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/ib_cm.h>
19 #include <rdma/rdma_cm.h>
20 #include <target/target_core_base.h>
21 #include <target/target_core_fabric.h>
22 #include <target/iscsi/iscsi_transport.h>
23 #include <linux/semaphore.h>
24 
25 #include "ib_isert.h"
26 
27 static int isert_debug_level;
28 module_param_named(debug_level, isert_debug_level, int, 0644);
29 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0 (default:0)");
30 
31 static int isert_sg_tablesize_set(const char *val,
32 				  const struct kernel_param *kp);
33 static const struct kernel_param_ops sg_tablesize_ops = {
34 	.set = isert_sg_tablesize_set,
35 	.get = param_get_int,
36 };
37 
38 static int isert_sg_tablesize = ISCSI_ISER_DEF_SG_TABLESIZE;
39 module_param_cb(sg_tablesize, &sg_tablesize_ops, &isert_sg_tablesize, 0644);
40 MODULE_PARM_DESC(sg_tablesize,
41 		 "Number of gather/scatter entries in a single scsi command, should >= 128 (default: 256, max: 4096)");
42 
43 static DEFINE_MUTEX(device_list_mutex);
44 static LIST_HEAD(device_list);
45 static struct workqueue_struct *isert_comp_wq;
46 static struct workqueue_struct *isert_release_wq;
47 
48 static int
49 isert_put_response(struct iscsi_conn *conn, struct iscsi_cmd *cmd);
50 static int
51 isert_login_post_recv(struct isert_conn *isert_conn);
52 static int
53 isert_rdma_accept(struct isert_conn *isert_conn);
54 struct rdma_cm_id *isert_setup_id(struct isert_np *isert_np);
55 
56 static void isert_release_work(struct work_struct *work);
57 static void isert_recv_done(struct ib_cq *cq, struct ib_wc *wc);
58 static void isert_send_done(struct ib_cq *cq, struct ib_wc *wc);
59 static void isert_login_recv_done(struct ib_cq *cq, struct ib_wc *wc);
60 static void isert_login_send_done(struct ib_cq *cq, struct ib_wc *wc);
61 
62 static int isert_sg_tablesize_set(const char *val, const struct kernel_param *kp)
63 {
64 	int n = 0, ret;
65 
66 	ret = kstrtoint(val, 10, &n);
67 	if (ret != 0 || n < ISCSI_ISER_MIN_SG_TABLESIZE ||
68 	    n > ISCSI_ISER_MAX_SG_TABLESIZE)
69 		return -EINVAL;
70 
71 	return param_set_int(val, kp);
72 }
73 
74 
75 static inline bool
76 isert_prot_cmd(struct isert_conn *conn, struct se_cmd *cmd)
77 {
78 	return (conn->pi_support &&
79 		cmd->prot_op != TARGET_PROT_NORMAL);
80 }
81 
82 
83 static void
84 isert_qp_event_callback(struct ib_event *e, void *context)
85 {
86 	struct isert_conn *isert_conn = context;
87 
88 	isert_err("%s (%d): conn %p\n",
89 		  ib_event_msg(e->event), e->event, isert_conn);
90 
91 	switch (e->event) {
92 	case IB_EVENT_COMM_EST:
93 		rdma_notify(isert_conn->cm_id, IB_EVENT_COMM_EST);
94 		break;
95 	case IB_EVENT_QP_LAST_WQE_REACHED:
96 		isert_warn("Reached TX IB_EVENT_QP_LAST_WQE_REACHED\n");
97 		break;
98 	default:
99 		break;
100 	}
101 }
102 
103 static struct ib_qp *
104 isert_create_qp(struct isert_conn *isert_conn,
105 		struct rdma_cm_id *cma_id)
106 {
107 	u32 cq_size = ISERT_QP_MAX_REQ_DTOS + ISERT_QP_MAX_RECV_DTOS + 2;
108 	struct isert_device *device = isert_conn->device;
109 	struct ib_device *ib_dev = device->ib_device;
110 	struct ib_qp_init_attr attr;
111 	int ret, factor;
112 
113 	isert_conn->cq = ib_cq_pool_get(ib_dev, cq_size, -1, IB_POLL_WORKQUEUE);
114 	if (IS_ERR(isert_conn->cq)) {
115 		isert_err("Unable to allocate cq\n");
116 		ret = PTR_ERR(isert_conn->cq);
117 		return ERR_PTR(ret);
118 	}
119 	isert_conn->cq_size = cq_size;
120 
121 	memset(&attr, 0, sizeof(struct ib_qp_init_attr));
122 	attr.event_handler = isert_qp_event_callback;
123 	attr.qp_context = isert_conn;
124 	attr.send_cq = isert_conn->cq;
125 	attr.recv_cq = isert_conn->cq;
126 	attr.cap.max_send_wr = ISERT_QP_MAX_REQ_DTOS + 1;
127 	attr.cap.max_recv_wr = ISERT_QP_MAX_RECV_DTOS + 1;
128 	factor = rdma_rw_mr_factor(device->ib_device, cma_id->port_num,
129 				   isert_sg_tablesize);
130 	attr.cap.max_rdma_ctxs = ISCSI_DEF_XMIT_CMDS_MAX * factor;
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 		ib_cq_pool_put(isert_conn->cq, isert_conn->cq_size);
142 
143 		return ERR_PTR(ret);
144 	}
145 
146 	return cma_id->qp;
147 }
148 
149 static int
150 isert_alloc_rx_descriptors(struct isert_conn *isert_conn)
151 {
152 	struct isert_device *device = isert_conn->device;
153 	struct ib_device *ib_dev = device->ib_device;
154 	struct iser_rx_desc *rx_desc;
155 	struct ib_sge *rx_sg;
156 	u64 dma_addr;
157 	int i, j;
158 
159 	isert_conn->rx_descs = kcalloc(ISERT_QP_MAX_RECV_DTOS,
160 				       sizeof(struct iser_rx_desc),
161 				       GFP_KERNEL);
162 	if (!isert_conn->rx_descs)
163 		return -ENOMEM;
164 
165 	rx_desc = isert_conn->rx_descs;
166 
167 	for (i = 0; i < ISERT_QP_MAX_RECV_DTOS; i++, rx_desc++)  {
168 		dma_addr = ib_dma_map_single(ib_dev, rx_desc->buf,
169 					ISER_RX_SIZE, DMA_FROM_DEVICE);
170 		if (ib_dma_mapping_error(ib_dev, dma_addr))
171 			goto dma_map_fail;
172 
173 		rx_desc->dma_addr = dma_addr;
174 
175 		rx_sg = &rx_desc->rx_sg;
176 		rx_sg->addr = rx_desc->dma_addr + isert_get_hdr_offset(rx_desc);
177 		rx_sg->length = ISER_RX_PAYLOAD_SIZE;
178 		rx_sg->lkey = device->pd->local_dma_lkey;
179 		rx_desc->rx_cqe.done = isert_recv_done;
180 	}
181 
182 	return 0;
183 
184 dma_map_fail:
185 	rx_desc = isert_conn->rx_descs;
186 	for (j = 0; j < i; j++, rx_desc++) {
187 		ib_dma_unmap_single(ib_dev, rx_desc->dma_addr,
188 				    ISER_RX_SIZE, DMA_FROM_DEVICE);
189 	}
190 	kfree(isert_conn->rx_descs);
191 	isert_conn->rx_descs = NULL;
192 	isert_err("conn %p failed to allocate rx descriptors\n", isert_conn);
193 	return -ENOMEM;
194 }
195 
196 static void
197 isert_free_rx_descriptors(struct isert_conn *isert_conn)
198 {
199 	struct ib_device *ib_dev = isert_conn->device->ib_device;
200 	struct iser_rx_desc *rx_desc;
201 	int i;
202 
203 	if (!isert_conn->rx_descs)
204 		return;
205 
206 	rx_desc = isert_conn->rx_descs;
207 	for (i = 0; i < ISERT_QP_MAX_RECV_DTOS; i++, rx_desc++)  {
208 		ib_dma_unmap_single(ib_dev, rx_desc->dma_addr,
209 				    ISER_RX_SIZE, DMA_FROM_DEVICE);
210 	}
211 
212 	kfree(isert_conn->rx_descs);
213 	isert_conn->rx_descs = NULL;
214 }
215 
216 static int
217 isert_create_device_ib_res(struct isert_device *device)
218 {
219 	struct ib_device *ib_dev = device->ib_device;
220 	int ret;
221 
222 	isert_dbg("devattr->max_send_sge: %d devattr->max_recv_sge %d\n",
223 		  ib_dev->attrs.max_send_sge, ib_dev->attrs.max_recv_sge);
224 	isert_dbg("devattr->max_sge_rd: %d\n", ib_dev->attrs.max_sge_rd);
225 
226 	device->pd = ib_alloc_pd(ib_dev, 0);
227 	if (IS_ERR(device->pd)) {
228 		ret = PTR_ERR(device->pd);
229 		isert_err("failed to allocate pd, device %p, ret=%d\n",
230 			  device, ret);
231 		return ret;
232 	}
233 
234 	/* Check signature cap */
235 	device->pi_capable = ib_dev->attrs.device_cap_flags &
236 			     IB_DEVICE_INTEGRITY_HANDOVER ? true : false;
237 
238 	return 0;
239 }
240 
241 static void
242 isert_free_device_ib_res(struct isert_device *device)
243 {
244 	isert_info("device %p\n", device);
245 
246 	ib_dealloc_pd(device->pd);
247 }
248 
249 static void
250 isert_device_put(struct isert_device *device)
251 {
252 	mutex_lock(&device_list_mutex);
253 	device->refcount--;
254 	isert_info("device %p refcount %d\n", device, device->refcount);
255 	if (!device->refcount) {
256 		isert_free_device_ib_res(device);
257 		list_del(&device->dev_node);
258 		kfree(device);
259 	}
260 	mutex_unlock(&device_list_mutex);
261 }
262 
263 static struct isert_device *
264 isert_device_get(struct rdma_cm_id *cma_id)
265 {
266 	struct isert_device *device;
267 	int ret;
268 
269 	mutex_lock(&device_list_mutex);
270 	list_for_each_entry(device, &device_list, dev_node) {
271 		if (device->ib_device->node_guid == cma_id->device->node_guid) {
272 			device->refcount++;
273 			isert_info("Found iser device %p refcount %d\n",
274 				   device, device->refcount);
275 			mutex_unlock(&device_list_mutex);
276 			return device;
277 		}
278 	}
279 
280 	device = kzalloc(sizeof(struct isert_device), GFP_KERNEL);
281 	if (!device) {
282 		mutex_unlock(&device_list_mutex);
283 		return ERR_PTR(-ENOMEM);
284 	}
285 
286 	INIT_LIST_HEAD(&device->dev_node);
287 
288 	device->ib_device = cma_id->device;
289 	ret = isert_create_device_ib_res(device);
290 	if (ret) {
291 		kfree(device);
292 		mutex_unlock(&device_list_mutex);
293 		return ERR_PTR(ret);
294 	}
295 
296 	device->refcount++;
297 	list_add_tail(&device->dev_node, &device_list);
298 	isert_info("Created a new iser device %p refcount %d\n",
299 		   device, device->refcount);
300 	mutex_unlock(&device_list_mutex);
301 
302 	return device;
303 }
304 
305 static void
306 isert_init_conn(struct isert_conn *isert_conn)
307 {
308 	isert_conn->state = ISER_CONN_INIT;
309 	INIT_LIST_HEAD(&isert_conn->node);
310 	init_completion(&isert_conn->login_comp);
311 	init_completion(&isert_conn->login_req_comp);
312 	init_waitqueue_head(&isert_conn->rem_wait);
313 	kref_init(&isert_conn->kref);
314 	mutex_init(&isert_conn->mutex);
315 	INIT_WORK(&isert_conn->release_work, isert_release_work);
316 }
317 
318 static void
319 isert_free_login_buf(struct isert_conn *isert_conn)
320 {
321 	struct ib_device *ib_dev = isert_conn->device->ib_device;
322 
323 	ib_dma_unmap_single(ib_dev, isert_conn->login_rsp_dma,
324 			    ISER_RX_PAYLOAD_SIZE, DMA_TO_DEVICE);
325 	kfree(isert_conn->login_rsp_buf);
326 
327 	ib_dma_unmap_single(ib_dev, isert_conn->login_desc->dma_addr,
328 			    ISER_RX_SIZE, DMA_FROM_DEVICE);
329 	kfree(isert_conn->login_desc);
330 }
331 
332 static int
333 isert_alloc_login_buf(struct isert_conn *isert_conn,
334 		      struct ib_device *ib_dev)
335 {
336 	int ret;
337 
338 	isert_conn->login_desc = kzalloc(sizeof(*isert_conn->login_desc),
339 			GFP_KERNEL);
340 	if (!isert_conn->login_desc)
341 		return -ENOMEM;
342 
343 	isert_conn->login_desc->dma_addr = ib_dma_map_single(ib_dev,
344 				isert_conn->login_desc->buf,
345 				ISER_RX_SIZE, DMA_FROM_DEVICE);
346 	ret = ib_dma_mapping_error(ib_dev, isert_conn->login_desc->dma_addr);
347 	if (ret) {
348 		isert_err("login_desc dma mapping error: %d\n", ret);
349 		isert_conn->login_desc->dma_addr = 0;
350 		goto out_free_login_desc;
351 	}
352 
353 	isert_conn->login_rsp_buf = kzalloc(ISER_RX_PAYLOAD_SIZE, GFP_KERNEL);
354 	if (!isert_conn->login_rsp_buf) {
355 		ret = -ENOMEM;
356 		goto out_unmap_login_desc;
357 	}
358 
359 	isert_conn->login_rsp_dma = ib_dma_map_single(ib_dev,
360 					isert_conn->login_rsp_buf,
361 					ISER_RX_PAYLOAD_SIZE, DMA_TO_DEVICE);
362 	ret = ib_dma_mapping_error(ib_dev, isert_conn->login_rsp_dma);
363 	if (ret) {
364 		isert_err("login_rsp_dma mapping error: %d\n", ret);
365 		isert_conn->login_rsp_dma = 0;
366 		goto out_free_login_rsp_buf;
367 	}
368 
369 	return 0;
370 
371 out_free_login_rsp_buf:
372 	kfree(isert_conn->login_rsp_buf);
373 out_unmap_login_desc:
374 	ib_dma_unmap_single(ib_dev, isert_conn->login_desc->dma_addr,
375 			    ISER_RX_SIZE, DMA_FROM_DEVICE);
376 out_free_login_desc:
377 	kfree(isert_conn->login_desc);
378 	return ret;
379 }
380 
381 static void
382 isert_set_nego_params(struct isert_conn *isert_conn,
383 		      struct rdma_conn_param *param)
384 {
385 	struct ib_device_attr *attr = &isert_conn->device->ib_device->attrs;
386 
387 	/* Set max inflight RDMA READ requests */
388 	isert_conn->initiator_depth = min_t(u8, param->initiator_depth,
389 				attr->max_qp_init_rd_atom);
390 	isert_dbg("Using initiator_depth: %u\n", isert_conn->initiator_depth);
391 
392 	if (param->private_data) {
393 		u8 flags = *(u8 *)param->private_data;
394 
395 		/*
396 		 * use remote invalidation if the both initiator
397 		 * and the HCA support it
398 		 */
399 		isert_conn->snd_w_inv = !(flags & ISER_SEND_W_INV_NOT_SUP) &&
400 					  (attr->device_cap_flags &
401 					   IB_DEVICE_MEM_MGT_EXTENSIONS);
402 		if (isert_conn->snd_w_inv)
403 			isert_info("Using remote invalidation\n");
404 	}
405 }
406 
407 static void
408 isert_destroy_qp(struct isert_conn *isert_conn)
409 {
410 	ib_destroy_qp(isert_conn->qp);
411 	ib_cq_pool_put(isert_conn->cq, isert_conn->cq_size);
412 }
413 
414 static int
415 isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
416 {
417 	struct isert_np *isert_np = cma_id->context;
418 	struct iscsi_np *np = isert_np->np;
419 	struct isert_conn *isert_conn;
420 	struct isert_device *device;
421 	int ret = 0;
422 
423 	spin_lock_bh(&np->np_thread_lock);
424 	if (!np->enabled) {
425 		spin_unlock_bh(&np->np_thread_lock);
426 		isert_dbg("iscsi_np is not enabled, reject connect request\n");
427 		return rdma_reject(cma_id, NULL, 0, IB_CM_REJ_CONSUMER_DEFINED);
428 	}
429 	spin_unlock_bh(&np->np_thread_lock);
430 
431 	isert_dbg("cma_id: %p, portal: %p\n",
432 		 cma_id, cma_id->context);
433 
434 	isert_conn = kzalloc(sizeof(struct isert_conn), GFP_KERNEL);
435 	if (!isert_conn)
436 		return -ENOMEM;
437 
438 	isert_init_conn(isert_conn);
439 	isert_conn->cm_id = cma_id;
440 
441 	ret = isert_alloc_login_buf(isert_conn, cma_id->device);
442 	if (ret)
443 		goto out;
444 
445 	device = isert_device_get(cma_id);
446 	if (IS_ERR(device)) {
447 		ret = PTR_ERR(device);
448 		goto out_rsp_dma_map;
449 	}
450 	isert_conn->device = device;
451 
452 	isert_set_nego_params(isert_conn, &event->param.conn);
453 
454 	isert_conn->qp = isert_create_qp(isert_conn, cma_id);
455 	if (IS_ERR(isert_conn->qp)) {
456 		ret = PTR_ERR(isert_conn->qp);
457 		goto out_conn_dev;
458 	}
459 
460 	ret = isert_login_post_recv(isert_conn);
461 	if (ret)
462 		goto out_destroy_qp;
463 
464 	ret = isert_rdma_accept(isert_conn);
465 	if (ret)
466 		goto out_destroy_qp;
467 
468 	mutex_lock(&isert_np->mutex);
469 	list_add_tail(&isert_conn->node, &isert_np->accepted);
470 	mutex_unlock(&isert_np->mutex);
471 
472 	return 0;
473 
474 out_destroy_qp:
475 	isert_destroy_qp(isert_conn);
476 out_conn_dev:
477 	isert_device_put(device);
478 out_rsp_dma_map:
479 	isert_free_login_buf(isert_conn);
480 out:
481 	kfree(isert_conn);
482 	rdma_reject(cma_id, NULL, 0, IB_CM_REJ_CONSUMER_DEFINED);
483 	return ret;
484 }
485 
486 static void
487 isert_connect_release(struct isert_conn *isert_conn)
488 {
489 	struct isert_device *device = isert_conn->device;
490 
491 	isert_dbg("conn %p\n", isert_conn);
492 
493 	BUG_ON(!device);
494 
495 	isert_free_rx_descriptors(isert_conn);
496 	if (isert_conn->cm_id &&
497 	    !isert_conn->dev_removed)
498 		rdma_destroy_id(isert_conn->cm_id);
499 
500 	if (isert_conn->qp)
501 		isert_destroy_qp(isert_conn);
502 
503 	if (isert_conn->login_desc)
504 		isert_free_login_buf(isert_conn);
505 
506 	isert_device_put(device);
507 
508 	if (isert_conn->dev_removed)
509 		wake_up_interruptible(&isert_conn->rem_wait);
510 	else
511 		kfree(isert_conn);
512 }
513 
514 static void
515 isert_connected_handler(struct rdma_cm_id *cma_id)
516 {
517 	struct isert_conn *isert_conn = cma_id->qp->qp_context;
518 	struct isert_np *isert_np = cma_id->context;
519 
520 	isert_info("conn %p\n", isert_conn);
521 
522 	mutex_lock(&isert_conn->mutex);
523 	isert_conn->state = ISER_CONN_UP;
524 	kref_get(&isert_conn->kref);
525 	mutex_unlock(&isert_conn->mutex);
526 
527 	mutex_lock(&isert_np->mutex);
528 	list_move_tail(&isert_conn->node, &isert_np->pending);
529 	mutex_unlock(&isert_np->mutex);
530 
531 	isert_info("np %p: Allow accept_np to continue\n", isert_np);
532 	up(&isert_np->sem);
533 }
534 
535 static void
536 isert_release_kref(struct kref *kref)
537 {
538 	struct isert_conn *isert_conn = container_of(kref,
539 				struct isert_conn, kref);
540 
541 	isert_info("conn %p final kref %s/%d\n", isert_conn, current->comm,
542 		   current->pid);
543 
544 	isert_connect_release(isert_conn);
545 }
546 
547 static void
548 isert_put_conn(struct isert_conn *isert_conn)
549 {
550 	kref_put(&isert_conn->kref, isert_release_kref);
551 }
552 
553 static void
554 isert_handle_unbound_conn(struct isert_conn *isert_conn)
555 {
556 	struct isert_np *isert_np = isert_conn->cm_id->context;
557 
558 	mutex_lock(&isert_np->mutex);
559 	if (!list_empty(&isert_conn->node)) {
560 		/*
561 		 * This means iscsi doesn't know this connection
562 		 * so schedule a cleanup ourselves
563 		 */
564 		list_del_init(&isert_conn->node);
565 		isert_put_conn(isert_conn);
566 		queue_work(isert_release_wq, &isert_conn->release_work);
567 	}
568 	mutex_unlock(&isert_np->mutex);
569 }
570 
571 /**
572  * isert_conn_terminate() - Initiate connection termination
573  * @isert_conn: isert connection struct
574  *
575  * Notes:
576  * In case the connection state is BOUND, move state
577  * to TEMINATING and start teardown sequence (rdma_disconnect).
578  * In case the connection state is UP, complete flush as well.
579  *
580  * This routine must be called with mutex held. Thus it is
581  * safe to call multiple times.
582  */
583 static void
584 isert_conn_terminate(struct isert_conn *isert_conn)
585 {
586 	int err;
587 
588 	if (isert_conn->state >= ISER_CONN_TERMINATING)
589 		return;
590 
591 	isert_info("Terminating conn %p state %d\n",
592 		   isert_conn, isert_conn->state);
593 	isert_conn->state = ISER_CONN_TERMINATING;
594 	err = rdma_disconnect(isert_conn->cm_id);
595 	if (err)
596 		isert_warn("Failed rdma_disconnect isert_conn %p\n",
597 			   isert_conn);
598 }
599 
600 static int
601 isert_np_cma_handler(struct isert_np *isert_np,
602 		     enum rdma_cm_event_type event)
603 {
604 	isert_dbg("%s (%d): isert np %p\n",
605 		  rdma_event_msg(event), event, isert_np);
606 
607 	switch (event) {
608 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
609 		isert_np->cm_id = NULL;
610 		break;
611 	case RDMA_CM_EVENT_ADDR_CHANGE:
612 		isert_np->cm_id = isert_setup_id(isert_np);
613 		if (IS_ERR(isert_np->cm_id)) {
614 			isert_err("isert np %p setup id failed: %ld\n",
615 				  isert_np, PTR_ERR(isert_np->cm_id));
616 			isert_np->cm_id = NULL;
617 		}
618 		break;
619 	default:
620 		isert_err("isert np %p Unexpected event %d\n",
621 			  isert_np, event);
622 	}
623 
624 	return -1;
625 }
626 
627 static int
628 isert_disconnected_handler(struct rdma_cm_id *cma_id,
629 			   enum rdma_cm_event_type event)
630 {
631 	struct isert_conn *isert_conn = cma_id->qp->qp_context;
632 
633 	mutex_lock(&isert_conn->mutex);
634 	switch (isert_conn->state) {
635 	case ISER_CONN_TERMINATING:
636 		break;
637 	case ISER_CONN_UP:
638 		isert_conn_terminate(isert_conn);
639 		ib_drain_qp(isert_conn->qp);
640 		isert_handle_unbound_conn(isert_conn);
641 		break;
642 	case ISER_CONN_BOUND:
643 	case ISER_CONN_FULL_FEATURE: /* FALLTHRU */
644 		iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
645 		break;
646 	default:
647 		isert_warn("conn %p terminating in state %d\n",
648 			   isert_conn, isert_conn->state);
649 	}
650 	mutex_unlock(&isert_conn->mutex);
651 
652 	return 0;
653 }
654 
655 static int
656 isert_connect_error(struct rdma_cm_id *cma_id)
657 {
658 	struct isert_conn *isert_conn = cma_id->qp->qp_context;
659 
660 	ib_drain_qp(isert_conn->qp);
661 	list_del_init(&isert_conn->node);
662 	isert_conn->cm_id = NULL;
663 	isert_put_conn(isert_conn);
664 
665 	return -1;
666 }
667 
668 static int
669 isert_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
670 {
671 	struct isert_np *isert_np = cma_id->context;
672 	struct isert_conn *isert_conn;
673 	int ret = 0;
674 
675 	isert_info("%s (%d): status %d id %p np %p\n",
676 		   rdma_event_msg(event->event), event->event,
677 		   event->status, cma_id, cma_id->context);
678 
679 	if (isert_np->cm_id == cma_id)
680 		return isert_np_cma_handler(cma_id->context, event->event);
681 
682 	switch (event->event) {
683 	case RDMA_CM_EVENT_CONNECT_REQUEST:
684 		ret = isert_connect_request(cma_id, event);
685 		if (ret)
686 			isert_err("failed handle connect request %d\n", ret);
687 		break;
688 	case RDMA_CM_EVENT_ESTABLISHED:
689 		isert_connected_handler(cma_id);
690 		break;
691 	case RDMA_CM_EVENT_ADDR_CHANGE:
692 	case RDMA_CM_EVENT_DISCONNECTED:
693 	case RDMA_CM_EVENT_TIMEWAIT_EXIT:  /* FALLTHRU */
694 		ret = isert_disconnected_handler(cma_id, event->event);
695 		break;
696 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
697 		isert_conn = cma_id->qp->qp_context;
698 		isert_conn->dev_removed = true;
699 		isert_disconnected_handler(cma_id, event->event);
700 		wait_event_interruptible(isert_conn->rem_wait,
701 					 isert_conn->state == ISER_CONN_DOWN);
702 		kfree(isert_conn);
703 		/*
704 		 * return non-zero from the callback to destroy
705 		 * the rdma cm id
706 		 */
707 		return 1;
708 	case RDMA_CM_EVENT_REJECTED:
709 		isert_info("Connection rejected: %s\n",
710 			   rdma_reject_msg(cma_id, event->status));
711 		fallthrough;
712 	case RDMA_CM_EVENT_UNREACHABLE:
713 	case RDMA_CM_EVENT_CONNECT_ERROR:
714 		ret = isert_connect_error(cma_id);
715 		break;
716 	default:
717 		isert_err("Unhandled RDMA CMA event: %d\n", event->event);
718 		break;
719 	}
720 
721 	return ret;
722 }
723 
724 static int
725 isert_post_recvm(struct isert_conn *isert_conn, u32 count)
726 {
727 	struct ib_recv_wr *rx_wr;
728 	int i, ret;
729 	struct iser_rx_desc *rx_desc;
730 
731 	for (rx_wr = isert_conn->rx_wr, i = 0; i < count; i++, rx_wr++) {
732 		rx_desc = &isert_conn->rx_descs[i];
733 
734 		rx_wr->wr_cqe = &rx_desc->rx_cqe;
735 		rx_wr->sg_list = &rx_desc->rx_sg;
736 		rx_wr->num_sge = 1;
737 		rx_wr->next = rx_wr + 1;
738 		rx_desc->in_use = false;
739 	}
740 	rx_wr--;
741 	rx_wr->next = NULL; /* mark end of work requests list */
742 
743 	ret = ib_post_recv(isert_conn->qp, isert_conn->rx_wr, NULL);
744 	if (ret)
745 		isert_err("ib_post_recv() failed with ret: %d\n", ret);
746 
747 	return ret;
748 }
749 
750 static int
751 isert_post_recv(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc)
752 {
753 	struct ib_recv_wr rx_wr;
754 	int ret;
755 
756 	if (!rx_desc->in_use) {
757 		/*
758 		 * if the descriptor is not in-use we already reposted it
759 		 * for recv, so just silently return
760 		 */
761 		return 0;
762 	}
763 
764 	rx_desc->in_use = false;
765 	rx_wr.wr_cqe = &rx_desc->rx_cqe;
766 	rx_wr.sg_list = &rx_desc->rx_sg;
767 	rx_wr.num_sge = 1;
768 	rx_wr.next = NULL;
769 
770 	ret = ib_post_recv(isert_conn->qp, &rx_wr, NULL);
771 	if (ret)
772 		isert_err("ib_post_recv() failed with ret: %d\n", ret);
773 
774 	return ret;
775 }
776 
777 static int
778 isert_login_post_send(struct isert_conn *isert_conn, struct iser_tx_desc *tx_desc)
779 {
780 	struct ib_device *ib_dev = isert_conn->cm_id->device;
781 	struct ib_send_wr send_wr;
782 	int ret;
783 
784 	ib_dma_sync_single_for_device(ib_dev, tx_desc->dma_addr,
785 				      ISER_HEADERS_LEN, DMA_TO_DEVICE);
786 
787 	tx_desc->tx_cqe.done = isert_login_send_done;
788 
789 	send_wr.next	= NULL;
790 	send_wr.wr_cqe	= &tx_desc->tx_cqe;
791 	send_wr.sg_list	= tx_desc->tx_sg;
792 	send_wr.num_sge	= tx_desc->num_sge;
793 	send_wr.opcode	= IB_WR_SEND;
794 	send_wr.send_flags = IB_SEND_SIGNALED;
795 
796 	ret = ib_post_send(isert_conn->qp, &send_wr, NULL);
797 	if (ret)
798 		isert_err("ib_post_send() failed, ret: %d\n", ret);
799 
800 	return ret;
801 }
802 
803 static void
804 __isert_create_send_desc(struct isert_device *device,
805 			 struct iser_tx_desc *tx_desc)
806 {
807 
808 	memset(&tx_desc->iser_header, 0, sizeof(struct iser_ctrl));
809 	tx_desc->iser_header.flags = ISCSI_CTRL;
810 
811 	tx_desc->num_sge = 1;
812 
813 	if (tx_desc->tx_sg[0].lkey != device->pd->local_dma_lkey) {
814 		tx_desc->tx_sg[0].lkey = device->pd->local_dma_lkey;
815 		isert_dbg("tx_desc %p lkey mismatch, fixing\n", tx_desc);
816 	}
817 }
818 
819 static void
820 isert_create_send_desc(struct isert_conn *isert_conn,
821 		       struct isert_cmd *isert_cmd,
822 		       struct iser_tx_desc *tx_desc)
823 {
824 	struct isert_device *device = isert_conn->device;
825 	struct ib_device *ib_dev = device->ib_device;
826 
827 	ib_dma_sync_single_for_cpu(ib_dev, tx_desc->dma_addr,
828 				   ISER_HEADERS_LEN, DMA_TO_DEVICE);
829 
830 	__isert_create_send_desc(device, tx_desc);
831 }
832 
833 static int
834 isert_init_tx_hdrs(struct isert_conn *isert_conn,
835 		   struct iser_tx_desc *tx_desc)
836 {
837 	struct isert_device *device = isert_conn->device;
838 	struct ib_device *ib_dev = device->ib_device;
839 	u64 dma_addr;
840 
841 	dma_addr = ib_dma_map_single(ib_dev, (void *)tx_desc,
842 			ISER_HEADERS_LEN, DMA_TO_DEVICE);
843 	if (ib_dma_mapping_error(ib_dev, dma_addr)) {
844 		isert_err("ib_dma_mapping_error() failed\n");
845 		return -ENOMEM;
846 	}
847 
848 	tx_desc->dma_addr = dma_addr;
849 	tx_desc->tx_sg[0].addr	= tx_desc->dma_addr;
850 	tx_desc->tx_sg[0].length = ISER_HEADERS_LEN;
851 	tx_desc->tx_sg[0].lkey = device->pd->local_dma_lkey;
852 
853 	isert_dbg("Setup tx_sg[0].addr: 0x%llx length: %u lkey: 0x%x\n",
854 		  tx_desc->tx_sg[0].addr, tx_desc->tx_sg[0].length,
855 		  tx_desc->tx_sg[0].lkey);
856 
857 	return 0;
858 }
859 
860 static void
861 isert_init_send_wr(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
862 		   struct ib_send_wr *send_wr)
863 {
864 	struct iser_tx_desc *tx_desc = &isert_cmd->tx_desc;
865 
866 	tx_desc->tx_cqe.done = isert_send_done;
867 	send_wr->wr_cqe = &tx_desc->tx_cqe;
868 
869 	if (isert_conn->snd_w_inv && isert_cmd->inv_rkey) {
870 		send_wr->opcode  = IB_WR_SEND_WITH_INV;
871 		send_wr->ex.invalidate_rkey = isert_cmd->inv_rkey;
872 	} else {
873 		send_wr->opcode = IB_WR_SEND;
874 	}
875 
876 	send_wr->sg_list = &tx_desc->tx_sg[0];
877 	send_wr->num_sge = isert_cmd->tx_desc.num_sge;
878 	send_wr->send_flags = IB_SEND_SIGNALED;
879 }
880 
881 static int
882 isert_login_post_recv(struct isert_conn *isert_conn)
883 {
884 	struct ib_recv_wr rx_wr;
885 	struct ib_sge sge;
886 	int ret;
887 
888 	memset(&sge, 0, sizeof(struct ib_sge));
889 	sge.addr = isert_conn->login_desc->dma_addr +
890 		isert_get_hdr_offset(isert_conn->login_desc);
891 	sge.length = ISER_RX_PAYLOAD_SIZE;
892 	sge.lkey = isert_conn->device->pd->local_dma_lkey;
893 
894 	isert_dbg("Setup sge: addr: %llx length: %d 0x%08x\n",
895 		sge.addr, sge.length, sge.lkey);
896 
897 	isert_conn->login_desc->rx_cqe.done = isert_login_recv_done;
898 
899 	memset(&rx_wr, 0, sizeof(struct ib_recv_wr));
900 	rx_wr.wr_cqe = &isert_conn->login_desc->rx_cqe;
901 	rx_wr.sg_list = &sge;
902 	rx_wr.num_sge = 1;
903 
904 	ret = ib_post_recv(isert_conn->qp, &rx_wr, NULL);
905 	if (ret)
906 		isert_err("ib_post_recv() failed: %d\n", ret);
907 
908 	return ret;
909 }
910 
911 static int
912 isert_put_login_tx(struct iscsi_conn *conn, struct iscsi_login *login,
913 		   u32 length)
914 {
915 	struct isert_conn *isert_conn = conn->context;
916 	struct isert_device *device = isert_conn->device;
917 	struct ib_device *ib_dev = device->ib_device;
918 	struct iser_tx_desc *tx_desc = &isert_conn->login_tx_desc;
919 	int ret;
920 
921 	__isert_create_send_desc(device, tx_desc);
922 
923 	memcpy(&tx_desc->iscsi_header, &login->rsp[0],
924 	       sizeof(struct iscsi_hdr));
925 
926 	isert_init_tx_hdrs(isert_conn, tx_desc);
927 
928 	if (length > 0) {
929 		struct ib_sge *tx_dsg = &tx_desc->tx_sg[1];
930 
931 		ib_dma_sync_single_for_cpu(ib_dev, isert_conn->login_rsp_dma,
932 					   length, DMA_TO_DEVICE);
933 
934 		memcpy(isert_conn->login_rsp_buf, login->rsp_buf, length);
935 
936 		ib_dma_sync_single_for_device(ib_dev, isert_conn->login_rsp_dma,
937 					      length, DMA_TO_DEVICE);
938 
939 		tx_dsg->addr	= isert_conn->login_rsp_dma;
940 		tx_dsg->length	= length;
941 		tx_dsg->lkey	= isert_conn->device->pd->local_dma_lkey;
942 		tx_desc->num_sge = 2;
943 	}
944 	if (!login->login_failed) {
945 		if (login->login_complete) {
946 			ret = isert_alloc_rx_descriptors(isert_conn);
947 			if (ret)
948 				return ret;
949 
950 			ret = isert_post_recvm(isert_conn,
951 					       ISERT_QP_MAX_RECV_DTOS);
952 			if (ret)
953 				return ret;
954 
955 			/* Now we are in FULL_FEATURE phase */
956 			mutex_lock(&isert_conn->mutex);
957 			isert_conn->state = ISER_CONN_FULL_FEATURE;
958 			mutex_unlock(&isert_conn->mutex);
959 			goto post_send;
960 		}
961 
962 		ret = isert_login_post_recv(isert_conn);
963 		if (ret)
964 			return ret;
965 	}
966 post_send:
967 	ret = isert_login_post_send(isert_conn, tx_desc);
968 	if (ret)
969 		return ret;
970 
971 	return 0;
972 }
973 
974 static void
975 isert_rx_login_req(struct isert_conn *isert_conn)
976 {
977 	struct iser_rx_desc *rx_desc = isert_conn->login_desc;
978 	int rx_buflen = isert_conn->login_req_len;
979 	struct iscsi_conn *conn = isert_conn->conn;
980 	struct iscsi_login *login = conn->conn_login;
981 	int size;
982 
983 	isert_info("conn %p\n", isert_conn);
984 
985 	WARN_ON_ONCE(!login);
986 
987 	if (login->first_request) {
988 		struct iscsi_login_req *login_req =
989 			(struct iscsi_login_req *)isert_get_iscsi_hdr(rx_desc);
990 		/*
991 		 * Setup the initial iscsi_login values from the leading
992 		 * login request PDU.
993 		 */
994 		login->leading_connection = (!login_req->tsih) ? 1 : 0;
995 		login->current_stage =
996 			(login_req->flags & ISCSI_FLAG_LOGIN_CURRENT_STAGE_MASK)
997 			 >> 2;
998 		login->version_min	= login_req->min_version;
999 		login->version_max	= login_req->max_version;
1000 		memcpy(login->isid, login_req->isid, 6);
1001 		login->cmd_sn		= be32_to_cpu(login_req->cmdsn);
1002 		login->init_task_tag	= login_req->itt;
1003 		login->initial_exp_statsn = be32_to_cpu(login_req->exp_statsn);
1004 		login->cid		= be16_to_cpu(login_req->cid);
1005 		login->tsih		= be16_to_cpu(login_req->tsih);
1006 	}
1007 
1008 	memcpy(&login->req[0], isert_get_iscsi_hdr(rx_desc), ISCSI_HDR_LEN);
1009 
1010 	size = min(rx_buflen, MAX_KEY_VALUE_PAIRS);
1011 	isert_dbg("Using login payload size: %d, rx_buflen: %d "
1012 		  "MAX_KEY_VALUE_PAIRS: %d\n", size, rx_buflen,
1013 		  MAX_KEY_VALUE_PAIRS);
1014 	memcpy(login->req_buf, isert_get_data(rx_desc), size);
1015 
1016 	if (login->first_request) {
1017 		complete(&isert_conn->login_comp);
1018 		return;
1019 	}
1020 	schedule_delayed_work(&conn->login_work, 0);
1021 }
1022 
1023 static struct iscsi_cmd
1024 *isert_allocate_cmd(struct iscsi_conn *conn, struct iser_rx_desc *rx_desc)
1025 {
1026 	struct isert_conn *isert_conn = conn->context;
1027 	struct isert_cmd *isert_cmd;
1028 	struct iscsi_cmd *cmd;
1029 
1030 	cmd = iscsit_allocate_cmd(conn, TASK_INTERRUPTIBLE);
1031 	if (!cmd) {
1032 		isert_err("Unable to allocate iscsi_cmd + isert_cmd\n");
1033 		return NULL;
1034 	}
1035 	isert_cmd = iscsit_priv_cmd(cmd);
1036 	isert_cmd->conn = isert_conn;
1037 	isert_cmd->iscsi_cmd = cmd;
1038 	isert_cmd->rx_desc = rx_desc;
1039 
1040 	return cmd;
1041 }
1042 
1043 static int
1044 isert_handle_scsi_cmd(struct isert_conn *isert_conn,
1045 		      struct isert_cmd *isert_cmd, struct iscsi_cmd *cmd,
1046 		      struct iser_rx_desc *rx_desc, unsigned char *buf)
1047 {
1048 	struct iscsi_conn *conn = isert_conn->conn;
1049 	struct iscsi_scsi_req *hdr = (struct iscsi_scsi_req *)buf;
1050 	int imm_data, imm_data_len, unsol_data, sg_nents, rc;
1051 	bool dump_payload = false;
1052 	unsigned int data_len;
1053 
1054 	rc = iscsit_setup_scsi_cmd(conn, cmd, buf);
1055 	if (rc < 0)
1056 		return rc;
1057 
1058 	imm_data = cmd->immediate_data;
1059 	imm_data_len = cmd->first_burst_len;
1060 	unsol_data = cmd->unsolicited_data;
1061 	data_len = cmd->se_cmd.data_length;
1062 
1063 	if (imm_data && imm_data_len == data_len)
1064 		cmd->se_cmd.se_cmd_flags |= SCF_PASSTHROUGH_SG_TO_MEM_NOALLOC;
1065 	rc = iscsit_process_scsi_cmd(conn, cmd, hdr);
1066 	if (rc < 0) {
1067 		return 0;
1068 	} else if (rc > 0) {
1069 		dump_payload = true;
1070 		goto sequence_cmd;
1071 	}
1072 
1073 	if (!imm_data)
1074 		return 0;
1075 
1076 	if (imm_data_len != data_len) {
1077 		sg_nents = max(1UL, DIV_ROUND_UP(imm_data_len, PAGE_SIZE));
1078 		sg_copy_from_buffer(cmd->se_cmd.t_data_sg, sg_nents,
1079 				    isert_get_data(rx_desc), imm_data_len);
1080 		isert_dbg("Copy Immediate sg_nents: %u imm_data_len: %d\n",
1081 			  sg_nents, imm_data_len);
1082 	} else {
1083 		sg_init_table(&isert_cmd->sg, 1);
1084 		cmd->se_cmd.t_data_sg = &isert_cmd->sg;
1085 		cmd->se_cmd.t_data_nents = 1;
1086 		sg_set_buf(&isert_cmd->sg, isert_get_data(rx_desc),
1087 				imm_data_len);
1088 		isert_dbg("Transfer Immediate imm_data_len: %d\n",
1089 			  imm_data_len);
1090 	}
1091 
1092 	cmd->write_data_done += imm_data_len;
1093 
1094 	if (cmd->write_data_done == cmd->se_cmd.data_length) {
1095 		spin_lock_bh(&cmd->istate_lock);
1096 		cmd->cmd_flags |= ICF_GOT_LAST_DATAOUT;
1097 		cmd->i_state = ISTATE_RECEIVED_LAST_DATAOUT;
1098 		spin_unlock_bh(&cmd->istate_lock);
1099 	}
1100 
1101 sequence_cmd:
1102 	rc = iscsit_sequence_cmd(conn, cmd, buf, hdr->cmdsn);
1103 
1104 	if (!rc && !dump_payload && unsol_data)
1105 		iscsit_set_unsolicited_dataout(cmd);
1106 	else if (dump_payload && imm_data)
1107 		target_put_sess_cmd(&cmd->se_cmd);
1108 
1109 	return 0;
1110 }
1111 
1112 static int
1113 isert_handle_iscsi_dataout(struct isert_conn *isert_conn,
1114 			   struct iser_rx_desc *rx_desc, unsigned char *buf)
1115 {
1116 	struct scatterlist *sg_start;
1117 	struct iscsi_conn *conn = isert_conn->conn;
1118 	struct iscsi_cmd *cmd = NULL;
1119 	struct iscsi_data *hdr = (struct iscsi_data *)buf;
1120 	u32 unsol_data_len = ntoh24(hdr->dlength);
1121 	int rc, sg_nents, sg_off, page_off;
1122 
1123 	rc = iscsit_check_dataout_hdr(conn, buf, &cmd);
1124 	if (rc < 0)
1125 		return rc;
1126 	else if (!cmd)
1127 		return 0;
1128 	/*
1129 	 * FIXME: Unexpected unsolicited_data out
1130 	 */
1131 	if (!cmd->unsolicited_data) {
1132 		isert_err("Received unexpected solicited data payload\n");
1133 		dump_stack();
1134 		return -1;
1135 	}
1136 
1137 	isert_dbg("Unsolicited DataOut unsol_data_len: %u, "
1138 		  "write_data_done: %u, data_length: %u\n",
1139 		  unsol_data_len,  cmd->write_data_done,
1140 		  cmd->se_cmd.data_length);
1141 
1142 	sg_off = cmd->write_data_done / PAGE_SIZE;
1143 	sg_start = &cmd->se_cmd.t_data_sg[sg_off];
1144 	sg_nents = max(1UL, DIV_ROUND_UP(unsol_data_len, PAGE_SIZE));
1145 	page_off = cmd->write_data_done % PAGE_SIZE;
1146 	/*
1147 	 * FIXME: Non page-aligned unsolicited_data out
1148 	 */
1149 	if (page_off) {
1150 		isert_err("unexpected non-page aligned data payload\n");
1151 		dump_stack();
1152 		return -1;
1153 	}
1154 	isert_dbg("Copying DataOut: sg_start: %p, sg_off: %u "
1155 		  "sg_nents: %u from %p %u\n", sg_start, sg_off,
1156 		  sg_nents, isert_get_data(rx_desc), unsol_data_len);
1157 
1158 	sg_copy_from_buffer(sg_start, sg_nents, isert_get_data(rx_desc),
1159 			    unsol_data_len);
1160 
1161 	rc = iscsit_check_dataout_payload(cmd, hdr, false);
1162 	if (rc < 0)
1163 		return rc;
1164 
1165 	/*
1166 	 * multiple data-outs on the same command can arrive -
1167 	 * so post the buffer before hand
1168 	 */
1169 	return isert_post_recv(isert_conn, rx_desc);
1170 }
1171 
1172 static int
1173 isert_handle_nop_out(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
1174 		     struct iscsi_cmd *cmd, struct iser_rx_desc *rx_desc,
1175 		     unsigned char *buf)
1176 {
1177 	struct iscsi_conn *conn = isert_conn->conn;
1178 	struct iscsi_nopout *hdr = (struct iscsi_nopout *)buf;
1179 	int rc;
1180 
1181 	rc = iscsit_setup_nop_out(conn, cmd, hdr);
1182 	if (rc < 0)
1183 		return rc;
1184 	/*
1185 	 * FIXME: Add support for NOPOUT payload using unsolicited RDMA payload
1186 	 */
1187 
1188 	return iscsit_process_nop_out(conn, cmd, hdr);
1189 }
1190 
1191 static int
1192 isert_handle_text_cmd(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
1193 		      struct iscsi_cmd *cmd, struct iser_rx_desc *rx_desc,
1194 		      struct iscsi_text *hdr)
1195 {
1196 	struct iscsi_conn *conn = isert_conn->conn;
1197 	u32 payload_length = ntoh24(hdr->dlength);
1198 	int rc;
1199 	unsigned char *text_in = NULL;
1200 
1201 	rc = iscsit_setup_text_cmd(conn, cmd, hdr);
1202 	if (rc < 0)
1203 		return rc;
1204 
1205 	if (payload_length) {
1206 		text_in = kzalloc(payload_length, GFP_KERNEL);
1207 		if (!text_in)
1208 			return -ENOMEM;
1209 	}
1210 	cmd->text_in_ptr = text_in;
1211 
1212 	memcpy(cmd->text_in_ptr, isert_get_data(rx_desc), payload_length);
1213 
1214 	return iscsit_process_text_cmd(conn, cmd, hdr);
1215 }
1216 
1217 static int
1218 isert_rx_opcode(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc,
1219 		uint32_t read_stag, uint64_t read_va,
1220 		uint32_t write_stag, uint64_t write_va)
1221 {
1222 	struct iscsi_hdr *hdr = isert_get_iscsi_hdr(rx_desc);
1223 	struct iscsi_conn *conn = isert_conn->conn;
1224 	struct iscsi_cmd *cmd;
1225 	struct isert_cmd *isert_cmd;
1226 	int ret = -EINVAL;
1227 	u8 opcode = (hdr->opcode & ISCSI_OPCODE_MASK);
1228 
1229 	if (conn->sess->sess_ops->SessionType &&
1230 	   (!(opcode & ISCSI_OP_TEXT) || !(opcode & ISCSI_OP_LOGOUT))) {
1231 		isert_err("Got illegal opcode: 0x%02x in SessionType=Discovery,"
1232 			  " ignoring\n", opcode);
1233 		return 0;
1234 	}
1235 
1236 	switch (opcode) {
1237 	case ISCSI_OP_SCSI_CMD:
1238 		cmd = isert_allocate_cmd(conn, rx_desc);
1239 		if (!cmd)
1240 			break;
1241 
1242 		isert_cmd = iscsit_priv_cmd(cmd);
1243 		isert_cmd->read_stag = read_stag;
1244 		isert_cmd->read_va = read_va;
1245 		isert_cmd->write_stag = write_stag;
1246 		isert_cmd->write_va = write_va;
1247 		isert_cmd->inv_rkey = read_stag ? read_stag : write_stag;
1248 
1249 		ret = isert_handle_scsi_cmd(isert_conn, isert_cmd, cmd,
1250 					rx_desc, (unsigned char *)hdr);
1251 		break;
1252 	case ISCSI_OP_NOOP_OUT:
1253 		cmd = isert_allocate_cmd(conn, rx_desc);
1254 		if (!cmd)
1255 			break;
1256 
1257 		isert_cmd = iscsit_priv_cmd(cmd);
1258 		ret = isert_handle_nop_out(isert_conn, isert_cmd, cmd,
1259 					   rx_desc, (unsigned char *)hdr);
1260 		break;
1261 	case ISCSI_OP_SCSI_DATA_OUT:
1262 		ret = isert_handle_iscsi_dataout(isert_conn, rx_desc,
1263 						(unsigned char *)hdr);
1264 		break;
1265 	case ISCSI_OP_SCSI_TMFUNC:
1266 		cmd = isert_allocate_cmd(conn, rx_desc);
1267 		if (!cmd)
1268 			break;
1269 
1270 		ret = iscsit_handle_task_mgt_cmd(conn, cmd,
1271 						(unsigned char *)hdr);
1272 		break;
1273 	case ISCSI_OP_LOGOUT:
1274 		cmd = isert_allocate_cmd(conn, rx_desc);
1275 		if (!cmd)
1276 			break;
1277 
1278 		ret = iscsit_handle_logout_cmd(conn, cmd, (unsigned char *)hdr);
1279 		break;
1280 	case ISCSI_OP_TEXT:
1281 		if (be32_to_cpu(hdr->ttt) != 0xFFFFFFFF)
1282 			cmd = iscsit_find_cmd_from_itt(conn, hdr->itt);
1283 		else
1284 			cmd = isert_allocate_cmd(conn, rx_desc);
1285 
1286 		if (!cmd)
1287 			break;
1288 
1289 		isert_cmd = iscsit_priv_cmd(cmd);
1290 		ret = isert_handle_text_cmd(isert_conn, isert_cmd, cmd,
1291 					    rx_desc, (struct iscsi_text *)hdr);
1292 		break;
1293 	default:
1294 		isert_err("Got unknown iSCSI OpCode: 0x%02x\n", opcode);
1295 		dump_stack();
1296 		break;
1297 	}
1298 
1299 	return ret;
1300 }
1301 
1302 static void
1303 isert_print_wc(struct ib_wc *wc, const char *type)
1304 {
1305 	if (wc->status != IB_WC_WR_FLUSH_ERR)
1306 		isert_err("%s failure: %s (%d) vend_err %x\n", type,
1307 			  ib_wc_status_msg(wc->status), wc->status,
1308 			  wc->vendor_err);
1309 	else
1310 		isert_dbg("%s failure: %s (%d)\n", type,
1311 			  ib_wc_status_msg(wc->status), wc->status);
1312 }
1313 
1314 static void
1315 isert_recv_done(struct ib_cq *cq, struct ib_wc *wc)
1316 {
1317 	struct isert_conn *isert_conn = wc->qp->qp_context;
1318 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1319 	struct iser_rx_desc *rx_desc = cqe_to_rx_desc(wc->wr_cqe);
1320 	struct iscsi_hdr *hdr = isert_get_iscsi_hdr(rx_desc);
1321 	struct iser_ctrl *iser_ctrl = isert_get_iser_hdr(rx_desc);
1322 	uint64_t read_va = 0, write_va = 0;
1323 	uint32_t read_stag = 0, write_stag = 0;
1324 
1325 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1326 		isert_print_wc(wc, "recv");
1327 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1328 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1329 		return;
1330 	}
1331 
1332 	rx_desc->in_use = true;
1333 
1334 	ib_dma_sync_single_for_cpu(ib_dev, rx_desc->dma_addr,
1335 			ISER_RX_SIZE, DMA_FROM_DEVICE);
1336 
1337 	isert_dbg("DMA: 0x%llx, iSCSI opcode: 0x%02x, ITT: 0x%08x, flags: 0x%02x dlen: %d\n",
1338 		 rx_desc->dma_addr, hdr->opcode, hdr->itt, hdr->flags,
1339 		 (int)(wc->byte_len - ISER_HEADERS_LEN));
1340 
1341 	switch (iser_ctrl->flags & 0xF0) {
1342 	case ISCSI_CTRL:
1343 		if (iser_ctrl->flags & ISER_RSV) {
1344 			read_stag = be32_to_cpu(iser_ctrl->read_stag);
1345 			read_va = be64_to_cpu(iser_ctrl->read_va);
1346 			isert_dbg("ISER_RSV: read_stag: 0x%x read_va: 0x%llx\n",
1347 				  read_stag, (unsigned long long)read_va);
1348 		}
1349 		if (iser_ctrl->flags & ISER_WSV) {
1350 			write_stag = be32_to_cpu(iser_ctrl->write_stag);
1351 			write_va = be64_to_cpu(iser_ctrl->write_va);
1352 			isert_dbg("ISER_WSV: write_stag: 0x%x write_va: 0x%llx\n",
1353 				  write_stag, (unsigned long long)write_va);
1354 		}
1355 
1356 		isert_dbg("ISER ISCSI_CTRL PDU\n");
1357 		break;
1358 	case ISER_HELLO:
1359 		isert_err("iSER Hello message\n");
1360 		break;
1361 	default:
1362 		isert_warn("Unknown iSER hdr flags: 0x%02x\n", iser_ctrl->flags);
1363 		break;
1364 	}
1365 
1366 	isert_rx_opcode(isert_conn, rx_desc,
1367 			read_stag, read_va, write_stag, write_va);
1368 
1369 	ib_dma_sync_single_for_device(ib_dev, rx_desc->dma_addr,
1370 			ISER_RX_SIZE, DMA_FROM_DEVICE);
1371 }
1372 
1373 static void
1374 isert_login_recv_done(struct ib_cq *cq, struct ib_wc *wc)
1375 {
1376 	struct isert_conn *isert_conn = wc->qp->qp_context;
1377 	struct ib_device *ib_dev = isert_conn->device->ib_device;
1378 
1379 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1380 		isert_print_wc(wc, "login recv");
1381 		return;
1382 	}
1383 
1384 	ib_dma_sync_single_for_cpu(ib_dev, isert_conn->login_desc->dma_addr,
1385 			ISER_RX_SIZE, DMA_FROM_DEVICE);
1386 
1387 	isert_conn->login_req_len = wc->byte_len - ISER_HEADERS_LEN;
1388 
1389 	if (isert_conn->conn) {
1390 		struct iscsi_login *login = isert_conn->conn->conn_login;
1391 
1392 		if (login && !login->first_request)
1393 			isert_rx_login_req(isert_conn);
1394 	}
1395 
1396 	mutex_lock(&isert_conn->mutex);
1397 	complete(&isert_conn->login_req_comp);
1398 	mutex_unlock(&isert_conn->mutex);
1399 
1400 	ib_dma_sync_single_for_device(ib_dev, isert_conn->login_desc->dma_addr,
1401 				ISER_RX_SIZE, DMA_FROM_DEVICE);
1402 }
1403 
1404 static void
1405 isert_rdma_rw_ctx_destroy(struct isert_cmd *cmd, struct isert_conn *conn)
1406 {
1407 	struct se_cmd *se_cmd = &cmd->iscsi_cmd->se_cmd;
1408 	enum dma_data_direction dir = target_reverse_dma_direction(se_cmd);
1409 
1410 	if (!cmd->rw.nr_ops)
1411 		return;
1412 
1413 	if (isert_prot_cmd(conn, se_cmd)) {
1414 		rdma_rw_ctx_destroy_signature(&cmd->rw, conn->qp,
1415 				conn->cm_id->port_num, se_cmd->t_data_sg,
1416 				se_cmd->t_data_nents, se_cmd->t_prot_sg,
1417 				se_cmd->t_prot_nents, dir);
1418 	} else {
1419 		rdma_rw_ctx_destroy(&cmd->rw, conn->qp, conn->cm_id->port_num,
1420 				se_cmd->t_data_sg, se_cmd->t_data_nents, dir);
1421 	}
1422 
1423 	cmd->rw.nr_ops = 0;
1424 }
1425 
1426 static void
1427 isert_put_cmd(struct isert_cmd *isert_cmd, bool comp_err)
1428 {
1429 	struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd;
1430 	struct isert_conn *isert_conn = isert_cmd->conn;
1431 	struct iscsi_conn *conn = isert_conn->conn;
1432 	struct iscsi_text_rsp *hdr;
1433 
1434 	isert_dbg("Cmd %p\n", isert_cmd);
1435 
1436 	switch (cmd->iscsi_opcode) {
1437 	case ISCSI_OP_SCSI_CMD:
1438 		spin_lock_bh(&conn->cmd_lock);
1439 		if (!list_empty(&cmd->i_conn_node))
1440 			list_del_init(&cmd->i_conn_node);
1441 		spin_unlock_bh(&conn->cmd_lock);
1442 
1443 		if (cmd->data_direction == DMA_TO_DEVICE) {
1444 			iscsit_stop_dataout_timer(cmd);
1445 			/*
1446 			 * Check for special case during comp_err where
1447 			 * WRITE_PENDING has been handed off from core,
1448 			 * but requires an extra target_put_sess_cmd()
1449 			 * before transport_generic_free_cmd() below.
1450 			 */
1451 			if (comp_err &&
1452 			    cmd->se_cmd.t_state == TRANSPORT_WRITE_PENDING) {
1453 				struct se_cmd *se_cmd = &cmd->se_cmd;
1454 
1455 				target_put_sess_cmd(se_cmd);
1456 			}
1457 		}
1458 
1459 		isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1460 		transport_generic_free_cmd(&cmd->se_cmd, 0);
1461 		break;
1462 	case ISCSI_OP_SCSI_TMFUNC:
1463 		spin_lock_bh(&conn->cmd_lock);
1464 		if (!list_empty(&cmd->i_conn_node))
1465 			list_del_init(&cmd->i_conn_node);
1466 		spin_unlock_bh(&conn->cmd_lock);
1467 
1468 		transport_generic_free_cmd(&cmd->se_cmd, 0);
1469 		break;
1470 	case ISCSI_OP_REJECT:
1471 	case ISCSI_OP_NOOP_OUT:
1472 	case ISCSI_OP_TEXT:
1473 		hdr = (struct iscsi_text_rsp *)&isert_cmd->tx_desc.iscsi_header;
1474 		/* If the continue bit is on, keep the command alive */
1475 		if (hdr->flags & ISCSI_FLAG_TEXT_CONTINUE)
1476 			break;
1477 
1478 		spin_lock_bh(&conn->cmd_lock);
1479 		if (!list_empty(&cmd->i_conn_node))
1480 			list_del_init(&cmd->i_conn_node);
1481 		spin_unlock_bh(&conn->cmd_lock);
1482 
1483 		/*
1484 		 * Handle special case for REJECT when iscsi_add_reject*() has
1485 		 * overwritten the original iscsi_opcode assignment, and the
1486 		 * associated cmd->se_cmd needs to be released.
1487 		 */
1488 		if (cmd->se_cmd.se_tfo != NULL) {
1489 			isert_dbg("Calling transport_generic_free_cmd for 0x%02x\n",
1490 				 cmd->iscsi_opcode);
1491 			transport_generic_free_cmd(&cmd->se_cmd, 0);
1492 			break;
1493 		}
1494 		fallthrough;
1495 	default:
1496 		iscsit_release_cmd(cmd);
1497 		break;
1498 	}
1499 }
1500 
1501 static void
1502 isert_unmap_tx_desc(struct iser_tx_desc *tx_desc, struct ib_device *ib_dev)
1503 {
1504 	if (tx_desc->dma_addr != 0) {
1505 		isert_dbg("unmap single for tx_desc->dma_addr\n");
1506 		ib_dma_unmap_single(ib_dev, tx_desc->dma_addr,
1507 				    ISER_HEADERS_LEN, DMA_TO_DEVICE);
1508 		tx_desc->dma_addr = 0;
1509 	}
1510 }
1511 
1512 static void
1513 isert_completion_put(struct iser_tx_desc *tx_desc, struct isert_cmd *isert_cmd,
1514 		     struct ib_device *ib_dev, bool comp_err)
1515 {
1516 	if (isert_cmd->pdu_buf_dma != 0) {
1517 		isert_dbg("unmap single for isert_cmd->pdu_buf_dma\n");
1518 		ib_dma_unmap_single(ib_dev, isert_cmd->pdu_buf_dma,
1519 				    isert_cmd->pdu_buf_len, DMA_TO_DEVICE);
1520 		isert_cmd->pdu_buf_dma = 0;
1521 	}
1522 
1523 	isert_unmap_tx_desc(tx_desc, ib_dev);
1524 	isert_put_cmd(isert_cmd, comp_err);
1525 }
1526 
1527 static int
1528 isert_check_pi_status(struct se_cmd *se_cmd, struct ib_mr *sig_mr)
1529 {
1530 	struct ib_mr_status mr_status;
1531 	int ret;
1532 
1533 	ret = ib_check_mr_status(sig_mr, IB_MR_CHECK_SIG_STATUS, &mr_status);
1534 	if (ret) {
1535 		isert_err("ib_check_mr_status failed, ret %d\n", ret);
1536 		goto fail_mr_status;
1537 	}
1538 
1539 	if (mr_status.fail_status & IB_MR_CHECK_SIG_STATUS) {
1540 		u64 sec_offset_err;
1541 		u32 block_size = se_cmd->se_dev->dev_attrib.block_size + 8;
1542 
1543 		switch (mr_status.sig_err.err_type) {
1544 		case IB_SIG_BAD_GUARD:
1545 			se_cmd->pi_err = TCM_LOGICAL_BLOCK_GUARD_CHECK_FAILED;
1546 			break;
1547 		case IB_SIG_BAD_REFTAG:
1548 			se_cmd->pi_err = TCM_LOGICAL_BLOCK_REF_TAG_CHECK_FAILED;
1549 			break;
1550 		case IB_SIG_BAD_APPTAG:
1551 			se_cmd->pi_err = TCM_LOGICAL_BLOCK_APP_TAG_CHECK_FAILED;
1552 			break;
1553 		}
1554 		sec_offset_err = mr_status.sig_err.sig_err_offset;
1555 		do_div(sec_offset_err, block_size);
1556 		se_cmd->sense_info = sec_offset_err + se_cmd->t_task_lba;
1557 
1558 		isert_err("PI error found type %d at sector 0x%llx "
1559 			  "expected 0x%x vs actual 0x%x\n",
1560 			  mr_status.sig_err.err_type,
1561 			  (unsigned long long)se_cmd->sense_info,
1562 			  mr_status.sig_err.expected,
1563 			  mr_status.sig_err.actual);
1564 		ret = 1;
1565 	}
1566 
1567 fail_mr_status:
1568 	return ret;
1569 }
1570 
1571 static void
1572 isert_rdma_write_done(struct ib_cq *cq, struct ib_wc *wc)
1573 {
1574 	struct isert_conn *isert_conn = wc->qp->qp_context;
1575 	struct isert_device *device = isert_conn->device;
1576 	struct iser_tx_desc *desc = cqe_to_tx_desc(wc->wr_cqe);
1577 	struct isert_cmd *isert_cmd = tx_desc_to_cmd(desc);
1578 	struct se_cmd *cmd = &isert_cmd->iscsi_cmd->se_cmd;
1579 	int ret = 0;
1580 
1581 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1582 		isert_print_wc(wc, "rdma write");
1583 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1584 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1585 		isert_completion_put(desc, isert_cmd, device->ib_device, true);
1586 		return;
1587 	}
1588 
1589 	isert_dbg("Cmd %p\n", isert_cmd);
1590 
1591 	ret = isert_check_pi_status(cmd, isert_cmd->rw.reg->mr);
1592 	isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1593 
1594 	if (ret) {
1595 		/*
1596 		 * transport_generic_request_failure() expects to have
1597 		 * plus two references to handle queue-full, so re-add
1598 		 * one here as target-core will have already dropped
1599 		 * it after the first isert_put_datain() callback.
1600 		 */
1601 		kref_get(&cmd->cmd_kref);
1602 		transport_generic_request_failure(cmd, cmd->pi_err);
1603 	} else {
1604 		/*
1605 		 * XXX: isert_put_response() failure is not retried.
1606 		 */
1607 		ret = isert_put_response(isert_conn->conn, isert_cmd->iscsi_cmd);
1608 		if (ret)
1609 			pr_warn_ratelimited("isert_put_response() ret: %d\n", ret);
1610 	}
1611 }
1612 
1613 static void
1614 isert_rdma_read_done(struct ib_cq *cq, struct ib_wc *wc)
1615 {
1616 	struct isert_conn *isert_conn = wc->qp->qp_context;
1617 	struct isert_device *device = isert_conn->device;
1618 	struct iser_tx_desc *desc = cqe_to_tx_desc(wc->wr_cqe);
1619 	struct isert_cmd *isert_cmd = tx_desc_to_cmd(desc);
1620 	struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd;
1621 	struct se_cmd *se_cmd = &cmd->se_cmd;
1622 	int ret = 0;
1623 
1624 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1625 		isert_print_wc(wc, "rdma read");
1626 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1627 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1628 		isert_completion_put(desc, isert_cmd, device->ib_device, true);
1629 		return;
1630 	}
1631 
1632 	isert_dbg("Cmd %p\n", isert_cmd);
1633 
1634 	iscsit_stop_dataout_timer(cmd);
1635 
1636 	if (isert_prot_cmd(isert_conn, se_cmd))
1637 		ret = isert_check_pi_status(se_cmd, isert_cmd->rw.reg->mr);
1638 	isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1639 	cmd->write_data_done = 0;
1640 
1641 	isert_dbg("Cmd: %p RDMA_READ comp calling execute_cmd\n", isert_cmd);
1642 	spin_lock_bh(&cmd->istate_lock);
1643 	cmd->cmd_flags |= ICF_GOT_LAST_DATAOUT;
1644 	cmd->i_state = ISTATE_RECEIVED_LAST_DATAOUT;
1645 	spin_unlock_bh(&cmd->istate_lock);
1646 
1647 	/*
1648 	 * transport_generic_request_failure() will drop the extra
1649 	 * se_cmd->cmd_kref reference after T10-PI error, and handle
1650 	 * any non-zero ->queue_status() callback error retries.
1651 	 */
1652 	if (ret)
1653 		transport_generic_request_failure(se_cmd, se_cmd->pi_err);
1654 	else
1655 		target_execute_cmd(se_cmd);
1656 }
1657 
1658 static void
1659 isert_do_control_comp(struct work_struct *work)
1660 {
1661 	struct isert_cmd *isert_cmd = container_of(work,
1662 			struct isert_cmd, comp_work);
1663 	struct isert_conn *isert_conn = isert_cmd->conn;
1664 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1665 	struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd;
1666 
1667 	isert_dbg("Cmd %p i_state %d\n", isert_cmd, cmd->i_state);
1668 
1669 	switch (cmd->i_state) {
1670 	case ISTATE_SEND_TASKMGTRSP:
1671 		iscsit_tmr_post_handler(cmd, cmd->conn);
1672 		fallthrough;
1673 	case ISTATE_SEND_REJECT:
1674 	case ISTATE_SEND_TEXTRSP:
1675 		cmd->i_state = ISTATE_SENT_STATUS;
1676 		isert_completion_put(&isert_cmd->tx_desc, isert_cmd,
1677 				     ib_dev, false);
1678 		break;
1679 	case ISTATE_SEND_LOGOUTRSP:
1680 		iscsit_logout_post_handler(cmd, cmd->conn);
1681 		break;
1682 	default:
1683 		isert_err("Unknown i_state %d\n", cmd->i_state);
1684 		dump_stack();
1685 		break;
1686 	}
1687 }
1688 
1689 static void
1690 isert_login_send_done(struct ib_cq *cq, struct ib_wc *wc)
1691 {
1692 	struct isert_conn *isert_conn = wc->qp->qp_context;
1693 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1694 	struct iser_tx_desc *tx_desc = cqe_to_tx_desc(wc->wr_cqe);
1695 
1696 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1697 		isert_print_wc(wc, "login send");
1698 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1699 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1700 	}
1701 
1702 	isert_unmap_tx_desc(tx_desc, ib_dev);
1703 }
1704 
1705 static void
1706 isert_send_done(struct ib_cq *cq, struct ib_wc *wc)
1707 {
1708 	struct isert_conn *isert_conn = wc->qp->qp_context;
1709 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1710 	struct iser_tx_desc *tx_desc = cqe_to_tx_desc(wc->wr_cqe);
1711 	struct isert_cmd *isert_cmd = tx_desc_to_cmd(tx_desc);
1712 
1713 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1714 		isert_print_wc(wc, "send");
1715 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1716 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1717 		isert_completion_put(tx_desc, isert_cmd, ib_dev, true);
1718 		return;
1719 	}
1720 
1721 	isert_dbg("Cmd %p\n", isert_cmd);
1722 
1723 	switch (isert_cmd->iscsi_cmd->i_state) {
1724 	case ISTATE_SEND_TASKMGTRSP:
1725 	case ISTATE_SEND_LOGOUTRSP:
1726 	case ISTATE_SEND_REJECT:
1727 	case ISTATE_SEND_TEXTRSP:
1728 		isert_unmap_tx_desc(tx_desc, ib_dev);
1729 
1730 		INIT_WORK(&isert_cmd->comp_work, isert_do_control_comp);
1731 		queue_work(isert_comp_wq, &isert_cmd->comp_work);
1732 		return;
1733 	default:
1734 		isert_cmd->iscsi_cmd->i_state = ISTATE_SENT_STATUS;
1735 		isert_completion_put(tx_desc, isert_cmd, ib_dev, false);
1736 		break;
1737 	}
1738 }
1739 
1740 static int
1741 isert_post_response(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd)
1742 {
1743 	int ret;
1744 
1745 	ret = isert_post_recv(isert_conn, isert_cmd->rx_desc);
1746 	if (ret)
1747 		return ret;
1748 
1749 	ret = ib_post_send(isert_conn->qp, &isert_cmd->tx_desc.send_wr, NULL);
1750 	if (ret) {
1751 		isert_err("ib_post_send failed with %d\n", ret);
1752 		return ret;
1753 	}
1754 	return ret;
1755 }
1756 
1757 static int
1758 isert_put_response(struct iscsi_conn *conn, struct iscsi_cmd *cmd)
1759 {
1760 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1761 	struct isert_conn *isert_conn = conn->context;
1762 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1763 	struct iscsi_scsi_rsp *hdr = (struct iscsi_scsi_rsp *)
1764 				&isert_cmd->tx_desc.iscsi_header;
1765 
1766 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1767 	iscsit_build_rsp_pdu(cmd, conn, true, hdr);
1768 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1769 	/*
1770 	 * Attach SENSE DATA payload to iSCSI Response PDU
1771 	 */
1772 	if (cmd->se_cmd.sense_buffer &&
1773 	    ((cmd->se_cmd.se_cmd_flags & SCF_TRANSPORT_TASK_SENSE) ||
1774 	    (cmd->se_cmd.se_cmd_flags & SCF_EMULATED_TASK_SENSE))) {
1775 		struct isert_device *device = isert_conn->device;
1776 		struct ib_device *ib_dev = device->ib_device;
1777 		struct ib_sge *tx_dsg = &isert_cmd->tx_desc.tx_sg[1];
1778 		u32 padding, pdu_len;
1779 
1780 		put_unaligned_be16(cmd->se_cmd.scsi_sense_length,
1781 				   cmd->sense_buffer);
1782 		cmd->se_cmd.scsi_sense_length += sizeof(__be16);
1783 
1784 		padding = -(cmd->se_cmd.scsi_sense_length) & 3;
1785 		hton24(hdr->dlength, (u32)cmd->se_cmd.scsi_sense_length);
1786 		pdu_len = cmd->se_cmd.scsi_sense_length + padding;
1787 
1788 		isert_cmd->pdu_buf_dma = ib_dma_map_single(ib_dev,
1789 				(void *)cmd->sense_buffer, pdu_len,
1790 				DMA_TO_DEVICE);
1791 		if (ib_dma_mapping_error(ib_dev, isert_cmd->pdu_buf_dma))
1792 			return -ENOMEM;
1793 
1794 		isert_cmd->pdu_buf_len = pdu_len;
1795 		tx_dsg->addr	= isert_cmd->pdu_buf_dma;
1796 		tx_dsg->length	= pdu_len;
1797 		tx_dsg->lkey	= device->pd->local_dma_lkey;
1798 		isert_cmd->tx_desc.num_sge = 2;
1799 	}
1800 
1801 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1802 
1803 	isert_dbg("Posting SCSI Response\n");
1804 
1805 	return isert_post_response(isert_conn, isert_cmd);
1806 }
1807 
1808 static void
1809 isert_aborted_task(struct iscsi_conn *conn, struct iscsi_cmd *cmd)
1810 {
1811 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1812 	struct isert_conn *isert_conn = conn->context;
1813 
1814 	spin_lock_bh(&conn->cmd_lock);
1815 	if (!list_empty(&cmd->i_conn_node))
1816 		list_del_init(&cmd->i_conn_node);
1817 	spin_unlock_bh(&conn->cmd_lock);
1818 
1819 	if (cmd->data_direction == DMA_TO_DEVICE)
1820 		iscsit_stop_dataout_timer(cmd);
1821 	isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1822 }
1823 
1824 static enum target_prot_op
1825 isert_get_sup_prot_ops(struct iscsi_conn *conn)
1826 {
1827 	struct isert_conn *isert_conn = conn->context;
1828 	struct isert_device *device = isert_conn->device;
1829 
1830 	if (conn->tpg->tpg_attrib.t10_pi) {
1831 		if (device->pi_capable) {
1832 			isert_info("conn %p PI offload enabled\n", isert_conn);
1833 			isert_conn->pi_support = true;
1834 			return TARGET_PROT_ALL;
1835 		}
1836 	}
1837 
1838 	isert_info("conn %p PI offload disabled\n", isert_conn);
1839 	isert_conn->pi_support = false;
1840 
1841 	return TARGET_PROT_NORMAL;
1842 }
1843 
1844 static int
1845 isert_put_nopin(struct iscsi_cmd *cmd, struct iscsi_conn *conn,
1846 		bool nopout_response)
1847 {
1848 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1849 	struct isert_conn *isert_conn = conn->context;
1850 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1851 
1852 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1853 	iscsit_build_nopin_rsp(cmd, conn, (struct iscsi_nopin *)
1854 			       &isert_cmd->tx_desc.iscsi_header,
1855 			       nopout_response);
1856 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1857 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1858 
1859 	isert_dbg("conn %p Posting NOPIN Response\n", isert_conn);
1860 
1861 	return isert_post_response(isert_conn, isert_cmd);
1862 }
1863 
1864 static int
1865 isert_put_logout_rsp(struct iscsi_cmd *cmd, struct iscsi_conn *conn)
1866 {
1867 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1868 	struct isert_conn *isert_conn = conn->context;
1869 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1870 
1871 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1872 	iscsit_build_logout_rsp(cmd, conn, (struct iscsi_logout_rsp *)
1873 				&isert_cmd->tx_desc.iscsi_header);
1874 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1875 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1876 
1877 	isert_dbg("conn %p Posting Logout Response\n", isert_conn);
1878 
1879 	return isert_post_response(isert_conn, isert_cmd);
1880 }
1881 
1882 static int
1883 isert_put_tm_rsp(struct iscsi_cmd *cmd, struct iscsi_conn *conn)
1884 {
1885 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1886 	struct isert_conn *isert_conn = conn->context;
1887 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1888 
1889 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1890 	iscsit_build_task_mgt_rsp(cmd, conn, (struct iscsi_tm_rsp *)
1891 				  &isert_cmd->tx_desc.iscsi_header);
1892 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1893 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1894 
1895 	isert_dbg("conn %p Posting Task Management Response\n", isert_conn);
1896 
1897 	return isert_post_response(isert_conn, isert_cmd);
1898 }
1899 
1900 static int
1901 isert_put_reject(struct iscsi_cmd *cmd, struct iscsi_conn *conn)
1902 {
1903 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1904 	struct isert_conn *isert_conn = conn->context;
1905 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1906 	struct isert_device *device = isert_conn->device;
1907 	struct ib_device *ib_dev = device->ib_device;
1908 	struct ib_sge *tx_dsg = &isert_cmd->tx_desc.tx_sg[1];
1909 	struct iscsi_reject *hdr =
1910 		(struct iscsi_reject *)&isert_cmd->tx_desc.iscsi_header;
1911 
1912 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1913 	iscsit_build_reject(cmd, conn, hdr);
1914 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1915 
1916 	hton24(hdr->dlength, ISCSI_HDR_LEN);
1917 	isert_cmd->pdu_buf_dma = ib_dma_map_single(ib_dev,
1918 			(void *)cmd->buf_ptr, ISCSI_HDR_LEN,
1919 			DMA_TO_DEVICE);
1920 	if (ib_dma_mapping_error(ib_dev, isert_cmd->pdu_buf_dma))
1921 		return -ENOMEM;
1922 	isert_cmd->pdu_buf_len = ISCSI_HDR_LEN;
1923 	tx_dsg->addr	= isert_cmd->pdu_buf_dma;
1924 	tx_dsg->length	= ISCSI_HDR_LEN;
1925 	tx_dsg->lkey	= device->pd->local_dma_lkey;
1926 	isert_cmd->tx_desc.num_sge = 2;
1927 
1928 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1929 
1930 	isert_dbg("conn %p Posting Reject\n", isert_conn);
1931 
1932 	return isert_post_response(isert_conn, isert_cmd);
1933 }
1934 
1935 static int
1936 isert_put_text_rsp(struct iscsi_cmd *cmd, struct iscsi_conn *conn)
1937 {
1938 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1939 	struct isert_conn *isert_conn = conn->context;
1940 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1941 	struct iscsi_text_rsp *hdr =
1942 		(struct iscsi_text_rsp *)&isert_cmd->tx_desc.iscsi_header;
1943 	u32 txt_rsp_len;
1944 	int rc;
1945 
1946 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1947 	rc = iscsit_build_text_rsp(cmd, conn, hdr, ISCSI_INFINIBAND);
1948 	if (rc < 0)
1949 		return rc;
1950 
1951 	txt_rsp_len = rc;
1952 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1953 
1954 	if (txt_rsp_len) {
1955 		struct isert_device *device = isert_conn->device;
1956 		struct ib_device *ib_dev = device->ib_device;
1957 		struct ib_sge *tx_dsg = &isert_cmd->tx_desc.tx_sg[1];
1958 		void *txt_rsp_buf = cmd->buf_ptr;
1959 
1960 		isert_cmd->pdu_buf_dma = ib_dma_map_single(ib_dev,
1961 				txt_rsp_buf, txt_rsp_len, DMA_TO_DEVICE);
1962 		if (ib_dma_mapping_error(ib_dev, isert_cmd->pdu_buf_dma))
1963 			return -ENOMEM;
1964 
1965 		isert_cmd->pdu_buf_len = txt_rsp_len;
1966 		tx_dsg->addr	= isert_cmd->pdu_buf_dma;
1967 		tx_dsg->length	= txt_rsp_len;
1968 		tx_dsg->lkey	= device->pd->local_dma_lkey;
1969 		isert_cmd->tx_desc.num_sge = 2;
1970 	}
1971 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1972 
1973 	isert_dbg("conn %p Text Response\n", isert_conn);
1974 
1975 	return isert_post_response(isert_conn, isert_cmd);
1976 }
1977 
1978 static inline void
1979 isert_set_dif_domain(struct se_cmd *se_cmd, struct ib_sig_domain *domain)
1980 {
1981 	domain->sig_type = IB_SIG_TYPE_T10_DIF;
1982 	domain->sig.dif.bg_type = IB_T10DIF_CRC;
1983 	domain->sig.dif.pi_interval = se_cmd->se_dev->dev_attrib.block_size;
1984 	domain->sig.dif.ref_tag = se_cmd->reftag_seed;
1985 	/*
1986 	 * At the moment we hard code those, but if in the future
1987 	 * the target core would like to use it, we will take it
1988 	 * from se_cmd.
1989 	 */
1990 	domain->sig.dif.apptag_check_mask = 0xffff;
1991 	domain->sig.dif.app_escape = true;
1992 	domain->sig.dif.ref_escape = true;
1993 	if (se_cmd->prot_type == TARGET_DIF_TYPE1_PROT ||
1994 	    se_cmd->prot_type == TARGET_DIF_TYPE2_PROT)
1995 		domain->sig.dif.ref_remap = true;
1996 };
1997 
1998 static int
1999 isert_set_sig_attrs(struct se_cmd *se_cmd, struct ib_sig_attrs *sig_attrs)
2000 {
2001 	memset(sig_attrs, 0, sizeof(*sig_attrs));
2002 
2003 	switch (se_cmd->prot_op) {
2004 	case TARGET_PROT_DIN_INSERT:
2005 	case TARGET_PROT_DOUT_STRIP:
2006 		sig_attrs->mem.sig_type = IB_SIG_TYPE_NONE;
2007 		isert_set_dif_domain(se_cmd, &sig_attrs->wire);
2008 		break;
2009 	case TARGET_PROT_DOUT_INSERT:
2010 	case TARGET_PROT_DIN_STRIP:
2011 		sig_attrs->wire.sig_type = IB_SIG_TYPE_NONE;
2012 		isert_set_dif_domain(se_cmd, &sig_attrs->mem);
2013 		break;
2014 	case TARGET_PROT_DIN_PASS:
2015 	case TARGET_PROT_DOUT_PASS:
2016 		isert_set_dif_domain(se_cmd, &sig_attrs->wire);
2017 		isert_set_dif_domain(se_cmd, &sig_attrs->mem);
2018 		break;
2019 	default:
2020 		isert_err("Unsupported PI operation %d\n", se_cmd->prot_op);
2021 		return -EINVAL;
2022 	}
2023 
2024 	if (se_cmd->prot_checks & TARGET_DIF_CHECK_GUARD)
2025 		sig_attrs->check_mask |= IB_SIG_CHECK_GUARD;
2026 	if (se_cmd->prot_checks & TARGET_DIF_CHECK_APPTAG)
2027 		sig_attrs->check_mask |= IB_SIG_CHECK_APPTAG;
2028 	if (se_cmd->prot_checks & TARGET_DIF_CHECK_REFTAG)
2029 		sig_attrs->check_mask |= IB_SIG_CHECK_REFTAG;
2030 
2031 	return 0;
2032 }
2033 
2034 static int
2035 isert_rdma_rw_ctx_post(struct isert_cmd *cmd, struct isert_conn *conn,
2036 		struct ib_cqe *cqe, struct ib_send_wr *chain_wr)
2037 {
2038 	struct se_cmd *se_cmd = &cmd->iscsi_cmd->se_cmd;
2039 	enum dma_data_direction dir = target_reverse_dma_direction(se_cmd);
2040 	u8 port_num = conn->cm_id->port_num;
2041 	u64 addr;
2042 	u32 rkey, offset;
2043 	int ret;
2044 
2045 	if (cmd->ctx_init_done)
2046 		goto rdma_ctx_post;
2047 
2048 	if (dir == DMA_FROM_DEVICE) {
2049 		addr = cmd->write_va;
2050 		rkey = cmd->write_stag;
2051 		offset = cmd->iscsi_cmd->write_data_done;
2052 	} else {
2053 		addr = cmd->read_va;
2054 		rkey = cmd->read_stag;
2055 		offset = 0;
2056 	}
2057 
2058 	if (isert_prot_cmd(conn, se_cmd)) {
2059 		struct ib_sig_attrs sig_attrs;
2060 
2061 		ret = isert_set_sig_attrs(se_cmd, &sig_attrs);
2062 		if (ret)
2063 			return ret;
2064 
2065 		WARN_ON_ONCE(offset);
2066 		ret = rdma_rw_ctx_signature_init(&cmd->rw, conn->qp, port_num,
2067 				se_cmd->t_data_sg, se_cmd->t_data_nents,
2068 				se_cmd->t_prot_sg, se_cmd->t_prot_nents,
2069 				&sig_attrs, addr, rkey, dir);
2070 	} else {
2071 		ret = rdma_rw_ctx_init(&cmd->rw, conn->qp, port_num,
2072 				se_cmd->t_data_sg, se_cmd->t_data_nents,
2073 				offset, addr, rkey, dir);
2074 	}
2075 
2076 	if (ret < 0) {
2077 		isert_err("Cmd: %p failed to prepare RDMA res\n", cmd);
2078 		return ret;
2079 	}
2080 
2081 	cmd->ctx_init_done = true;
2082 
2083 rdma_ctx_post:
2084 	ret = rdma_rw_ctx_post(&cmd->rw, conn->qp, port_num, cqe, chain_wr);
2085 	if (ret < 0)
2086 		isert_err("Cmd: %p failed to post RDMA res\n", cmd);
2087 	return ret;
2088 }
2089 
2090 static int
2091 isert_put_datain(struct iscsi_conn *conn, struct iscsi_cmd *cmd)
2092 {
2093 	struct se_cmd *se_cmd = &cmd->se_cmd;
2094 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2095 	struct isert_conn *isert_conn = conn->context;
2096 	struct ib_cqe *cqe = NULL;
2097 	struct ib_send_wr *chain_wr = NULL;
2098 	int rc;
2099 
2100 	isert_dbg("Cmd: %p RDMA_WRITE data_length: %u\n",
2101 		 isert_cmd, se_cmd->data_length);
2102 
2103 	if (isert_prot_cmd(isert_conn, se_cmd)) {
2104 		isert_cmd->tx_desc.tx_cqe.done = isert_rdma_write_done;
2105 		cqe = &isert_cmd->tx_desc.tx_cqe;
2106 	} else {
2107 		/*
2108 		 * Build isert_conn->tx_desc for iSCSI response PDU and attach
2109 		 */
2110 		isert_create_send_desc(isert_conn, isert_cmd,
2111 				       &isert_cmd->tx_desc);
2112 		iscsit_build_rsp_pdu(cmd, conn, true, (struct iscsi_scsi_rsp *)
2113 				     &isert_cmd->tx_desc.iscsi_header);
2114 		isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
2115 		isert_init_send_wr(isert_conn, isert_cmd,
2116 				   &isert_cmd->tx_desc.send_wr);
2117 
2118 		rc = isert_post_recv(isert_conn, isert_cmd->rx_desc);
2119 		if (rc)
2120 			return rc;
2121 
2122 		chain_wr = &isert_cmd->tx_desc.send_wr;
2123 	}
2124 
2125 	rc = isert_rdma_rw_ctx_post(isert_cmd, isert_conn, cqe, chain_wr);
2126 	isert_dbg("Cmd: %p posted RDMA_WRITE for iSER Data READ rc: %d\n",
2127 		  isert_cmd, rc);
2128 	return rc;
2129 }
2130 
2131 static int
2132 isert_get_dataout(struct iscsi_conn *conn, struct iscsi_cmd *cmd, bool recovery)
2133 {
2134 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2135 	int ret;
2136 
2137 	isert_dbg("Cmd: %p RDMA_READ data_length: %u write_data_done: %u\n",
2138 		 isert_cmd, cmd->se_cmd.data_length, cmd->write_data_done);
2139 
2140 	isert_cmd->tx_desc.tx_cqe.done = isert_rdma_read_done;
2141 	ret = isert_rdma_rw_ctx_post(isert_cmd, conn->context,
2142 				     &isert_cmd->tx_desc.tx_cqe, NULL);
2143 
2144 	isert_dbg("Cmd: %p posted RDMA_READ memory for ISER Data WRITE rc: %d\n",
2145 		 isert_cmd, ret);
2146 	return ret;
2147 }
2148 
2149 static int
2150 isert_immediate_queue(struct iscsi_conn *conn, struct iscsi_cmd *cmd, int state)
2151 {
2152 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2153 	int ret = 0;
2154 
2155 	switch (state) {
2156 	case ISTATE_REMOVE:
2157 		spin_lock_bh(&conn->cmd_lock);
2158 		list_del_init(&cmd->i_conn_node);
2159 		spin_unlock_bh(&conn->cmd_lock);
2160 		isert_put_cmd(isert_cmd, true);
2161 		break;
2162 	case ISTATE_SEND_NOPIN_WANT_RESPONSE:
2163 		ret = isert_put_nopin(cmd, conn, false);
2164 		break;
2165 	default:
2166 		isert_err("Unknown immediate state: 0x%02x\n", state);
2167 		ret = -EINVAL;
2168 		break;
2169 	}
2170 
2171 	return ret;
2172 }
2173 
2174 static int
2175 isert_response_queue(struct iscsi_conn *conn, struct iscsi_cmd *cmd, int state)
2176 {
2177 	struct isert_conn *isert_conn = conn->context;
2178 	int ret;
2179 
2180 	switch (state) {
2181 	case ISTATE_SEND_LOGOUTRSP:
2182 		ret = isert_put_logout_rsp(cmd, conn);
2183 		if (!ret)
2184 			isert_conn->logout_posted = true;
2185 		break;
2186 	case ISTATE_SEND_NOPIN:
2187 		ret = isert_put_nopin(cmd, conn, true);
2188 		break;
2189 	case ISTATE_SEND_TASKMGTRSP:
2190 		ret = isert_put_tm_rsp(cmd, conn);
2191 		break;
2192 	case ISTATE_SEND_REJECT:
2193 		ret = isert_put_reject(cmd, conn);
2194 		break;
2195 	case ISTATE_SEND_TEXTRSP:
2196 		ret = isert_put_text_rsp(cmd, conn);
2197 		break;
2198 	case ISTATE_SEND_STATUS:
2199 		/*
2200 		 * Special case for sending non GOOD SCSI status from TX thread
2201 		 * context during pre se_cmd excecution failure.
2202 		 */
2203 		ret = isert_put_response(conn, cmd);
2204 		break;
2205 	default:
2206 		isert_err("Unknown response state: 0x%02x\n", state);
2207 		ret = -EINVAL;
2208 		break;
2209 	}
2210 
2211 	return ret;
2212 }
2213 
2214 struct rdma_cm_id *
2215 isert_setup_id(struct isert_np *isert_np)
2216 {
2217 	struct iscsi_np *np = isert_np->np;
2218 	struct rdma_cm_id *id;
2219 	struct sockaddr *sa;
2220 	int ret;
2221 
2222 	sa = (struct sockaddr *)&np->np_sockaddr;
2223 	isert_dbg("ksockaddr: %p, sa: %p\n", &np->np_sockaddr, sa);
2224 
2225 	id = rdma_create_id(&init_net, isert_cma_handler, isert_np,
2226 			    RDMA_PS_TCP, IB_QPT_RC);
2227 	if (IS_ERR(id)) {
2228 		isert_err("rdma_create_id() failed: %ld\n", PTR_ERR(id));
2229 		ret = PTR_ERR(id);
2230 		goto out;
2231 	}
2232 	isert_dbg("id %p context %p\n", id, id->context);
2233 
2234 	ret = rdma_bind_addr(id, sa);
2235 	if (ret) {
2236 		isert_err("rdma_bind_addr() failed: %d\n", ret);
2237 		goto out_id;
2238 	}
2239 
2240 	ret = rdma_listen(id, 0);
2241 	if (ret) {
2242 		isert_err("rdma_listen() failed: %d\n", ret);
2243 		goto out_id;
2244 	}
2245 
2246 	return id;
2247 out_id:
2248 	rdma_destroy_id(id);
2249 out:
2250 	return ERR_PTR(ret);
2251 }
2252 
2253 static int
2254 isert_setup_np(struct iscsi_np *np,
2255 	       struct sockaddr_storage *ksockaddr)
2256 {
2257 	struct isert_np *isert_np;
2258 	struct rdma_cm_id *isert_lid;
2259 	int ret;
2260 
2261 	isert_np = kzalloc(sizeof(struct isert_np), GFP_KERNEL);
2262 	if (!isert_np)
2263 		return -ENOMEM;
2264 
2265 	sema_init(&isert_np->sem, 0);
2266 	mutex_init(&isert_np->mutex);
2267 	INIT_LIST_HEAD(&isert_np->accepted);
2268 	INIT_LIST_HEAD(&isert_np->pending);
2269 	isert_np->np = np;
2270 
2271 	/*
2272 	 * Setup the np->np_sockaddr from the passed sockaddr setup
2273 	 * in iscsi_target_configfs.c code..
2274 	 */
2275 	memcpy(&np->np_sockaddr, ksockaddr,
2276 	       sizeof(struct sockaddr_storage));
2277 
2278 	isert_lid = isert_setup_id(isert_np);
2279 	if (IS_ERR(isert_lid)) {
2280 		ret = PTR_ERR(isert_lid);
2281 		goto out;
2282 	}
2283 
2284 	isert_np->cm_id = isert_lid;
2285 	np->np_context = isert_np;
2286 
2287 	return 0;
2288 
2289 out:
2290 	kfree(isert_np);
2291 
2292 	return ret;
2293 }
2294 
2295 static int
2296 isert_rdma_accept(struct isert_conn *isert_conn)
2297 {
2298 	struct rdma_cm_id *cm_id = isert_conn->cm_id;
2299 	struct rdma_conn_param cp;
2300 	int ret;
2301 	struct iser_cm_hdr rsp_hdr;
2302 
2303 	memset(&cp, 0, sizeof(struct rdma_conn_param));
2304 	cp.initiator_depth = isert_conn->initiator_depth;
2305 	cp.retry_count = 7;
2306 	cp.rnr_retry_count = 7;
2307 
2308 	memset(&rsp_hdr, 0, sizeof(rsp_hdr));
2309 	rsp_hdr.flags = ISERT_ZBVA_NOT_USED;
2310 	if (!isert_conn->snd_w_inv)
2311 		rsp_hdr.flags = rsp_hdr.flags | ISERT_SEND_W_INV_NOT_USED;
2312 	cp.private_data = (void *)&rsp_hdr;
2313 	cp.private_data_len = sizeof(rsp_hdr);
2314 
2315 	ret = rdma_accept(cm_id, &cp);
2316 	if (ret) {
2317 		isert_err("rdma_accept() failed with: %d\n", ret);
2318 		return ret;
2319 	}
2320 
2321 	return 0;
2322 }
2323 
2324 static int
2325 isert_get_login_rx(struct iscsi_conn *conn, struct iscsi_login *login)
2326 {
2327 	struct isert_conn *isert_conn = conn->context;
2328 	int ret;
2329 
2330 	isert_info("before login_req comp conn: %p\n", isert_conn);
2331 	ret = wait_for_completion_interruptible(&isert_conn->login_req_comp);
2332 	if (ret) {
2333 		isert_err("isert_conn %p interrupted before got login req\n",
2334 			  isert_conn);
2335 		return ret;
2336 	}
2337 	reinit_completion(&isert_conn->login_req_comp);
2338 
2339 	/*
2340 	 * For login requests after the first PDU, isert_rx_login_req() will
2341 	 * kick schedule_delayed_work(&conn->login_work) as the packet is
2342 	 * received, which turns this callback from iscsi_target_do_login_rx()
2343 	 * into a NOP.
2344 	 */
2345 	if (!login->first_request)
2346 		return 0;
2347 
2348 	isert_rx_login_req(isert_conn);
2349 
2350 	isert_info("before login_comp conn: %p\n", conn);
2351 	ret = wait_for_completion_interruptible(&isert_conn->login_comp);
2352 	if (ret)
2353 		return ret;
2354 
2355 	isert_info("processing login->req: %p\n", login->req);
2356 
2357 	return 0;
2358 }
2359 
2360 static void
2361 isert_set_conn_info(struct iscsi_np *np, struct iscsi_conn *conn,
2362 		    struct isert_conn *isert_conn)
2363 {
2364 	struct rdma_cm_id *cm_id = isert_conn->cm_id;
2365 	struct rdma_route *cm_route = &cm_id->route;
2366 
2367 	conn->login_family = np->np_sockaddr.ss_family;
2368 
2369 	conn->login_sockaddr = cm_route->addr.dst_addr;
2370 	conn->local_sockaddr = cm_route->addr.src_addr;
2371 }
2372 
2373 static int
2374 isert_accept_np(struct iscsi_np *np, struct iscsi_conn *conn)
2375 {
2376 	struct isert_np *isert_np = np->np_context;
2377 	struct isert_conn *isert_conn;
2378 	int ret;
2379 
2380 accept_wait:
2381 	ret = down_interruptible(&isert_np->sem);
2382 	if (ret)
2383 		return -ENODEV;
2384 
2385 	spin_lock_bh(&np->np_thread_lock);
2386 	if (np->np_thread_state >= ISCSI_NP_THREAD_RESET) {
2387 		spin_unlock_bh(&np->np_thread_lock);
2388 		isert_dbg("np_thread_state %d\n",
2389 			 np->np_thread_state);
2390 		/**
2391 		 * No point in stalling here when np_thread
2392 		 * is in state RESET/SHUTDOWN/EXIT - bail
2393 		 **/
2394 		return -ENODEV;
2395 	}
2396 	spin_unlock_bh(&np->np_thread_lock);
2397 
2398 	mutex_lock(&isert_np->mutex);
2399 	if (list_empty(&isert_np->pending)) {
2400 		mutex_unlock(&isert_np->mutex);
2401 		goto accept_wait;
2402 	}
2403 	isert_conn = list_first_entry(&isert_np->pending,
2404 			struct isert_conn, node);
2405 	list_del_init(&isert_conn->node);
2406 	mutex_unlock(&isert_np->mutex);
2407 
2408 	conn->context = isert_conn;
2409 	isert_conn->conn = conn;
2410 	isert_conn->state = ISER_CONN_BOUND;
2411 
2412 	isert_set_conn_info(np, conn, isert_conn);
2413 
2414 	isert_dbg("Processing isert_conn: %p\n", isert_conn);
2415 
2416 	return 0;
2417 }
2418 
2419 static void
2420 isert_free_np(struct iscsi_np *np)
2421 {
2422 	struct isert_np *isert_np = np->np_context;
2423 	struct isert_conn *isert_conn, *n;
2424 
2425 	if (isert_np->cm_id)
2426 		rdma_destroy_id(isert_np->cm_id);
2427 
2428 	/*
2429 	 * FIXME: At this point we don't have a good way to insure
2430 	 * that at this point we don't have hanging connections that
2431 	 * completed RDMA establishment but didn't start iscsi login
2432 	 * process. So work-around this by cleaning up what ever piled
2433 	 * up in accepted and pending lists.
2434 	 */
2435 	mutex_lock(&isert_np->mutex);
2436 	if (!list_empty(&isert_np->pending)) {
2437 		isert_info("Still have isert pending connections\n");
2438 		list_for_each_entry_safe(isert_conn, n,
2439 					 &isert_np->pending,
2440 					 node) {
2441 			isert_info("cleaning isert_conn %p state (%d)\n",
2442 				   isert_conn, isert_conn->state);
2443 			isert_connect_release(isert_conn);
2444 		}
2445 	}
2446 
2447 	if (!list_empty(&isert_np->accepted)) {
2448 		isert_info("Still have isert accepted connections\n");
2449 		list_for_each_entry_safe(isert_conn, n,
2450 					 &isert_np->accepted,
2451 					 node) {
2452 			isert_info("cleaning isert_conn %p state (%d)\n",
2453 				   isert_conn, isert_conn->state);
2454 			isert_connect_release(isert_conn);
2455 		}
2456 	}
2457 	mutex_unlock(&isert_np->mutex);
2458 
2459 	np->np_context = NULL;
2460 	kfree(isert_np);
2461 }
2462 
2463 static void isert_release_work(struct work_struct *work)
2464 {
2465 	struct isert_conn *isert_conn = container_of(work,
2466 						     struct isert_conn,
2467 						     release_work);
2468 
2469 	isert_info("Starting release conn %p\n", isert_conn);
2470 
2471 	mutex_lock(&isert_conn->mutex);
2472 	isert_conn->state = ISER_CONN_DOWN;
2473 	mutex_unlock(&isert_conn->mutex);
2474 
2475 	isert_info("Destroying conn %p\n", isert_conn);
2476 	isert_put_conn(isert_conn);
2477 }
2478 
2479 static void
2480 isert_wait4logout(struct isert_conn *isert_conn)
2481 {
2482 	struct iscsi_conn *conn = isert_conn->conn;
2483 
2484 	isert_info("conn %p\n", isert_conn);
2485 
2486 	if (isert_conn->logout_posted) {
2487 		isert_info("conn %p wait for conn_logout_comp\n", isert_conn);
2488 		wait_for_completion_timeout(&conn->conn_logout_comp,
2489 					    SECONDS_FOR_LOGOUT_COMP * HZ);
2490 	}
2491 }
2492 
2493 static void
2494 isert_wait4cmds(struct iscsi_conn *conn)
2495 {
2496 	isert_info("iscsi_conn %p\n", conn);
2497 
2498 	if (conn->sess) {
2499 		target_stop_session(conn->sess->se_sess);
2500 		target_wait_for_sess_cmds(conn->sess->se_sess);
2501 	}
2502 }
2503 
2504 /**
2505  * isert_put_unsol_pending_cmds() - Drop commands waiting for
2506  *     unsolicitate dataout
2507  * @conn:    iscsi connection
2508  *
2509  * We might still have commands that are waiting for unsolicited
2510  * dataouts messages. We must put the extra reference on those
2511  * before blocking on the target_wait_for_session_cmds
2512  */
2513 static void
2514 isert_put_unsol_pending_cmds(struct iscsi_conn *conn)
2515 {
2516 	struct iscsi_cmd *cmd, *tmp;
2517 	static LIST_HEAD(drop_cmd_list);
2518 
2519 	spin_lock_bh(&conn->cmd_lock);
2520 	list_for_each_entry_safe(cmd, tmp, &conn->conn_cmd_list, i_conn_node) {
2521 		if ((cmd->cmd_flags & ICF_NON_IMMEDIATE_UNSOLICITED_DATA) &&
2522 		    (cmd->write_data_done < conn->sess->sess_ops->FirstBurstLength) &&
2523 		    (cmd->write_data_done < cmd->se_cmd.data_length))
2524 			list_move_tail(&cmd->i_conn_node, &drop_cmd_list);
2525 	}
2526 	spin_unlock_bh(&conn->cmd_lock);
2527 
2528 	list_for_each_entry_safe(cmd, tmp, &drop_cmd_list, i_conn_node) {
2529 		list_del_init(&cmd->i_conn_node);
2530 		if (cmd->i_state != ISTATE_REMOVE) {
2531 			struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2532 
2533 			isert_info("conn %p dropping cmd %p\n", conn, cmd);
2534 			isert_put_cmd(isert_cmd, true);
2535 		}
2536 	}
2537 }
2538 
2539 static void isert_wait_conn(struct iscsi_conn *conn)
2540 {
2541 	struct isert_conn *isert_conn = conn->context;
2542 
2543 	isert_info("Starting conn %p\n", isert_conn);
2544 
2545 	mutex_lock(&isert_conn->mutex);
2546 	isert_conn_terminate(isert_conn);
2547 	mutex_unlock(&isert_conn->mutex);
2548 
2549 	ib_drain_qp(isert_conn->qp);
2550 	isert_put_unsol_pending_cmds(conn);
2551 	isert_wait4cmds(conn);
2552 	isert_wait4logout(isert_conn);
2553 
2554 	queue_work(isert_release_wq, &isert_conn->release_work);
2555 }
2556 
2557 static void isert_free_conn(struct iscsi_conn *conn)
2558 {
2559 	struct isert_conn *isert_conn = conn->context;
2560 
2561 	ib_drain_qp(isert_conn->qp);
2562 	isert_put_conn(isert_conn);
2563 }
2564 
2565 static void isert_get_rx_pdu(struct iscsi_conn *conn)
2566 {
2567 	struct completion comp;
2568 
2569 	init_completion(&comp);
2570 
2571 	wait_for_completion_interruptible(&comp);
2572 }
2573 
2574 static struct iscsit_transport iser_target_transport = {
2575 	.name			= "IB/iSER",
2576 	.transport_type		= ISCSI_INFINIBAND,
2577 	.rdma_shutdown		= true,
2578 	.priv_size		= sizeof(struct isert_cmd),
2579 	.owner			= THIS_MODULE,
2580 	.iscsit_setup_np	= isert_setup_np,
2581 	.iscsit_accept_np	= isert_accept_np,
2582 	.iscsit_free_np		= isert_free_np,
2583 	.iscsit_wait_conn	= isert_wait_conn,
2584 	.iscsit_free_conn	= isert_free_conn,
2585 	.iscsit_get_login_rx	= isert_get_login_rx,
2586 	.iscsit_put_login_tx	= isert_put_login_tx,
2587 	.iscsit_immediate_queue	= isert_immediate_queue,
2588 	.iscsit_response_queue	= isert_response_queue,
2589 	.iscsit_get_dataout	= isert_get_dataout,
2590 	.iscsit_queue_data_in	= isert_put_datain,
2591 	.iscsit_queue_status	= isert_put_response,
2592 	.iscsit_aborted_task	= isert_aborted_task,
2593 	.iscsit_get_rx_pdu	= isert_get_rx_pdu,
2594 	.iscsit_get_sup_prot_ops = isert_get_sup_prot_ops,
2595 };
2596 
2597 static int __init isert_init(void)
2598 {
2599 	int ret;
2600 
2601 	isert_comp_wq = alloc_workqueue("isert_comp_wq",
2602 					WQ_UNBOUND | WQ_HIGHPRI, 0);
2603 	if (!isert_comp_wq) {
2604 		isert_err("Unable to allocate isert_comp_wq\n");
2605 		return -ENOMEM;
2606 	}
2607 
2608 	isert_release_wq = alloc_workqueue("isert_release_wq", WQ_UNBOUND,
2609 					WQ_UNBOUND_MAX_ACTIVE);
2610 	if (!isert_release_wq) {
2611 		isert_err("Unable to allocate isert_release_wq\n");
2612 		ret = -ENOMEM;
2613 		goto destroy_comp_wq;
2614 	}
2615 
2616 	iscsit_register_transport(&iser_target_transport);
2617 	isert_info("iSER_TARGET[0] - Loaded iser_target_transport\n");
2618 
2619 	return 0;
2620 
2621 destroy_comp_wq:
2622 	destroy_workqueue(isert_comp_wq);
2623 
2624 	return ret;
2625 }
2626 
2627 static void __exit isert_exit(void)
2628 {
2629 	flush_scheduled_work();
2630 	destroy_workqueue(isert_release_wq);
2631 	destroy_workqueue(isert_comp_wq);
2632 	iscsit_unregister_transport(&iser_target_transport);
2633 	isert_info("iSER_TARGET[0] - Released iser_target_transport\n");
2634 }
2635 
2636 MODULE_DESCRIPTION("iSER-Target for mainline target infrastructure");
2637 MODULE_AUTHOR("nab@Linux-iSCSI.org");
2638 MODULE_LICENSE("GPL");
2639 
2640 module_init(isert_init);
2641 module_exit(isert_exit);
2642