failed;
810 	int i, ret;
811 	struct iser_rx_desc *rx_desc;
812 
813 	for (rx_wr = isert_conn->rx_wr, i = 0; i < count; i++, rx_wr++) {
814 		rx_desc = &isert_conn->rx_descs[i];
815 
816 		rx_wr->wr_cqe = &rx_desc->rx_cqe;
817 		rx_wr->sg_list = &rx_desc->rx_sg;
818 		rx_wr->num_sge = 1;
819 		rx_wr->next = rx_wr + 1;
820 		rx_desc->in_use = false;
821 	}
822 	rx_wr--;
823 	rx_wr->next = NULL; /* mark end of work requests list */
824 
825 	ret = ib_post_recv(isert_conn->qp, isert_conn->rx_wr,
826 			   &rx_wr_failed);
827 	if (ret)
828 		isert_err("ib_post_recv() failed with ret: %d\n", ret);
829 
830 	return ret;
831 }
832 
833 static int
834 isert_post_recv(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc)
835 {
836 	struct ib_recv_wr *rx_wr_failed, rx_wr;
837 	int ret;
838 
839 	if (!rx_desc->in_use) {
840 		/*
841 		 * if the descriptor is not in-use we already reposted it
842 		 * for recv, so just silently return
843 		 */
844 		return 0;
845 	}
846 
847 	rx_desc->in_use = false;
848 	rx_wr.wr_cqe = &rx_desc->rx_cqe;
849 	rx_wr.sg_list = &rx_desc->rx_sg;
850 	rx_wr.num_sge = 1;
851 	rx_wr.next = NULL;
852 
853 	ret = ib_post_recv(isert_conn->qp, &rx_wr, &rx_wr_failed);
854 	if (ret)
855 		isert_err("ib_post_recv() failed with ret: %d\n", ret);
856 
857 	return ret;
858 }
859 
860 static int
861 isert_login_post_send(struct isert_conn *isert_conn, struct iser_tx_desc *tx_desc)
862 {
863 	struct ib_device *ib_dev = isert_conn->cm_id->device;
864 	struct ib_send_wr send_wr, *send_wr_failed;
865 	int ret;
866 
867 	ib_dma_sync_single_for_device(ib_dev, tx_desc->dma_addr,
868 				      ISER_HEADERS_LEN, DMA_TO_DEVICE);
869 
870 	tx_desc->tx_cqe.done = isert_login_send_done;
871 
872 	send_wr.next	= NULL;
873 	send_wr.wr_cqe	= &tx_desc->tx_cqe;
874 	send_wr.sg_list	= tx_desc->tx_sg;
875 	send_wr.num_sge	= tx_desc->num_sge;
876 	send_wr.opcode	= IB_WR_SEND;
877 	send_wr.send_flags = IB_SEND_SIGNALED;
878 
879 	ret = ib_post_send(isert_conn->qp, &send_wr, &send_wr_failed);
880 	if (ret)
881 		isert_err("ib_post_send() failed, ret: %d\n", ret);
882 
883 	return ret;
884 }
885 
886 static void
887 isert_create_send_desc(struct isert_conn *isert_conn,
888 		       struct isert_cmd *isert_cmd,
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 
894 	ib_dma_sync_single_for_cpu(ib_dev, tx_desc->dma_addr,
895 				   ISER_HEADERS_LEN, DMA_TO_DEVICE);
896 
897 	memset(&tx_desc->iser_header, 0, sizeof(struct iser_ctrl));
898 	tx_desc->iser_header.flags = ISCSI_CTRL;
899 
900 	tx_desc->num_sge = 1;
901 
902 	if (tx_desc->tx_sg[0].lkey != device->pd->local_dma_lkey) {
903 		tx_desc->tx_sg[0].lkey = device->pd->local_dma_lkey;
904 		isert_dbg("tx_desc %p lkey mismatch, fixing\n", tx_desc);
905 	}
906 }
907 
908 static int
909 isert_init_tx_hdrs(struct isert_conn *isert_conn,
910 		   struct iser_tx_desc *tx_desc)
911 {
912 	struct isert_device *device = isert_conn->device;
913 	struct ib_device *ib_dev = device->ib_device;
914 	u64 dma_addr;
915 
916 	dma_addr = ib_dma_map_single(ib_dev, (void *)tx_desc,
917 			ISER_HEADERS_LEN, DMA_TO_DEVICE);
918 	if (ib_dma_mapping_error(ib_dev, dma_addr)) {
919 		isert_err("ib_dma_mapping_error() failed\n");
920 		return -ENOMEM;
921 	}
922 
923 	tx_desc->dma_addr = dma_addr;
924 	tx_desc->tx_sg[0].addr	= tx_desc->dma_addr;
925 	tx_desc->tx_sg[0].length = ISER_HEADERS_LEN;
926 	tx_desc->tx_sg[0].lkey = device->pd->local_dma_lkey;
927 
928 	isert_dbg("Setup tx_sg[0].addr: 0x%llx length: %u lkey: 0x%x\n",
929 		  tx_desc->tx_sg[0].addr, tx_desc->tx_sg[0].length,
930 		  tx_desc->tx_sg[0].lkey);
931 
932 	return 0;
933 }
934 
935 static void
936 isert_init_send_wr(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
937 		   struct ib_send_wr *send_wr)
938 {
939 	struct iser_tx_desc *tx_desc = &isert_cmd->tx_desc;
940 
941 	tx_desc->tx_cqe.done = isert_send_done;
942 	send_wr->wr_cqe = &tx_desc->tx_cqe;
943 
944 	if (isert_conn->snd_w_inv && isert_cmd->inv_rkey) {
945 		send_wr->opcode  = IB_WR_SEND_WITH_INV;
946 		send_wr->ex.invalidate_rkey = isert_cmd->inv_rkey;
947 	} else {
948 		send_wr->opcode = IB_WR_SEND;
949 	}
950 
951 	send_wr->sg_list = &tx_desc->tx_sg[0];
952 	send_wr->num_sge = isert_cmd->tx_desc.num_sge;
953 	send_wr->send_flags = IB_SEND_SIGNALED;
954 }
955 
956 static int
957 isert_login_post_recv(struct isert_conn *isert_conn)
958 {
959 	struct ib_recv_wr rx_wr, *rx_wr_fail;
960 	struct ib_sge sge;
961 	int ret;
962 
963 	memset(&sge, 0, sizeof(struct ib_sge));
964 	sge.addr = isert_conn->login_req_dma;
965 	sge.length = ISER_RX_PAYLOAD_SIZE;
966 	sge.lkey = isert_conn->device->pd->local_dma_lkey;
967 
968 	isert_dbg("Setup sge: addr: %llx length: %d 0x%08x\n",
969 		sge.addr, sge.length, sge.lkey);
970 
971 	isert_conn->login_req_buf->rx_cqe.done = isert_login_recv_done;
972 
973 	memset(&rx_wr, 0, sizeof(struct ib_recv_wr));
974 	rx_wr.wr_cqe = &isert_conn->login_req_buf->rx_cqe;
975 	rx_wr.sg_list = &sge;
976 	rx_wr.num_sge = 1;
977 
978 	ret = ib_post_recv(isert_conn->qp, &rx_wr, &rx_wr_fail);
979 	if (ret)
980 		isert_err("ib_post_recv() failed: %d\n", ret);
981 
982 	return ret;
983 }
984 
985 static int
986 isert_put_login_tx(struct iscsi_conn *conn, struct iscsi_login *login,
987 		   u32 length)
988 {
989 	struct isert_conn *isert_conn = conn->context;
990 	struct isert_device *device = isert_conn->device;
991 	struct ib_device *ib_dev = device->ib_device;
992 	struct iser_tx_desc *tx_desc = &isert_conn->login_tx_desc;
993 	int ret;
994 
995 	isert_create_send_desc(isert_conn, NULL, tx_desc);
996 
997 	memcpy(&tx_desc->iscsi_header, &login->rsp[0],
998 	       sizeof(struct iscsi_hdr));
999 
1000 	isert_init_tx_hdrs(isert_conn, tx_desc);
1001 
1002 	if (length > 0) {
1003 		struct ib_sge *tx_dsg = &tx_desc->tx_sg[1];
1004 
1005 		ib_dma_sync_single_for_cpu(ib_dev, isert_conn->login_rsp_dma,
1006 					   length, DMA_TO_DEVICE);
1007 
1008 		memcpy(isert_conn->login_rsp_buf, login->rsp_buf, length);
1009 
1010 		ib_dma_sync_single_for_device(ib_dev, isert_conn->login_rsp_dma,
1011 					      length, DMA_TO_DEVICE);
1012 
1013 		tx_dsg->addr	= isert_conn->login_rsp_dma;
1014 		tx_dsg->length	= length;
1015 		tx_dsg->lkey	= isert_conn->device->pd->local_dma_lkey;
1016 		tx_desc->num_sge = 2;
1017 	}
1018 	if (!login->login_failed) {
1019 		if (login->login_complete) {
1020 			ret = isert_alloc_rx_descriptors(isert_conn);
1021 			if (ret)
1022 				return ret;
1023 
1024 			ret = isert_post_recvm(isert_conn,
1025 					       ISERT_QP_MAX_RECV_DTOS);
1026 			if (ret)
1027 				return ret;
1028 
1029 			/* Now we are in FULL_FEATURE phase */
1030 			mutex_lock(&isert_conn->mutex);
1031 			isert_conn->state = ISER_CONN_FULL_FEATURE;
1032 			mutex_unlock(&isert_conn->mutex);
1033 			goto post_send;
1034 		}
1035 
1036 		ret = isert_login_post_recv(isert_conn);
1037 		if (ret)
1038 			return ret;
1039 	}
1040 post_send:
1041 	ret = isert_login_post_send(isert_conn, tx_desc);
1042 	if (ret)
1043 		return ret;
1044 
1045 	return 0;
1046 }
1047 
1048 static void
1049 isert_rx_login_req(struct isert_conn *isert_conn)
1050 {
1051 	struct iser_rx_desc *rx_desc = isert_conn->login_req_buf;
1052 	int rx_buflen = isert_conn->login_req_len;
1053 	struct iscsi_conn *conn = isert_conn->conn;
1054 	struct iscsi_login *login = conn->conn_login;
1055 	int size;
1056 
1057 	isert_info("conn %p\n", isert_conn);
1058 
1059 	WARN_ON_ONCE(!login);
1060 
1061 	if (login->first_request) {
1062 		struct iscsi_login_req *login_req =
1063 			(struct iscsi_login_req *)&rx_desc->iscsi_header;
1064 		/*
1065 		 * Setup the initial iscsi_login values from the leading
1066 		 * login request PDU.
1067 		 */
1068 		login->leading_connection = (!login_req->tsih) ? 1 : 0;
1069 		login->current_stage =
1070 			(login_req->flags & ISCSI_FLAG_LOGIN_CURRENT_STAGE_MASK)
1071 			 >> 2;
1072 		login->version_min	= login_req->min_version;
1073 		login->version_max	= login_req->max_version;
1074 		memcpy(login->isid, login_req->isid, 6);
1075 		login->cmd_sn		= be32_to_cpu(login_req->cmdsn);
1076 		login->init_task_tag	= login_req->itt;
1077 		login->initial_exp_statsn = be32_to_cpu(login_req->exp_statsn);
1078 		login->cid		= be16_to_cpu(login_req->cid);
1079 		login->tsih		= be16_to_cpu(login_req->tsih);
1080 	}
1081 
1082 	memcpy(&login->req[0], (void *)&rx_desc->iscsi_header, ISCSI_HDR_LEN);
1083 
1084 	size = min(rx_buflen, MAX_KEY_VALUE_PAIRS);
1085 	isert_dbg("Using login payload size: %d, rx_buflen: %d "
1086 		  "MAX_KEY_VALUE_PAIRS: %d\n", size, rx_buflen,
1087 		  MAX_KEY_VALUE_PAIRS);
1088 	memcpy(login->req_buf, &rx_desc->data[0], size);
1089 
1090 	if (login->first_request) {
1091 		complete(&isert_conn->login_comp);
1092 		return;
1093 	}
1094 	schedule_delayed_work(&conn->login_work, 0);
1095 }
1096 
1097 static struct iscsi_cmd
1098 *isert_allocate_cmd(struct iscsi_conn *conn, struct iser_rx_desc *rx_desc)
1099 {
1100 	struct isert_conn *isert_conn = conn->context;
1101 	struct isert_cmd *isert_cmd;
1102 	struct iscsi_cmd *cmd;
1103 
1104 	cmd = iscsit_allocate_cmd(conn, TASK_INTERRUPTIBLE);
1105 	if (!cmd) {
1106 		isert_err("Unable to allocate iscsi_cmd + isert_cmd\n");
1107 		return NULL;
1108 	}
1109 	isert_cmd = iscsit_priv_cmd(cmd);
1110 	isert_cmd->conn = isert_conn;
1111 	isert_cmd->iscsi_cmd = cmd;
1112 	isert_cmd->rx_desc = rx_desc;
1113 
1114 	return cmd;
1115 }
1116 
1117 static int
1118 isert_handle_scsi_cmd(struct isert_conn *isert_conn,
1119 		      struct isert_cmd *isert_cmd, struct iscsi_cmd *cmd,
1120 		      struct iser_rx_desc *rx_desc, unsigned char *buf)
1121 {
1122 	struct iscsi_conn *conn = isert_conn->conn;
1123 	struct iscsi_scsi_req *hdr = (struct iscsi_scsi_req *)buf;
1124 	int imm_data, imm_data_len, unsol_data, sg_nents, rc;
1125 	bool dump_payload = false;
1126 	unsigned int data_len;
1127 
1128 	rc = iscsit_setup_scsi_cmd(conn, cmd, buf);
1129 	if (rc < 0)
1130 		return rc;
1131 
1132 	imm_data = cmd->immediate_data;
1133 	imm_data_len = cmd->first_burst_len;
1134 	unsol_data = cmd->unsolicited_data;
1135 	data_len = cmd->se_cmd.data_length;
1136 
1137 	if (imm_data && imm_data_len == data_len)
1138 		cmd->se_cmd.se_cmd_flags |= SCF_PASSTHROUGH_SG_TO_MEM_NOALLOC;
1139 	rc = iscsit_process_scsi_cmd(conn, cmd, hdr);
1140 	if (rc < 0) {
1141 		return 0;
1142 	} else if (rc > 0) {
1143 		dump_payload = true;
1144 		goto sequence_cmd;
1145 	}
1146 
1147 	if (!imm_data)
1148 		return 0;
1149 
1150 	if (imm_data_len != data_len) {
1151 		sg_nents = max(1UL, DIV_ROUND_UP(imm_data_len, PAGE_SIZE));
1152 		sg_copy_from_buffer(cmd->se_cmd.t_data_sg, sg_nents,
1153 				    &rx_desc->data[0], imm_data_len);
1154 		isert_dbg("Copy Immediate sg_nents: %u imm_data_len: %d\n",
1155 			  sg_nents, imm_data_len);
1156 	} else {
1157 		sg_init_table(&isert_cmd->sg, 1);
1158 		cmd->se_cmd.t_data_sg = &isert_cmd->sg;
1159 		cmd->se_cmd.t_data_nents = 1;
1160 		sg_set_buf(&isert_cmd->sg, &rx_desc->data[0], imm_data_len);
1161 		isert_dbg("Transfer Immediate imm_data_len: %d\n",
1162 			  imm_data_len);
1163 	}
1164 
1165 	cmd->write_data_done += imm_data_len;
1166 
1167 	if (cmd->write_data_done == cmd->se_cmd.data_length) {
1168 		spin_lock_bh(&cmd->istate_lock);
1169 		cmd->cmd_flags |= ICF_GOT_LAST_DATAOUT;
1170 		cmd->i_state = ISTATE_RECEIVED_LAST_DATAOUT;
1171 		spin_unlock_bh(&cmd->istate_lock);
1172 	}
1173 
1174 sequence_cmd:
1175 	rc = iscsit_sequence_cmd(conn, cmd, buf, hdr->cmdsn);
1176 
1177 	if (!rc && dump_payload == false && unsol_data)
1178 		iscsit_set_unsoliticed_dataout(cmd);
1179 	else if (dump_payload && imm_data)
1180 		target_put_sess_cmd(&cmd->se_cmd);
1181 
1182 	return 0;
1183 }
1184 
1185 static int
1186 isert_handle_iscsi_dataout(struct isert_conn *isert_conn,
1187 			   struct iser_rx_desc *rx_desc, unsigned char *buf)
1188 {
1189 	struct scatterlist *sg_start;
1190 	struct iscsi_conn *conn = isert_conn->conn;
1191 	struct iscsi_cmd *cmd = NULL;
1192 	struct iscsi_data *hdr = (struct iscsi_data *)buf;
1193 	u32 unsol_data_len = ntoh24(hdr->dlength);
1194 	int rc, sg_nents, sg_off, page_off;
1195 
1196 	rc = iscsit_check_dataout_hdr(conn, buf, &cmd);
1197 	if (rc < 0)
1198 		return rc;
1199 	else if (!cmd)
1200 		return 0;
1201 	/*
1202 	 * FIXME: Unexpected unsolicited_data out
1203 	 */
1204 	if (!cmd->unsolicited_data) {
1205 		isert_err("Received unexpected solicited data payload\n");
1206 		dump_stack();
1207 		return -1;
1208 	}
1209 
1210 	isert_dbg("Unsolicited DataOut unsol_data_len: %u, "
1211 		  "write_data_done: %u, data_length: %u\n",
1212 		  unsol_data_len,  cmd->write_data_done,
1213 		  cmd->se_cmd.data_length);
1214 
1215 	sg_off = cmd->write_data_done / PAGE_SIZE;
1216 	sg_start = &cmd->se_cmd.t_data_sg[sg_off];
1217 	sg_nents = max(1UL, DIV_ROUND_UP(unsol_data_len, PAGE_SIZE));
1218 	page_off = cmd->write_data_done % PAGE_SIZE;
1219 	/*
1220 	 * FIXME: Non page-aligned unsolicited_data out
1221 	 */
1222 	if (page_off) {
1223 		isert_err("unexpected non-page aligned data payload\n");
1224 		dump_stack();
1225 		return -1;
1226 	}
1227 	isert_dbg("Copying DataOut: sg_start: %p, sg_off: %u "
1228 		  "sg_nents: %u from %p %u\n", sg_start, sg_off,
1229 		  sg_nents, &rx_desc->data[0], unsol_data_len);
1230 
1231 	sg_copy_from_buffer(sg_start, sg_nents, &rx_desc->data[0],
1232 			    unsol_data_len);
1233 
1234 	rc = iscsit_check_dataout_payload(cmd, hdr, false);
1235 	if (rc < 0)
1236 		return rc;
1237 
1238 	/*
1239 	 * multiple data-outs on the same command can arrive -
1240 	 * so post the buffer before hand
1241 	 */
1242 	rc = isert_post_recv(isert_conn, rx_desc);
1243 	if (rc) {
1244 		isert_err("ib_post_recv failed with %d\n", rc);
1245 		return rc;
1246 	}
1247 	return 0;
1248 }
1249 
1250 static int
1251 isert_handle_nop_out(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
1252 		     struct iscsi_cmd *cmd, struct iser_rx_desc *rx_desc,
1253 		     unsigned char *buf)
1254 {
1255 	struct iscsi_conn *conn = isert_conn->conn;
1256 	struct iscsi_nopout *hdr = (struct iscsi_nopout *)buf;
1257 	int rc;
1258 
1259 	rc = iscsit_setup_nop_out(conn, cmd, hdr);
1260 	if (rc < 0)
1261 		return rc;
1262 	/*
1263 	 * FIXME: Add support for NOPOUT payload using unsolicited RDMA payload
1264 	 */
1265 
1266 	return iscsit_process_nop_out(conn, cmd, hdr);
1267 }
1268 
1269 static int
1270 isert_handle_text_cmd(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
1271 		      struct iscsi_cmd *cmd, struct iser_rx_desc *rx_desc,
1272 		      struct iscsi_text *hdr)
1273 {
1274 	struct iscsi_conn *conn = isert_conn->conn;
1275 	u32 payload_length = ntoh24(hdr->dlength);
1276 	int rc;
1277 	unsigned char *text_in = NULL;
1278 
1279 	rc = iscsit_setup_text_cmd(conn, cmd, hdr);
1280 	if (rc < 0)
1281 		return rc;
1282 
1283 	if (payload_length) {
1284 		text_in = kzalloc(payload_length, GFP_KERNEL);
1285 		if (!text_in)
1286 			return -ENOMEM;
1287 	}
1288 	cmd->text_in_ptr = text_in;
1289 
1290 	memcpy(cmd->text_in_ptr, &rx_desc->data[0], payload_length);
1291 
1292 	return iscsit_process_text_cmd(conn, cmd, hdr);
1293 }
1294 
1295 static int
1296 isert_rx_opcode(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc,
1297 		uint32_t read_stag, uint64_t read_va,
1298 		uint32_t write_stag, uint64_t write_va)
1299 {
1300 	struct iscsi_hdr *hdr = &rx_desc->iscsi_header;
1301 	struct iscsi_conn *conn = isert_conn->conn;
1302 	struct iscsi_cmd *cmd;
1303 	struct isert_cmd *isert_cmd;
1304 	int ret = -EINVAL;
1305 	u8 opcode = (hdr->opcode & ISCSI_OPCODE_MASK);
1306 
1307 	if (conn->sess->sess_ops->SessionType &&
1308 	   (!(opcode & ISCSI_OP_TEXT) || !(opcode & ISCSI_OP_LOGOUT))) {
1309 		isert_err("Got illegal opcode: 0x%02x in SessionType=Discovery,"
1310 			  " ignoring\n", opcode);
1311 		return 0;
1312 	}
1313 
1314 	switch (opcode) {
1315 	case ISCSI_OP_SCSI_CMD:
1316 		cmd = isert_allocate_cmd(conn, rx_desc);
1317 		if (!cmd)
1318 			break;
1319 
1320 		isert_cmd = iscsit_priv_cmd(cmd);
1321 		isert_cmd->read_stag = read_stag;
1322 		isert_cmd->read_va = read_va;
1323 		isert_cmd->write_stag = write_stag;
1324 		isert_cmd->write_va = write_va;
1325 		isert_cmd->inv_rkey = read_stag ? read_stag : write_stag;
1326 
1327 		ret = isert_handle_scsi_cmd(isert_conn, isert_cmd, cmd,
1328 					rx_desc, (unsigned char *)hdr);
1329 		break;
1330 	case ISCSI_OP_NOOP_OUT:
1331 		cmd = isert_allocate_cmd(conn, rx_desc);
1332 		if (!cmd)
1333 			break;
1334 
1335 		isert_cmd = iscsit_priv_cmd(cmd);
1336 		ret = isert_handle_nop_out(isert_conn, isert_cmd, cmd,
1337 					   rx_desc, (unsigned char *)hdr);
1338 		break;
1339 	case ISCSI_OP_SCSI_DATA_OUT:
1340 		ret = isert_handle_iscsi_dataout(isert_conn, rx_desc,
1341 						(unsigned char *)hdr);
1342 		break;
1343 	case ISCSI_OP_SCSI_TMFUNC:
1344 		cmd = isert_allocate_cmd(conn, rx_desc);
1345 		if (!cmd)
1346 			break;
1347 
1348 		ret = iscsit_handle_task_mgt_cmd(conn, cmd,
1349 						(unsigned char *)hdr);
1350 		break;
1351 	case ISCSI_OP_LOGOUT:
1352 		cmd = isert_allocate_cmd(conn, rx_desc);
1353 		if (!cmd)
1354 			break;
1355 
1356 		ret = iscsit_handle_logout_cmd(conn, cmd, (unsigned char *)hdr);
1357 		break;
1358 	case ISCSI_OP_TEXT:
1359 		if (be32_to_cpu(hdr->ttt) != 0xFFFFFFFF)
1360 			cmd = iscsit_find_cmd_from_itt(conn, hdr->itt);
1361 		else
1362 			cmd = isert_allocate_cmd(conn, rx_desc);
1363 
1364 		if (!cmd)
1365 			break;
1366 
1367 		isert_cmd = iscsit_priv_cmd(cmd);
1368 		ret = isert_handle_text_cmd(isert_conn, isert_cmd, cmd,
1369 					    rx_desc, (struct iscsi_text *)hdr);
1370 		break;
1371 	default:
1372 		isert_err("Got unknown iSCSI OpCode: 0x%02x\n", opcode);
1373 		dump_stack();
1374 		break;
1375 	}
1376 
1377 	return ret;
1378 }
1379 
1380 static void
1381 isert_print_wc(struct ib_wc *wc, const char *type)
1382 {
1383 	if (wc->status != IB_WC_WR_FLUSH_ERR)
1384 		isert_err("%s failure: %s (%d) vend_err %x\n", type,
1385 			  ib_wc_status_msg(wc->status), wc->status,
1386 			  wc->vendor_err);
1387 	else
1388 		isert_dbg("%s failure: %s (%d)\n", type,
1389 			  ib_wc_status_msg(wc->status), wc->status);
1390 }
1391 
1392 static void
1393 isert_recv_done(struct ib_cq *cq, struct ib_wc *wc)
1394 {
1395 	struct isert_conn *isert_conn = wc->qp->qp_context;
1396 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1397 	struct iser_rx_desc *rx_desc = cqe_to_rx_desc(wc->wr_cqe);
1398 	struct iscsi_hdr *hdr = &rx_desc->iscsi_header;
1399 	struct iser_ctrl *iser_ctrl = &rx_desc->iser_header;
1400 	uint64_t read_va = 0, write_va = 0;
1401 	uint32_t read_stag = 0, write_stag = 0;
1402 
1403 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1404 		isert_print_wc(wc, "recv");
1405 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1406 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1407 		return;
1408 	}
1409 
1410 	rx_desc->in_use = true;
1411 
1412 	ib_dma_sync_single_for_cpu(ib_dev, rx_desc->dma_addr,
1413 			ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
1414 
1415 	isert_dbg("DMA: 0x%llx, iSCSI opcode: 0x%02x, ITT: 0x%08x, flags: 0x%02x dlen: %d\n",
1416 		 rx_desc->dma_addr, hdr->opcode, hdr->itt, hdr->flags,
1417 		 (int)(wc->byte_len - ISER_HEADERS_LEN));
1418 
1419 	switch (iser_ctrl->flags & 0xF0) {
1420 	case ISCSI_CTRL:
1421 		if (iser_ctrl->flags & ISER_RSV) {
1422 			read_stag = be32_to_cpu(iser_ctrl->read_stag);
1423 			read_va = be64_to_cpu(iser_ctrl->read_va);
1424 			isert_dbg("ISER_RSV: read_stag: 0x%x read_va: 0x%llx\n",
1425 				  read_stag, (unsigned long long)read_va);
1426 		}
1427 		if (iser_ctrl->flags & ISER_WSV) {
1428 			write_stag = be32_to_cpu(iser_ctrl->write_stag);
1429 			write_va = be64_to_cpu(iser_ctrl->write_va);
1430 			isert_dbg("ISER_WSV: write_stag: 0x%x write_va: 0x%llx\n",
1431 				  write_stag, (unsigned long long)write_va);
1432 		}
1433 
1434 		isert_dbg("ISER ISCSI_CTRL PDU\n");
1435 		break;
1436 	case ISER_HELLO:
1437 		isert_err("iSER Hello message\n");
1438 		break;
1439 	default:
1440 		isert_warn("Unknown iSER hdr flags: 0x%02x\n", iser_ctrl->flags);
1441 		break;
1442 	}
1443 
1444 	isert_rx_opcode(isert_conn, rx_desc,
1445 			read_stag, read_va, write_stag, write_va);
1446 
1447 	ib_dma_sync_single_for_device(ib_dev, rx_desc->dma_addr,
1448 			ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
1449 }
1450 
1451 static void
1452 isert_login_recv_done(struct ib_cq *cq, struct ib_wc *wc)
1453 {
1454 	struct isert_conn *isert_conn = wc->qp->qp_context;
1455 	struct ib_device *ib_dev = isert_conn->device->ib_device;
1456 
1457 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1458 		isert_print_wc(wc, "login recv");
1459 		return;
1460 	}
1461 
1462 	ib_dma_sync_single_for_cpu(ib_dev, isert_conn->login_req_dma,
1463 			ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
1464 
1465 	isert_conn->login_req_len = wc->byte_len - ISER_HEADERS_LEN;
1466 
1467 	if (isert_conn->conn) {
1468 		struct iscsi_login *login = isert_conn->conn->conn_login;
1469 
1470 		if (login && !login->first_request)
1471 			isert_rx_login_req(isert_conn);
1472 	}
1473 
1474 	mutex_lock(&isert_conn->mutex);
1475 	complete(&isert_conn->login_req_comp);
1476 	mutex_unlock(&isert_conn->mutex);
1477 
1478 	ib_dma_sync_single_for_device(ib_dev, isert_conn->login_req_dma,
1479 				ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
1480 }
1481 
1482 static void
1483 isert_rdma_rw_ctx_destroy(struct isert_cmd *cmd, struct isert_conn *conn)
1484 {
1485 	struct se_cmd *se_cmd = &cmd->iscsi_cmd->se_cmd;
1486 	enum dma_data_direction dir = target_reverse_dma_direction(se_cmd);
1487 
1488 	if (!cmd->rw.nr_ops)
1489 		return;
1490 
1491 	if (isert_prot_cmd(conn, se_cmd)) {
1492 		rdma_rw_ctx_destroy_signature(&cmd->rw, conn->qp,
1493 				conn->cm_id->port_num, se_cmd->t_data_sg,
1494 				se_cmd->t_data_nents, se_cmd->t_prot_sg,
1495 				se_cmd->t_prot_nents, dir);
1496 	} else {
1497 		rdma_rw_ctx_destroy(&cmd->rw, conn->qp, conn->cm_id->port_num,
1498 				se_cmd->t_data_sg, se_cmd->t_data_nents, dir);
1499 	}
1500 
1501 	cmd->rw.nr_ops = 0;
1502 }
1503 
1504 static void
1505 isert_put_cmd(struct isert_cmd *isert_cmd, bool comp_err)
1506 {
1507 	struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd;
1508 	struct isert_conn *isert_conn = isert_cmd->conn;
1509 	struct iscsi_conn *conn = isert_conn->conn;
1510 	struct iscsi_text_rsp *hdr;
1511 
1512 	isert_dbg("Cmd %p\n", isert_cmd);
1513 
1514 	switch (cmd->iscsi_opcode) {
1515 	case ISCSI_OP_SCSI_CMD:
1516 		spin_lock_bh(&conn->cmd_lock);
1517 		if (!list_empty(&cmd->i_conn_node))
1518 			list_del_init(&cmd->i_conn_node);
1519 		spin_unlock_bh(&conn->cmd_lock);
1520 
1521 		if (cmd->data_direction == DMA_TO_DEVICE) {
1522 			iscsit_stop_dataout_timer(cmd);
1523 			/*
1524 			 * Check for special case during comp_err where
1525 			 * WRITE_PENDING has been handed off from core,
1526 			 * but requires an extra target_put_sess_cmd()
1527 			 * before transport_generic_free_cmd() below.
1528 			 */
1529 			if (comp_err &&
1530 			    cmd->se_cmd.t_state == TRANSPORT_WRITE_PENDING) {
1531 				struct se_cmd *se_cmd = &cmd->se_cmd;
1532 
1533 				target_put_sess_cmd(se_cmd);
1534 			}
1535 		}
1536 
1537 		isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1538 		transport_generic_free_cmd(&cmd->se_cmd, 0);
1539 		break;
1540 	case ISCSI_OP_SCSI_TMFUNC:
1541 		spin_lock_bh(&conn->cmd_lock);
1542 		if (!list_empty(&cmd->i_conn_node))
1543 			list_del_init(&cmd->i_conn_node);
1544 		spin_unlock_bh(&conn->cmd_lock);
1545 
1546 		transport_generic_free_cmd(&cmd->se_cmd, 0);
1547 		break;
1548 	case ISCSI_OP_REJECT:
1549 	case ISCSI_OP_NOOP_OUT:
1550 	case ISCSI_OP_TEXT:
1551 		hdr = (struct iscsi_text_rsp *)&isert_cmd->tx_desc.iscsi_header;
1552 		/* If the continue bit is on, keep the command alive */
1553 		if (hdr->flags & ISCSI_FLAG_TEXT_CONTINUE)
1554 			break;
1555 
1556 		spin_lock_bh(&conn->cmd_lock);
1557 		if (!list_empty(&cmd->i_conn_node))
1558 			list_del_init(&cmd->i_conn_node);
1559 		spin_unlock_bh(&conn->cmd_lock);
1560 
1561 		/*
1562 		 * Handle special case for REJECT when iscsi_add_reject*() has
1563 		 * overwritten the original iscsi_opcode assignment, and the
1564 		 * associated cmd->se_cmd needs to be released.
1565 		 */
1566 		if (cmd->se_cmd.se_tfo != NULL) {
1567 			isert_dbg("Calling transport_generic_free_cmd for 0x%02x\n",
1568 				 cmd->iscsi_opcode);
1569 			transport_generic_free_cmd(&cmd->se_cmd, 0);
1570 			break;
1571 		}
1572 		/*
1573 		 * Fall-through
1574 		 */
1575 	default:
1576 		iscsit_release_cmd(cmd);
1577 		break;
1578 	}
1579 }
1580 
1581 static void
1582 isert_unmap_tx_desc(struct iser_tx_desc *tx_desc, struct ib_device *ib_dev)
1583 {
1584 	if (tx_desc->dma_addr != 0) {
1585 		isert_dbg("unmap single for tx_desc->dma_addr\n");
1586 		ib_dma_unmap_single(ib_dev, tx_desc->dma_addr,
1587 				    ISER_HEADERS_LEN, DMA_TO_DEVICE);
1588 		tx_desc->dma_addr = 0;
1589 	}
1590 }
1591 
1592 static void
1593 isert_completion_put(struct iser_tx_desc *tx_desc, struct isert_cmd *isert_cmd,
1594 		     struct ib_device *ib_dev, bool comp_err)
1595 {
1596 	if (isert_cmd->pdu_buf_dma != 0) {
1597 		isert_dbg("unmap single for isert_cmd->pdu_buf_dma\n");
1598 		ib_dma_unmap_single(ib_dev, isert_cmd->pdu_buf_dma,
1599 				    isert_cmd->pdu_buf_len, DMA_TO_DEVICE);
1600 		isert_cmd->pdu_buf_dma = 0;
1601 	}
1602 
1603 	isert_unmap_tx_desc(tx_desc, ib_dev);
1604 	isert_put_cmd(isert_cmd, comp_err);
1605 }
1606 
1607 static int
1608 isert_check_pi_status(struct se_cmd *se_cmd, struct ib_mr *sig_mr)
1609 {
1610 	struct ib_mr_status mr_status;
1611 	int ret;
1612 
1613 	ret = ib_check_mr_status(sig_mr, IB_MR_CHECK_SIG_STATUS, &mr_status);
1614 	if (ret) {
1615 		isert_err("ib_check_mr_status failed, ret %d\n", ret);
1616 		goto fail_mr_status;
1617 	}
1618 
1619 	if (mr_status.fail_status & IB_MR_CHECK_SIG_STATUS) {
1620 		u64 sec_offset_err;
1621 		u32 block_size = se_cmd->se_dev->dev_attrib.block_size + 8;
1622 
1623 		switch (mr_status.sig_err.err_type) {
1624 		case IB_SIG_BAD_GUARD:
1625 			se_cmd->pi_err = TCM_LOGICAL_BLOCK_GUARD_CHECK_FAILED;
1626 			break;
1627 		case IB_SIG_BAD_REFTAG:
1628 			se_cmd->pi_err = TCM_LOGICAL_BLOCK_REF_TAG_CHECK_FAILED;
1629 			break;
1630 		case IB_SIG_BAD_APPTAG:
1631 			se_cmd->pi_err = TCM_LOGICAL_BLOCK_APP_TAG_CHECK_FAILED;
1632 			break;
1633 		}
1634 		sec_offset_err = mr_status.sig_err.sig_err_offset;
1635 		do_div(sec_offset_err, block_size);
1636 		se_cmd->bad_sector = sec_offset_err + se_cmd->t_task_lba;
1637 
1638 		isert_err("PI error found type %d at sector 0x%llx "
1639 			  "expected 0x%x vs actual 0x%x\n",
1640 			  mr_status.sig_err.err_type,
1641 			  (unsigned long long)se_cmd->bad_sector,
1642 			  mr_status.sig_err.expected,
1643 			  mr_status.sig_err.actual);
1644 		ret = 1;
1645 	}
1646 
1647 fail_mr_status:
1648 	return ret;
1649 }
1650 
1651 static void
1652 isert_rdma_write_done(struct ib_cq *cq, struct ib_wc *wc)
1653 {
1654 	struct isert_conn *isert_conn = wc->qp->qp_context;
1655 	struct isert_device *device = isert_conn->device;
1656 	struct iser_tx_desc *desc = cqe_to_tx_desc(wc->wr_cqe);
1657 	struct isert_cmd *isert_cmd = tx_desc_to_cmd(desc);
1658 	struct se_cmd *cmd = &isert_cmd->iscsi_cmd->se_cmd;
1659 	int ret = 0;
1660 
1661 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1662 		isert_print_wc(wc, "rdma write");
1663 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1664 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1665 		isert_completion_put(desc, isert_cmd, device->ib_device, true);
1666 		return;
1667 	}
1668 
1669 	isert_dbg("Cmd %p\n", isert_cmd);
1670 
1671 	ret = isert_check_pi_status(cmd, isert_cmd->rw.sig->sig_mr);
1672 	isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1673 
1674 	if (ret) {
1675 		/*
1676 		 * transport_generic_request_failure() expects to have
1677 		 * plus two references to handle queue-full, so re-add
1678 		 * one here as target-core will have already dropped
1679 		 * it after the first isert_put_datain() callback.
1680 		 */
1681 		kref_get(&cmd->cmd_kref);
1682 		transport_generic_request_failure(cmd, cmd->pi_err);
1683 	} else {
1684 		/*
1685 		 * XXX: isert_put_response() failure is not retried.
1686 		 */
1687 		ret = isert_put_response(isert_conn->conn, isert_cmd->iscsi_cmd);
1688 		if (ret)
1689 			pr_warn_ratelimited("isert_put_response() ret: %d\n", ret);
1690 	}
1691 }
1692 
1693 static void
1694 isert_rdma_read_done(struct ib_cq *cq, struct ib_wc *wc)
1695 {
1696 	struct isert_conn *isert_conn = wc->qp->qp_context;
1697 	struct isert_device *device = isert_conn->device;
1698 	struct iser_tx_desc *desc = cqe_to_tx_desc(wc->wr_cqe);
1699 	struct isert_cmd *isert_cmd = tx_desc_to_cmd(desc);
1700 	struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd;
1701 	struct se_cmd *se_cmd = &cmd->se_cmd;
1702 	int ret = 0;
1703 
1704 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1705 		isert_print_wc(wc, "rdma read");
1706 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1707 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1708 		isert_completion_put(desc, isert_cmd, device->ib_device, true);
1709 		return;
1710 	}
1711 
1712 	isert_dbg("Cmd %p\n", isert_cmd);
1713 
1714 	iscsit_stop_dataout_timer(cmd);
1715 
1716 	if (isert_prot_cmd(isert_conn, se_cmd))
1717 		ret = isert_check_pi_status(se_cmd, isert_cmd->rw.sig->sig_mr);
1718 	isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1719 	cmd->write_data_done = 0;
1720 
1721 	isert_dbg("Cmd: %p RDMA_READ comp calling execute_cmd\n", isert_cmd);
1722 	spin_lock_bh(&cmd->istate_lock);
1723 	cmd->cmd_flags |= ICF_GOT_LAST_DATAOUT;
1724 	cmd->i_state = ISTATE_RECEIVED_LAST_DATAOUT;
1725 	spin_unlock_bh(&cmd->istate_lock);
1726 
1727 	/*
1728 	 * transport_generic_request_failure() will drop the extra
1729 	 * se_cmd->cmd_kref reference after T10-PI error, and handle
1730 	 * any non-zero ->queue_status() callback error retries.
1731 	 */
1732 	if (ret)
1733 		transport_generic_request_failure(se_cmd, se_cmd->pi_err);
1734 	else
1735 		target_execute_cmd(se_cmd);
1736 }
1737 
1738 static void
1739 isert_do_control_comp(struct work_struct *work)
1740 {
1741 	struct isert_cmd *isert_cmd = container_of(work,
1742 			struct isert_cmd, comp_work);
1743 	struct isert_conn *isert_conn = isert_cmd->conn;
1744 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1745 	struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd;
1746 
1747 	isert_dbg("Cmd %p i_state %d\n", isert_cmd, cmd->i_state);
1748 
1749 	switch (cmd->i_state) {
1750 	case ISTATE_SEND_TASKMGTRSP:
1751 		iscsit_tmr_post_handler(cmd, cmd->conn);
1752 	case ISTATE_SEND_REJECT:   /* FALLTHRU */
1753 	case ISTATE_SEND_TEXTRSP:  /* FALLTHRU */
1754 		cmd->i_state = ISTATE_SENT_STATUS;
1755 		isert_completion_put(&isert_cmd->tx_desc, isert_cmd,
1756 				     ib_dev, false);
1757 		break;
1758 	case ISTATE_SEND_LOGOUTRSP:
1759 		iscsit_logout_post_handler(cmd, cmd->conn);
1760 		break;
1761 	default:
1762 		isert_err("Unknown i_state %d\n", cmd->i_state);
1763 		dump_stack();
1764 		break;
1765 	}
1766 }
1767 
1768 static void
1769 isert_login_send_done(struct ib_cq *cq, struct ib_wc *wc)
1770 {
1771 	struct isert_conn *isert_conn = wc->qp->qp_context;
1772 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1773 	struct iser_tx_desc *tx_desc = cqe_to_tx_desc(wc->wr_cqe);
1774 
1775 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1776 		isert_print_wc(wc, "login send");
1777 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1778 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1779 	}
1780 
1781 	isert_unmap_tx_desc(tx_desc, ib_dev);
1782 }
1783 
1784 static void
1785 isert_send_done(struct ib_cq *cq, struct ib_wc *wc)
1786 {
1787 	struct isert_conn *isert_conn = wc->qp->qp_context;
1788 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1789 	struct iser_tx_desc *tx_desc = cqe_to_tx_desc(wc->wr_cqe);
1790 	struct isert_cmd *isert_cmd = tx_desc_to_cmd(tx_desc);
1791 
1792 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1793 		isert_print_wc(wc, "send");
1794 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1795 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1796 		isert_completion_put(tx_desc, isert_cmd, ib_dev, true);
1797 		return;
1798 	}
1799 
1800 	isert_dbg("Cmd %p\n", isert_cmd);
1801 
1802 	switch (isert_cmd->iscsi_cmd->i_state) {
1803 	case ISTATE_SEND_TASKMGTRSP:
1804 	case ISTATE_SEND_LOGOUTRSP:
1805 	case ISTATE_SEND_REJECT:
1806 	case ISTATE_SEND_TEXTRSP:
1807 		isert_unmap_tx_desc(tx_desc, ib_dev);
1808 
1809 		INIT_WORK(&isert_cmd->comp_work, isert_do_control_comp);
1810 		queue_work(isert_comp_wq, &isert_cmd->comp_work);
1811 		return;
1812 	default:
1813 		isert_cmd->iscsi_cmd->i_state = ISTATE_SENT_STATUS;
1814 		isert_completion_put(tx_desc, isert_cmd, ib_dev, false);
1815 		break;
1816 	}
1817 }
1818 
1819 static int
1820 isert_post_response(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd)
1821 {
1822 	struct ib_send_wr *wr_failed;
1823 	int ret;
1824 
1825 	ret = isert_post_recv(isert_conn, isert_cmd->rx_desc);
1826 	if (ret) {
1827 		isert_err("ib_post_recv failed with %d\n", ret);
1828 		return ret;
1829 	}
1830 
1831 	ret = ib_post_send(isert_conn->qp, &isert_cmd->tx_desc.send_wr,
1832 			   &wr_failed);
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_attrs *sig_attrs,
2063 		     struct ib_sig_domain *domain)
2064 {
2065 	domain->sig_type = IB_SIG_TYPE_T10_DIF;
2066 	domain->sig.dif.bg_type = IB_T10DIF_CRC;
2067 	domain->sig.dif.pi_interval = se_cmd->se_dev->dev_attrib.block_size;
2068 	domain->sig.dif.ref_tag = se_cmd->reftag_seed;
2069 	/*
2070 	 * At the moment we hard code those, but if in the future
2071 	 * the target core would like to use it, we will take it
2072 	 * from se_cmd.
2073 	 */
2074 	domain->sig.dif.apptag_check_mask = 0xffff;
2075 	domain->sig.dif.app_escape = true;
2076 	domain->sig.dif.ref_escape = true;
2077 	if (se_cmd->prot_type == TARGET_DIF_TYPE1_PROT ||
2078 	    se_cmd->prot_type == TARGET_DIF_TYPE2_PROT)
2079 		domain->sig.dif.ref_remap = true;
2080 };
2081 
2082 static int
2083 isert_set_sig_attrs(struct se_cmd *se_cmd, struct ib_sig_attrs *sig_attrs)
2084 {
2085 	memset(sig_attrs, 0, sizeof(*sig_attrs));
2086 
2087 	switch (se_cmd->prot_op) {
2088 	case TARGET_PROT_DIN_INSERT:
2089 	case TARGET_PROT_DOUT_STRIP:
2090 		sig_attrs->mem.sig_type = IB_SIG_TYPE_NONE;
2091 		isert_set_dif_domain(se_cmd, sig_attrs, &sig_attrs->wire);
2092 		break;
2093 	case TARGET_PROT_DOUT_INSERT:
2094 	case TARGET_PROT_DIN_STRIP:
2095 		sig_attrs->wire.sig_type = IB_SIG_TYPE_NONE;
2096 		isert_set_dif_domain(se_cmd, sig_attrs, &sig_attrs->mem);
2097 		break;
2098 	case TARGET_PROT_DIN_PASS:
2099 	case TARGET_PROT_DOUT_PASS:
2100 		isert_set_dif_domain(se_cmd, sig_attrs, &sig_attrs->wire);
2101 		isert_set_dif_domain(se_cmd, sig_attrs, &sig_attrs->mem);
2102 		break;
2103 	default:
2104 		isert_err("Unsupported PI operation %d\n", se_cmd->prot_op);
2105 		return -EINVAL;
2106 	}
2107 
2108 	sig_attrs->check_mask =
2109 	       (se_cmd->prot_checks & TARGET_DIF_CHECK_GUARD  ? 0xc0 : 0) |
2110 	       (se_cmd->prot_checks & TARGET_DIF_CHECK_REFTAG ? 0x30 : 0) |
2111 	       (se_cmd->prot_checks & TARGET_DIF_CHECK_REFTAG ? 0x0f : 0);
2112 	return 0;
2113 }
2114 
2115 static int
2116 isert_rdma_rw_ctx_post(struct isert_cmd *cmd, struct isert_conn *conn,
2117 		struct ib_cqe *cqe, struct ib_send_wr *chain_wr)
2118 {
2119 	struct se_cmd *se_cmd = &cmd->iscsi_cmd->se_cmd;
2120 	enum dma_data_direction dir = target_reverse_dma_direction(se_cmd);
2121 	u8 port_num = conn->cm_id->port_num;
2122 	u64 addr;
2123 	u32 rkey, offset;
2124 	int ret;
2125 
2126 	if (dir == DMA_FROM_DEVICE) {
2127 		addr = cmd->write_va;
2128 		rkey = cmd->write_stag;
2129 		offset = cmd->iscsi_cmd->write_data_done;
2130 	} else {
2131 		addr = cmd->read_va;
2132 		rkey = cmd->read_stag;
2133 		offset = 0;
2134 	}
2135 
2136 	if (isert_prot_cmd(conn, se_cmd)) {
2137 		struct ib_sig_attrs sig_attrs;
2138 
2139 		ret = isert_set_sig_attrs(se_cmd, &sig_attrs);
2140 		if (ret)
2141 			return ret;
2142 
2143 		WARN_ON_ONCE(offset);
2144 		ret = rdma_rw_ctx_signature_init(&cmd->rw, conn->qp, port_num,
2145 				se_cmd->t_data_sg, se_cmd->t_data_nents,
2146 				se_cmd->t_prot_sg, se_cmd->t_prot_nents,
2147 				&sig_attrs, addr, rkey, dir);
2148 	} else {
2149 		ret = rdma_rw_ctx_init(&cmd->rw, conn->qp, port_num,
2150 				se_cmd->t_data_sg, se_cmd->t_data_nents,
2151 				offset, addr, rkey, dir);
2152 	}
2153 	if (ret < 0) {
2154 		isert_err("Cmd: %p failed to prepare RDMA res\n", cmd);
2155 		return ret;
2156 	}
2157 
2158 	ret = rdma_rw_ctx_post(&cmd->rw, conn->qp, port_num, cqe, chain_wr);
2159 	if (ret < 0)
2160 		isert_err("Cmd: %p failed to post RDMA res\n", cmd);
2161 	return ret;
2162 }
2163 
2164 static int
2165 isert_put_datain(struct iscsi_conn *conn, struct iscsi_cmd *cmd)
2166 {
2167 	struct se_cmd *se_cmd = &cmd->se_cmd;
2168 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2169 	struct isert_conn *isert_conn = conn->context;
2170 	struct ib_cqe *cqe = NULL;
2171 	struct ib_send_wr *chain_wr = NULL;
2172 	int rc;
2173 
2174 	isert_dbg("Cmd: %p RDMA_WRITE data_length: %u\n",
2175 		 isert_cmd, se_cmd->data_length);
2176 
2177 	if (isert_prot_cmd(isert_conn, se_cmd)) {
2178 		isert_cmd->tx_desc.tx_cqe.done = isert_rdma_write_done;
2179 		cqe = &isert_cmd->tx_desc.tx_cqe;
2180 	} else {
2181 		/*
2182 		 * Build isert_conn->tx_desc for iSCSI response PDU and attach
2183 		 */
2184 		isert_create_send_desc(isert_conn, isert_cmd,
2185 				       &isert_cmd->tx_desc);
2186 		iscsit_build_rsp_pdu(cmd, conn, true, (struct iscsi_scsi_rsp *)
2187 				     &isert_cmd->tx_desc.iscsi_header);
2188 		isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
2189 		isert_init_send_wr(isert_conn, isert_cmd,
2190 				   &isert_cmd->tx_desc.send_wr);
2191 
2192 		rc = isert_post_recv(isert_conn, isert_cmd->rx_desc);
2193 		if (rc) {
2194 			isert_err("ib_post_recv failed with %d\n", rc);
2195 			return rc;
2196 		}
2197 
2198 		chain_wr = &isert_cmd->tx_desc.send_wr;
2199 	}
2200 
2201 	rc = isert_rdma_rw_ctx_post(isert_cmd, isert_conn, cqe, chain_wr);
2202 	isert_dbg("Cmd: %p posted RDMA_WRITE for iSER Data READ rc: %d\n",
2203 		  isert_cmd, rc);
2204 	return rc;
2205 }
2206 
2207 static int
2208 isert_get_dataout(struct iscsi_conn *conn, struct iscsi_cmd *cmd, bool recovery)
2209 {
2210 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2211 	int ret;
2212 
2213 	isert_dbg("Cmd: %p RDMA_READ data_length: %u write_data_done: %u\n",
2214 		 isert_cmd, cmd->se_cmd.data_length, cmd->write_data_done);
2215 
2216 	isert_cmd->tx_desc.tx_cqe.done = isert_rdma_read_done;
2217 	ret = isert_rdma_rw_ctx_post(isert_cmd, conn->context,
2218 				     &isert_cmd->tx_desc.tx_cqe, NULL);
2219 
2220 	isert_dbg("Cmd: %p posted RDMA_READ memory for ISER Data WRITE rc: %d\n",
2221 		 isert_cmd, ret);
2222 	return ret;
2223 }
2224 
2225 static int
2226 isert_immediate_queue(struct iscsi_conn *conn, struct iscsi_cmd *cmd, int state)
2227 {
2228 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2229 	int ret = 0;
2230 
2231 	switch (state) {
2232 	case ISTATE_REMOVE:
2233 		spin_lock_bh(&conn->cmd_lock);
2234 		list_del_init(&cmd->i_conn_node);
2235 		spin_unlock_bh(&conn->cmd_lock);
2236 		isert_put_cmd(isert_cmd, true);
2237 		break;
2238 	case ISTATE_SEND_NOPIN_WANT_RESPONSE:
2239 		ret = isert_put_nopin(cmd, conn, false);
2240 		break;
2241 	default:
2242 		isert_err("Unknown immediate state: 0x%02x\n", state);
2243 		ret = -EINVAL;
2244 		break;
2245 	}
2246 
2247 	return ret;
2248 }
2249 
2250 static int
2251 isert_response_queue(struct iscsi_conn *conn, struct iscsi_cmd *cmd, int state)
2252 {
2253 	struct isert_conn *isert_conn = conn->context;
2254 	int ret;
2255 
2256 	switch (state) {
2257 	case ISTATE_SEND_LOGOUTRSP:
2258 		ret = isert_put_logout_rsp(cmd, conn);
2259 		if (!ret)
2260 			isert_conn->logout_posted = true;
2261 		break;
2262 	case ISTATE_SEND_NOPIN:
2263 		ret = isert_put_nopin(cmd, conn, true);
2264 		break;
2265 	case ISTATE_SEND_TASKMGTRSP:
2266 		ret = isert_put_tm_rsp(cmd, conn);
2267 		break;
2268 	case ISTATE_SEND_REJECT:
2269 		ret = isert_put_reject(cmd, conn);
2270 		break;
2271 	case ISTATE_SEND_TEXTRSP:
2272 		ret = isert_put_text_rsp(cmd, conn);
2273 		break;
2274 	case ISTATE_SEND_STATUS:
2275 		/*
2276 		 * Special case for sending non GOOD SCSI status from TX thread
2277 		 * context during pre se_cmd excecution failure.
2278 		 */
2279 		ret = isert_put_response(conn, cmd);
2280 		break;
2281 	default:
2282 		isert_err("Unknown response state: 0x%02x\n", state);
2283 		ret = -EINVAL;
2284 		break;
2285 	}
2286 
2287 	return ret;
2288 }
2289 
2290 struct rdma_cm_id *
2291 isert_setup_id(struct isert_np *isert_np)
2292 {
2293 	struct iscsi_np *np = isert_np->np;
2294 	struct rdma_cm_id *id;
2295 	struct sockaddr *sa;
2296 	int ret;
2297 
2298 	sa = (struct sockaddr *)&np->np_sockaddr;
2299 	isert_dbg("ksockaddr: %p, sa: %p\n", &np->np_sockaddr, sa);
2300 
2301 	id = rdma_create_id(&init_net, isert_cma_handler, isert_np,
2302 			    RDMA_PS_TCP, IB_QPT_RC);
2303 	if (IS_ERR(id)) {
2304 		isert_err("rdma_create_id() failed: %ld\n", PTR_ERR(id));
2305 		ret = PTR_ERR(id);
2306 		goto out;
2307 	}
2308 	isert_dbg("id %p context %p\n", id, id->context);
2309 
2310 	ret = rdma_bind_addr(id, sa);
2311 	if (ret) {
2312 		isert_err("rdma_bind_addr() failed: %d\n", ret);
2313 		goto out_id;
2314 	}
2315 
2316 	ret = rdma_listen(id, 0);
2317 	if (ret) {
2318 		isert_err("rdma_listen() failed: %d\n", ret);
2319 		goto out_id;
2320 	}
2321 
2322 	return id;
2323 out_id:
2324 	rdma_destroy_id(id);
2325 out:
2326 	return ERR_PTR(ret);
2327 }
2328 
2329 static int
2330 isert_setup_np(struct iscsi_np *np,
2331 	       struct sockaddr_storage *ksockaddr)
2332 {
2333 	struct isert_np *isert_np;
2334 	struct rdma_cm_id *isert_lid;
2335 	int ret;
2336 
2337 	isert_np = kzalloc(sizeof(struct isert_np), GFP_KERNEL);
2338 	if (!isert_np)
2339 		return -ENOMEM;
2340 
2341 	sema_init(&isert_np->sem, 0);
2342 	mutex_init(&isert_np->mutex);
2343 	INIT_LIST_HEAD(&isert_np->accepted);
2344 	INIT_LIST_HEAD(&isert_np->pending);
2345 	isert_np->np = np;
2346 
2347 	/*
2348 	 * Setup the np->np_sockaddr from the passed sockaddr setup
2349 	 * in iscsi_target_configfs.c code..
2350 	 */
2351 	memcpy(&np->np_sockaddr, ksockaddr,
2352 	       sizeof(struct sockaddr_storage));
2353 
2354 	isert_lid = isert_setup_id(isert_np);
2355 	if (IS_ERR(isert_lid)) {
2356 		ret = PTR_ERR(isert_lid);
2357 		goto out;
2358 	}
2359 
2360 	isert_np->cm_id = isert_lid;
2361 	np->np_context = isert_np;
2362 
2363 	return 0;
2364 
2365 out:
2366 	kfree(isert_np);
2367 
2368 	return ret;
2369 }
2370 
2371 static int
2372 isert_rdma_accept(struct isert_conn *isert_conn)
2373 {
2374 	struct rdma_cm_id *cm_id = isert_conn->cm_id;
2375 	struct rdma_conn_param cp;
2376 	int ret;
2377 	struct iser_cm_hdr rsp_hdr;
2378 
2379 	memset(&cp, 0, sizeof(struct rdma_conn_param));
2380 	cp.initiator_depth = isert_conn->initiator_depth;
2381 	cp.retry_count = 7;
2382 	cp.rnr_retry_count = 7;
2383 
2384 	memset(&rsp_hdr, 0, sizeof(rsp_hdr));
2385 	rsp_hdr.flags = ISERT_ZBVA_NOT_USED;
2386 	if (!isert_conn->snd_w_inv)
2387 		rsp_hdr.flags = rsp_hdr.flags | ISERT_SEND_W_INV_NOT_USED;
2388 	cp.private_data = (void *)&rsp_hdr;
2389 	cp.private_data_len = sizeof(rsp_hdr);
2390 
2391 	ret = rdma_accept(cm_id, &cp);
2392 	if (ret) {
2393 		isert_err("rdma_accept() failed with: %d\n", ret);
2394 		return ret;
2395 	}
2396 
2397 	return 0;
2398 }
2399 
2400 static int
2401 isert_get_login_rx(struct iscsi_conn *conn, struct iscsi_login *login)
2402 {
2403 	struct isert_conn *isert_conn = conn->context;
2404 	int ret;
2405 
2406 	isert_info("before login_req comp conn: %p\n", isert_conn);
2407 	ret = wait_for_completion_interruptible(&isert_conn->login_req_comp);
2408 	if (ret) {
2409 		isert_err("isert_conn %p interrupted before got login req\n",
2410 			  isert_conn);
2411 		return ret;
2412 	}
2413 	reinit_completion(&isert_conn->login_req_comp);
2414 
2415 	/*
2416 	 * For login requests after the first PDU, isert_rx_login_req() will
2417 	 * kick schedule_delayed_work(&conn->login_work) as the packet is
2418 	 * received, which turns this callback from iscsi_target_do_login_rx()
2419 	 * into a NOP.
2420 	 */
2421 	if (!login->first_request)
2422 		return 0;
2423 
2424 	isert_rx_login_req(isert_conn);
2425 
2426 	isert_info("before login_comp conn: %p\n", conn);
2427 	ret = wait_for_completion_interruptible(&isert_conn->login_comp);
2428 	if (ret)
2429 		return ret;
2430 
2431 	isert_info("processing login->req: %p\n", login->req);
2432 
2433 	return 0;
2434 }
2435 
2436 static void
2437 isert_set_conn_info(struct iscsi_np *np, struct iscsi_conn *conn,
2438 		    struct isert_conn *isert_conn)
2439 {
2440 	struct rdma_cm_id *cm_id = isert_conn->cm_id;
2441 	struct rdma_route *cm_route = &cm_id->route;
2442 
2443 	conn->login_family = np->np_sockaddr.ss_family;
2444 
2445 	conn->login_sockaddr = cm_route->addr.dst_addr;
2446 	conn->local_sockaddr = cm_route->addr.src_addr;
2447 }
2448 
2449 static int
2450 isert_accept_np(struct iscsi_np *np, struct iscsi_conn *conn)
2451 {
2452 	struct isert_np *isert_np = np->np_context;
2453 	struct isert_conn *isert_conn;
2454 	int ret;
2455 
2456 accept_wait:
2457 	ret = down_interruptible(&isert_np->sem);
2458 	if (ret)
2459 		return -ENODEV;
2460 
2461 	spin_lock_bh(&np->np_thread_lock);
2462 	if (np->np_thread_state >= ISCSI_NP_THREAD_RESET) {
2463 		spin_unlock_bh(&np->np_thread_lock);
2464 		isert_dbg("np_thread_state %d\n",
2465 			 np->np_thread_state);
2466 		/**
2467 		 * No point in stalling here when np_thread
2468 		 * is in state RESET/SHUTDOWN/EXIT - bail
2469 		 **/
2470 		return -ENODEV;
2471 	}
2472 	spin_unlock_bh(&np->np_thread_lock);
2473 
2474 	mutex_lock(&isert_np->mutex);
2475 	if (list_empty(&isert_np->pending)) {
2476 		mutex_unlock(&isert_np->mutex);
2477 		goto accept_wait;
2478 	}
2479 	isert_conn = list_first_entry(&isert_np->pending,
2480 			struct isert_conn, node);
2481 	list_del_init(&isert_conn->node);
2482 	mutex_unlock(&isert_np->mutex);
2483 
2484 	conn->context = isert_conn;
2485 	isert_conn->conn = conn;
2486 	isert_conn->state = ISER_CONN_BOUND;
2487 
2488 	isert_set_conn_info(np, conn, isert_conn);
2489 
2490 	isert_dbg("Processing isert_conn: %p\n", isert_conn);
2491 
2492 	return 0;
2493 }
2494 
2495 static void
2496 isert_free_np(struct iscsi_np *np)
2497 {
2498 	struct isert_np *isert_np = np->np_context;
2499 	struct isert_conn *isert_conn, *n;
2500 
2501 	if (isert_np->cm_id)
2502 		rdma_destroy_id(isert_np->cm_id);
2503 
2504 	/*
2505 	 * FIXME: At this point we don't have a good way to insure
2506 	 * that at this point we don't have hanging connections that
2507 	 * completed RDMA establishment but didn't start iscsi login
2508 	 * process. So work-around this by cleaning up what ever piled
2509 	 * up in accepted and pending lists.
2510 	 */
2511 	mutex_lock(&isert_np->mutex);
2512 	if (!list_empty(&isert_np->pending)) {
2513 		isert_info("Still have isert pending connections\n");
2514 		list_for_each_entry_safe(isert_conn, n,
2515 					 &isert_np->pending,
2516 					 node) {
2517 			isert_info("cleaning isert_conn %p state (%d)\n",
2518 				   isert_conn, isert_conn->state);
2519 			isert_connect_release(isert_conn);
2520 		}
2521 	}
2522 
2523 	if (!list_empty(&isert_np->accepted)) {
2524 		isert_info("Still have isert accepted connections\n");
2525 		list_for_each_entry_safe(isert_conn, n,
2526 					 &isert_np->accepted,
2527 					 node) {
2528 			isert_info("cleaning isert_conn %p state (%d)\n",
2529 				   isert_conn, isert_conn->state);
2530 			isert_connect_release(isert_conn);
2531 		}
2532 	}
2533 	mutex_unlock(&isert_np->mutex);
2534 
2535 	np->np_context = NULL;
2536 	kfree(isert_np);
2537 }
2538 
2539 static void isert_release_work(struct work_struct *work)
2540 {
2541 	struct isert_conn *isert_conn = container_of(work,
2542 						     struct isert_conn,
2543 						     release_work);
2544 
2545 	isert_info("Starting release conn %p\n", isert_conn);
2546 
2547 	mutex_lock(&isert_conn->mutex);
2548 	isert_conn->state = ISER_CONN_DOWN;
2549 	mutex_unlock(&isert_conn->mutex);
2550 
2551 	isert_info("Destroying conn %p\n", isert_conn);
2552 	isert_put_conn(isert_conn);
2553 }
2554 
2555 static void
2556 isert_wait4logout(struct isert_conn *isert_conn)
2557 {
2558 	struct iscsi_conn *conn = isert_conn->conn;
2559 
2560 	isert_info("conn %p\n", isert_conn);
2561 
2562 	if (isert_conn->logout_posted) {
2563 		isert_info("conn %p wait for conn_logout_comp\n", isert_conn);
2564 		wait_for_completion_timeout(&conn->conn_logout_comp,
2565 					    SECONDS_FOR_LOGOUT_COMP * HZ);
2566 	}
2567 }
2568 
2569 static void
2570 isert_wait4cmds(struct iscsi_conn *conn)
2571 {
2572 	isert_info("iscsi_conn %p\n", conn);
2573 
2574 	if (conn->sess) {
2575 		target_sess_cmd_list_set_waiting(conn->sess->se_sess);
2576 		target_wait_for_sess_cmds(conn->sess->se_sess);
2577 	}
2578 }
2579 
2580 /**
2581  * isert_put_unsol_pending_cmds() - Drop commands waiting for
2582  *     unsolicitate dataout
2583  * @conn:    iscsi connection
2584  *
2585  * We might still have commands that are waiting for unsolicited
2586  * dataouts messages. We must put the extra reference on those
2587  * before blocking on the target_wait_for_session_cmds
2588  */
2589 static void
2590 isert_put_unsol_pending_cmds(struct iscsi_conn *conn)
2591 {
2592 	struct iscsi_cmd *cmd, *tmp;
2593 	static LIST_HEAD(drop_cmd_list);
2594 
2595 	spin_lock_bh(&conn->cmd_lock);
2596 	list_for_each_entry_safe(cmd, tmp, &conn->conn_cmd_list, i_conn_node) {
2597 		if ((cmd->cmd_flags & ICF_NON_IMMEDIATE_UNSOLICITED_DATA) &&
2598 		    (cmd->write_data_done < conn->sess->sess_ops->FirstBurstLength) &&
2599 		    (cmd->write_data_done < cmd->se_cmd.data_length))
2600 			list_move_tail(&cmd->i_conn_node, &drop_cmd_list);
2601 	}
2602 	spin_unlock_bh(&conn->cmd_lock);
2603 
2604 	list_for_each_entry_safe(cmd, tmp, &drop_cmd_list, i_conn_node) {
2605 		list_del_init(&cmd->i_conn_node);
2606 		if (cmd->i_state != ISTATE_REMOVE) {
2607 			struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2608 
2609 			isert_info("conn %p dropping cmd %p\n", conn, cmd);
2610 			isert_put_cmd(isert_cmd, true);
2611 		}
2612 	}
2613 }
2614 
2615 static void isert_wait_conn(struct iscsi_conn *conn)
2616 {
2617 	struct isert_conn *isert_conn = conn->context;
2618 
2619 	isert_info("Starting conn %p\n", isert_conn);
2620 
2621 	mutex_lock(&isert_conn->mutex);
2622 	isert_conn_terminate(isert_conn);
2623 	mutex_unlock(&isert_conn->mutex);
2624 
2625 	ib_drain_qp(isert_conn->qp);
2626 	isert_put_unsol_pending_cmds(conn);
2627 	isert_wait4cmds(conn);
2628 	isert_wait4logout(isert_conn);
2629 
2630 	queue_work(isert_release_wq, &isert_conn->release_work);
2631 }
2632 
2633 static void isert_free_conn(struct iscsi_conn *conn)
2634 {
2635 	struct isert_conn *isert_conn = conn->context;
2636 
2637 	ib_drain_qp(isert_conn->qp);
2638 	isert_put_conn(isert_conn);
2639 }
2640 
2641 static void isert_get_rx_pdu(struct iscsi_conn *conn)
2642 {
2643 	struct completion comp;
2644 
2645 	init_completion(&comp);
2646 
2647 	wait_for_completion_interruptible(&comp);
2648 }
2649 
2650 static struct iscsit_transport iser_target_transport = {
2651 	.name			= "IB/iSER",
2652 	.transport_type		= ISCSI_INFINIBAND,
2653 	.rdma_shutdown		= true,
2654 	.priv_size		= sizeof(struct isert_cmd),
2655 	.owner			= THIS_MODULE,
2656 	.iscsit_setup_np	= isert_setup_np,
2657 	.iscsit_accept_np	= isert_accept_np,
2658 	.iscsit_free_np		= isert_free_np,
2659 	.iscsit_wait_conn	= isert_wait_conn,
2660 	.iscsit_free_conn	= isert_free_conn,
2661 	.iscsit_get_login_rx	= isert_get_login_rx,
2662 	.iscsit_put_login_tx	= isert_put_login_tx,
2663 	.iscsit_immediate_queue	= isert_immediate_queue,
2664 	.iscsit_response_queue	= isert_response_queue,
2665 	.iscsit_get_dataout	= isert_get_dataout,
2666 	.iscsit_queue_data_in	= isert_put_datain,
2667 	.iscsit_queue_status	= isert_put_response,
2668 	.iscsit_aborted_task	= isert_aborted_task,
2669 	.iscsit_get_rx_pdu	= isert_get_rx_pdu,
2670 	.iscsit_get_sup_prot_ops = isert_get_sup_prot_ops,
2671 };
2672 
2673 static int __init isert_init(void)
2674 {
2675 	int ret;
2676 
2677 	isert_comp_wq = alloc_workqueue("isert_comp_wq",
2678 					WQ_UNBOUND | WQ_HIGHPRI, 0);
2679 	if (!isert_comp_wq) {
2680 		isert_err("Unable to allocate isert_comp_wq\n");
2681 		return -ENOMEM;
2682 	}
2683 
2684 	isert_release_wq = alloc_workqueue("isert_release_wq", WQ_UNBOUND,
2685 					WQ_UNBOUND_MAX_ACTIVE);
2686 	if (!isert_release_wq) {
2687 		isert_err("Unable to allocate isert_release_wq\n");
2688 		ret = -ENOMEM;
2689 		goto destroy_comp_wq;
2690 	}
2691 
2692 	iscsit_register_transport(&iser_target_transport);
2693 	isert_info("iSER_TARGET[0] - Loaded iser_target_transport\n");
2694 
2695 	return 0;
2696 
2697 destroy_comp_wq:
2698 	destroy_workqueue(isert_comp_wq);
2699 
2700 	return ret;
2701 }
2702 
2703 static void __exit isert_exit(void)
2704 {
2705 	flush_scheduled_work();
2706 	destroy_workqueue(isert_release_wq);
2707 	destroy_workqueue(isert_comp_wq);
2708 	iscsit_unregister_transport(&iser_target_transport);
2709 	isert_info("iSER_TARGET[0] - Released iser_target_transport\n");
2710 }
2711 
2712 MODULE_DESCRIPTION("iSER-Target for mainline target infrastructure");
2713 MODULE_VERSION("1.0");
2714 MODULE_AUTHOR("nab@Linux-iSCSI.org");
2715 MODULE_LICENSE("GPL");
2716 
2717 module_init(isert_init);
2718 module_exit(isert_exit);
2719