xref: /linux/drivers/infiniband/ulp/isert/ib_isert.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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_MIN_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: 128, max: 4096)");
42 
43 static DEFINE_MUTEX(device_list_mutex);
44 static LIST_HEAD(device_list);
45 static struct workqueue_struct *isert_login_wq;
46 static struct workqueue_struct *isert_comp_wq;
47 static struct workqueue_struct *isert_release_wq;
48 
49 static int
50 isert_put_response(struct iscsit_conn *conn, struct iscsit_cmd *cmd);
51 static int
52 isert_login_post_recv(struct isert_conn *isert_conn);
53 static int
54 isert_rdma_accept(struct isert_conn *isert_conn);
55 struct rdma_cm_id *isert_setup_id(struct isert_np *isert_np);
56 
57 static void isert_release_work(struct work_struct *work);
58 static void isert_recv_done(struct ib_cq *cq, struct ib_wc *wc);
59 static void isert_send_done(struct ib_cq *cq, struct ib_wc *wc);
60 static void isert_login_recv_done(struct ib_cq *cq, struct ib_wc *wc);
61 static void isert_login_send_done(struct ib_cq *cq, struct ib_wc *wc);
62 
63 static int isert_sg_tablesize_set(const char *val, const struct kernel_param *kp)
64 {
65 	int n = 0, ret;
66 
67 	ret = kstrtoint(val, 10, &n);
68 	if (ret != 0 || n < ISCSI_ISER_MIN_SG_TABLESIZE ||
69 	    n > ISCSI_ISER_MAX_SG_TABLESIZE)
70 		return -EINVAL;
71 
72 	return param_set_int(val, kp);
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 static void
83 isert_qp_event_callback(struct ib_event *e, void *context)
84 {
85 	struct isert_conn *isert_conn = context;
86 
87 	isert_err("%s (%d): conn %p\n",
88 		  ib_event_msg(e->event), e->event, isert_conn);
89 
90 	switch (e->event) {
91 	case IB_EVENT_COMM_EST:
92 		rdma_notify(isert_conn->cm_id, IB_EVENT_COMM_EST);
93 		break;
94 	default:
95 		break;
96 	}
97 }
98 
99 static struct ib_qp *
100 isert_create_qp(struct isert_conn *isert_conn,
101 		struct rdma_cm_id *cma_id)
102 {
103 	u32 cq_size = ISERT_QP_MAX_REQ_DTOS + ISERT_QP_MAX_RECV_DTOS + 2;
104 	struct isert_device *device = isert_conn->device;
105 	struct ib_device *ib_dev = device->ib_device;
106 	struct ib_qp_init_attr attr;
107 	int ret, factor;
108 
109 	isert_conn->cq = ib_cq_pool_get(ib_dev, cq_size, -1, IB_POLL_WORKQUEUE);
110 	if (IS_ERR(isert_conn->cq)) {
111 		isert_err("Unable to allocate cq\n");
112 		ret = PTR_ERR(isert_conn->cq);
113 		return ERR_PTR(ret);
114 	}
115 	isert_conn->cq_size = cq_size;
116 
117 	memset(&attr, 0, sizeof(struct ib_qp_init_attr));
118 	attr.event_handler = isert_qp_event_callback;
119 	attr.qp_context = isert_conn;
120 	attr.send_cq = isert_conn->cq;
121 	attr.recv_cq = isert_conn->cq;
122 	attr.cap.max_send_wr = ISERT_QP_MAX_REQ_DTOS + 1;
123 	attr.cap.max_recv_wr = ISERT_QP_MAX_RECV_DTOS + 1;
124 	factor = rdma_rw_mr_factor(device->ib_device, cma_id->port_num,
125 				   isert_sg_tablesize);
126 	attr.cap.max_rdma_ctxs = ISCSI_DEF_XMIT_CMDS_MAX * factor;
127 	attr.cap.max_send_sge = device->ib_device->attrs.max_send_sge;
128 	attr.cap.max_recv_sge = 1;
129 	attr.sq_sig_type = IB_SIGNAL_REQ_WR;
130 	attr.qp_type = IB_QPT_RC;
131 	if (device->pi_capable)
132 		attr.create_flags |= IB_QP_CREATE_INTEGRITY_EN;
133 
134 	ret = rdma_create_qp(cma_id, device->pd, &attr);
135 	if (ret) {
136 		isert_err("rdma_create_qp failed for cma_id %d\n", ret);
137 		ib_cq_pool_put(isert_conn->cq, isert_conn->cq_size);
138 
139 		return ERR_PTR(ret);
140 	}
141 
142 	return cma_id->qp;
143 }
144 
145 static int
146 isert_alloc_rx_descriptors(struct isert_conn *isert_conn)
147 {
148 	struct isert_device *device = isert_conn->device;
149 	struct ib_device *ib_dev = device->ib_device;
150 	struct iser_rx_desc *rx_desc;
151 	struct ib_sge *rx_sg;
152 	u64 dma_addr;
153 	int i, j;
154 
155 	isert_conn->rx_descs = kcalloc(ISERT_QP_MAX_RECV_DTOS,
156 				       sizeof(struct iser_rx_desc),
157 				       GFP_KERNEL);
158 	if (!isert_conn->rx_descs)
159 		return -ENOMEM;
160 
161 	rx_desc = isert_conn->rx_descs;
162 
163 	for (i = 0; i < ISERT_QP_MAX_RECV_DTOS; i++, rx_desc++)  {
164 		dma_addr = ib_dma_map_single(ib_dev, rx_desc->buf,
165 					ISER_RX_SIZE, DMA_FROM_DEVICE);
166 		if (ib_dma_mapping_error(ib_dev, dma_addr))
167 			goto dma_map_fail;
168 
169 		rx_desc->dma_addr = dma_addr;
170 
171 		rx_sg = &rx_desc->rx_sg;
172 		rx_sg->addr = rx_desc->dma_addr + isert_get_hdr_offset(rx_desc);
173 		rx_sg->length = ISER_RX_PAYLOAD_SIZE;
174 		rx_sg->lkey = device->pd->local_dma_lkey;
175 		rx_desc->rx_cqe.done = isert_recv_done;
176 	}
177 
178 	return 0;
179 
180 dma_map_fail:
181 	rx_desc = isert_conn->rx_descs;
182 	for (j = 0; j < i; j++, rx_desc++) {
183 		ib_dma_unmap_single(ib_dev, rx_desc->dma_addr,
184 				    ISER_RX_SIZE, DMA_FROM_DEVICE);
185 	}
186 	kfree(isert_conn->rx_descs);
187 	isert_conn->rx_descs = NULL;
188 	isert_err("conn %p failed to allocate rx descriptors\n", isert_conn);
189 	return -ENOMEM;
190 }
191 
192 static void
193 isert_free_rx_descriptors(struct isert_conn *isert_conn)
194 {
195 	struct ib_device *ib_dev = isert_conn->device->ib_device;
196 	struct iser_rx_desc *rx_desc;
197 	int i;
198 
199 	if (!isert_conn->rx_descs)
200 		return;
201 
202 	rx_desc = isert_conn->rx_descs;
203 	for (i = 0; i < ISERT_QP_MAX_RECV_DTOS; i++, rx_desc++)  {
204 		ib_dma_unmap_single(ib_dev, rx_desc->dma_addr,
205 				    ISER_RX_SIZE, DMA_FROM_DEVICE);
206 	}
207 
208 	kfree(isert_conn->rx_descs);
209 	isert_conn->rx_descs = NULL;
210 }
211 
212 static int
213 isert_create_device_ib_res(struct isert_device *device)
214 {
215 	struct ib_device *ib_dev = device->ib_device;
216 	int ret;
217 
218 	isert_dbg("devattr->max_send_sge: %d devattr->max_recv_sge %d\n",
219 		  ib_dev->attrs.max_send_sge, ib_dev->attrs.max_recv_sge);
220 	isert_dbg("devattr->max_sge_rd: %d\n", ib_dev->attrs.max_sge_rd);
221 
222 	device->pd = ib_alloc_pd(ib_dev, 0);
223 	if (IS_ERR(device->pd)) {
224 		ret = PTR_ERR(device->pd);
225 		isert_err("failed to allocate pd, device %p, ret=%d\n",
226 			  device, ret);
227 		return ret;
228 	}
229 
230 	/* Check signature cap */
231 	if (ib_dev->attrs.kernel_cap_flags & IBK_INTEGRITY_HANDOVER)
232 		device->pi_capable = true;
233 	else
234 		device->pi_capable = false;
235 
236 	return 0;
237 }
238 
239 static void
240 isert_free_device_ib_res(struct isert_device *device)
241 {
242 	isert_info("device %p\n", device);
243 
244 	ib_dealloc_pd(device->pd);
245 }
246 
247 static void
248 isert_device_put(struct isert_device *device)
249 {
250 	mutex_lock(&device_list_mutex);
251 	device->refcount--;
252 	isert_info("device %p refcount %d\n", device, device->refcount);
253 	if (!device->refcount) {
254 		isert_free_device_ib_res(device);
255 		list_del(&device->dev_node);
256 		kfree(device);
257 	}
258 	mutex_unlock(&device_list_mutex);
259 }
260 
261 static struct isert_device *
262 isert_device_get(struct rdma_cm_id *cma_id)
263 {
264 	struct isert_device *device;
265 	int ret;
266 
267 	mutex_lock(&device_list_mutex);
268 	list_for_each_entry(device, &device_list, dev_node) {
269 		if (device->ib_device->node_guid == cma_id->device->node_guid) {
270 			device->refcount++;
271 			isert_info("Found iser device %p refcount %d\n",
272 				   device, device->refcount);
273 			mutex_unlock(&device_list_mutex);
274 			return device;
275 		}
276 	}
277 
278 	device = kzalloc(sizeof(struct isert_device), GFP_KERNEL);
279 	if (!device) {
280 		mutex_unlock(&device_list_mutex);
281 		return ERR_PTR(-ENOMEM);
282 	}
283 
284 	INIT_LIST_HEAD(&device->dev_node);
285 
286 	device->ib_device = cma_id->device;
287 	ret = isert_create_device_ib_res(device);
288 	if (ret) {
289 		kfree(device);
290 		mutex_unlock(&device_list_mutex);
291 		return ERR_PTR(ret);
292 	}
293 
294 	device->refcount++;
295 	list_add_tail(&device->dev_node, &device_list);
296 	isert_info("Created a new iser device %p refcount %d\n",
297 		   device, device->refcount);
298 	mutex_unlock(&device_list_mutex);
299 
300 	return device;
301 }
302 
303 static void
304 isert_init_conn(struct isert_conn *isert_conn)
305 {
306 	isert_conn->state = ISER_CONN_INIT;
307 	INIT_LIST_HEAD(&isert_conn->node);
308 	init_completion(&isert_conn->login_comp);
309 	init_completion(&isert_conn->login_req_comp);
310 	init_waitqueue_head(&isert_conn->rem_wait);
311 	kref_init(&isert_conn->kref);
312 	mutex_init(&isert_conn->mutex);
313 	INIT_WORK(&isert_conn->release_work, isert_release_work);
314 }
315 
316 static void
317 isert_free_login_buf(struct isert_conn *isert_conn)
318 {
319 	struct ib_device *ib_dev = isert_conn->device->ib_device;
320 
321 	ib_dma_unmap_single(ib_dev, isert_conn->login_rsp_dma,
322 			    ISER_RX_PAYLOAD_SIZE, DMA_TO_DEVICE);
323 	kfree(isert_conn->login_rsp_buf);
324 
325 	ib_dma_unmap_single(ib_dev, isert_conn->login_desc->dma_addr,
326 			    ISER_RX_SIZE, DMA_FROM_DEVICE);
327 	kfree(isert_conn->login_desc);
328 }
329 
330 static int
331 isert_alloc_login_buf(struct isert_conn *isert_conn,
332 		      struct ib_device *ib_dev)
333 {
334 	int ret;
335 
336 	isert_conn->login_desc = kzalloc(sizeof(*isert_conn->login_desc),
337 			GFP_KERNEL);
338 	if (!isert_conn->login_desc)
339 		return -ENOMEM;
340 
341 	isert_conn->login_desc->dma_addr = ib_dma_map_single(ib_dev,
342 				isert_conn->login_desc->buf,
343 				ISER_RX_SIZE, DMA_FROM_DEVICE);
344 	ret = ib_dma_mapping_error(ib_dev, isert_conn->login_desc->dma_addr);
345 	if (ret) {
346 		isert_err("login_desc dma mapping error: %d\n", ret);
347 		isert_conn->login_desc->dma_addr = 0;
348 		goto out_free_login_desc;
349 	}
350 
351 	isert_conn->login_rsp_buf = kzalloc(ISER_RX_PAYLOAD_SIZE, GFP_KERNEL);
352 	if (!isert_conn->login_rsp_buf) {
353 		ret = -ENOMEM;
354 		goto out_unmap_login_desc;
355 	}
356 
357 	isert_conn->login_rsp_dma = ib_dma_map_single(ib_dev,
358 					isert_conn->login_rsp_buf,
359 					ISER_RX_PAYLOAD_SIZE, DMA_TO_DEVICE);
360 	ret = ib_dma_mapping_error(ib_dev, isert_conn->login_rsp_dma);
361 	if (ret) {
362 		isert_err("login_rsp_dma mapping error: %d\n", ret);
363 		isert_conn->login_rsp_dma = 0;
364 		goto out_free_login_rsp_buf;
365 	}
366 
367 	return 0;
368 
369 out_free_login_rsp_buf:
370 	kfree(isert_conn->login_rsp_buf);
371 out_unmap_login_desc:
372 	ib_dma_unmap_single(ib_dev, isert_conn->login_desc->dma_addr,
373 			    ISER_RX_SIZE, DMA_FROM_DEVICE);
374 out_free_login_desc:
375 	kfree(isert_conn->login_desc);
376 	return ret;
377 }
378 
379 static void
380 isert_set_nego_params(struct isert_conn *isert_conn,
381 		      struct rdma_conn_param *param)
382 {
383 	struct ib_device_attr *attr = &isert_conn->device->ib_device->attrs;
384 
385 	/* Set max inflight RDMA READ requests */
386 	isert_conn->initiator_depth = min_t(u8, param->initiator_depth,
387 				attr->max_qp_init_rd_atom);
388 	isert_dbg("Using initiator_depth: %u\n", isert_conn->initiator_depth);
389 
390 	if (param->private_data) {
391 		u8 flags = *(u8 *)param->private_data;
392 
393 		/*
394 		 * use remote invalidation if the both initiator
395 		 * and the HCA support it
396 		 */
397 		isert_conn->snd_w_inv = !(flags & ISER_SEND_W_INV_NOT_SUP) &&
398 					  (attr->device_cap_flags &
399 					   IB_DEVICE_MEM_MGT_EXTENSIONS);
400 		if (isert_conn->snd_w_inv)
401 			isert_info("Using remote invalidation\n");
402 	}
403 }
404 
405 static void
406 isert_destroy_qp(struct isert_conn *isert_conn)
407 {
408 	ib_destroy_qp(isert_conn->qp);
409 	ib_cq_pool_put(isert_conn->cq, isert_conn->cq_size);
410 }
411 
412 static int
413 isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
414 {
415 	struct isert_np *isert_np = cma_id->context;
416 	struct iscsi_np *np = isert_np->np;
417 	struct isert_conn *isert_conn;
418 	struct isert_device *device;
419 	int ret = 0;
420 
421 	spin_lock_bh(&np->np_thread_lock);
422 	if (!np->enabled) {
423 		spin_unlock_bh(&np->np_thread_lock);
424 		isert_dbg("iscsi_np is not enabled, reject connect request\n");
425 		return rdma_reject(cma_id, NULL, 0, IB_CM_REJ_CONSUMER_DEFINED);
426 	}
427 	spin_unlock_bh(&np->np_thread_lock);
428 
429 	isert_dbg("cma_id: %p, portal: %p\n",
430 		 cma_id, cma_id->context);
431 
432 	isert_conn = kzalloc(sizeof(struct isert_conn), GFP_KERNEL);
433 	if (!isert_conn)
434 		return -ENOMEM;
435 
436 	isert_init_conn(isert_conn);
437 	isert_conn->cm_id = cma_id;
438 
439 	device = isert_device_get(cma_id);
440 	if (IS_ERR(device)) {
441 		ret = PTR_ERR(device);
442 		goto out;
443 	}
444 	isert_conn->device = device;
445 
446 	ret = isert_alloc_login_buf(isert_conn, cma_id->device);
447 	if (ret)
448 		goto out_conn_dev;
449 
450 	isert_set_nego_params(isert_conn, &event->param.conn);
451 
452 	isert_conn->qp = isert_create_qp(isert_conn, cma_id);
453 	if (IS_ERR(isert_conn->qp)) {
454 		ret = PTR_ERR(isert_conn->qp);
455 		goto out_rsp_dma_map;
456 	}
457 
458 	ret = isert_login_post_recv(isert_conn);
459 	if (ret)
460 		goto out_destroy_qp;
461 
462 	ret = isert_rdma_accept(isert_conn);
463 	if (ret)
464 		goto out_destroy_qp;
465 
466 	mutex_lock(&isert_np->mutex);
467 	list_add_tail(&isert_conn->node, &isert_np->accepted);
468 	mutex_unlock(&isert_np->mutex);
469 
470 	return 0;
471 
472 out_destroy_qp:
473 	isert_destroy_qp(isert_conn);
474 out_rsp_dma_map:
475 	isert_free_login_buf(isert_conn);
476 out_conn_dev:
477 	isert_device_put(device);
478 out:
479 	kfree(isert_conn);
480 	rdma_reject(cma_id, NULL, 0, IB_CM_REJ_CONSUMER_DEFINED);
481 	return ret;
482 }
483 
484 static void
485 isert_connect_release(struct isert_conn *isert_conn)
486 {
487 	struct isert_device *device = isert_conn->device;
488 
489 	isert_dbg("conn %p\n", isert_conn);
490 
491 	BUG_ON(!device);
492 
493 	isert_free_rx_descriptors(isert_conn);
494 	if (isert_conn->cm_id &&
495 	    !isert_conn->dev_removed)
496 		rdma_destroy_id(isert_conn->cm_id);
497 
498 	if (isert_conn->qp)
499 		isert_destroy_qp(isert_conn);
500 
501 	if (isert_conn->login_desc)
502 		isert_free_login_buf(isert_conn);
503 
504 	isert_device_put(device);
505 
506 	if (isert_conn->dev_removed)
507 		wake_up_interruptible(&isert_conn->rem_wait);
508 	else
509 		kfree(isert_conn);
510 }
511 
512 static void
513 isert_connected_handler(struct rdma_cm_id *cma_id)
514 {
515 	struct isert_conn *isert_conn = cma_id->qp->qp_context;
516 	struct isert_np *isert_np = cma_id->context;
517 
518 	isert_info("conn %p\n", isert_conn);
519 
520 	mutex_lock(&isert_conn->mutex);
521 	isert_conn->state = ISER_CONN_UP;
522 	kref_get(&isert_conn->kref);
523 	mutex_unlock(&isert_conn->mutex);
524 
525 	mutex_lock(&isert_np->mutex);
526 	list_move_tail(&isert_conn->node, &isert_np->pending);
527 	mutex_unlock(&isert_np->mutex);
528 
529 	isert_info("np %p: Allow accept_np to continue\n", isert_np);
530 	up(&isert_np->sem);
531 }
532 
533 static void
534 isert_release_kref(struct kref *kref)
535 {
536 	struct isert_conn *isert_conn = container_of(kref,
537 				struct isert_conn, kref);
538 
539 	isert_info("conn %p final kref %s/%d\n", isert_conn, current->comm,
540 		   current->pid);
541 
542 	isert_connect_release(isert_conn);
543 }
544 
545 static void
546 isert_put_conn(struct isert_conn *isert_conn)
547 {
548 	kref_put(&isert_conn->kref, isert_release_kref);
549 }
550 
551 static void
552 isert_handle_unbound_conn(struct isert_conn *isert_conn)
553 {
554 	struct isert_np *isert_np = isert_conn->cm_id->context;
555 
556 	mutex_lock(&isert_np->mutex);
557 	if (!list_empty(&isert_conn->node)) {
558 		/*
559 		 * This means iscsi doesn't know this connection
560 		 * so schedule a cleanup ourselves
561 		 */
562 		list_del_init(&isert_conn->node);
563 		isert_put_conn(isert_conn);
564 		queue_work(isert_release_wq, &isert_conn->release_work);
565 	}
566 	mutex_unlock(&isert_np->mutex);
567 }
568 
569 /**
570  * isert_conn_terminate() - Initiate connection termination
571  * @isert_conn: isert connection struct
572  *
573  * Notes:
574  * In case the connection state is BOUND, move state
575  * to TEMINATING and start teardown sequence (rdma_disconnect).
576  * In case the connection state is UP, complete flush as well.
577  *
578  * This routine must be called with mutex held. Thus it is
579  * safe to call multiple times.
580  */
581 static void
582 isert_conn_terminate(struct isert_conn *isert_conn)
583 {
584 	int err;
585 
586 	if (isert_conn->state >= ISER_CONN_TERMINATING)
587 		return;
588 
589 	isert_info("Terminating conn %p state %d\n",
590 		   isert_conn, isert_conn->state);
591 	isert_conn->state = ISER_CONN_TERMINATING;
592 	err = rdma_disconnect(isert_conn->cm_id);
593 	if (err)
594 		isert_warn("Failed rdma_disconnect isert_conn %p\n",
595 			   isert_conn);
596 }
597 
598 static int
599 isert_np_cma_handler(struct isert_np *isert_np,
600 		     enum rdma_cm_event_type event)
601 {
602 	isert_dbg("%s (%d): isert np %p\n",
603 		  rdma_event_msg(event), event, isert_np);
604 
605 	switch (event) {
606 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
607 		isert_np->cm_id = NULL;
608 		break;
609 	case RDMA_CM_EVENT_ADDR_CHANGE:
610 		isert_np->cm_id = isert_setup_id(isert_np);
611 		if (IS_ERR(isert_np->cm_id)) {
612 			isert_err("isert np %p setup id failed: %ld\n",
613 				  isert_np, PTR_ERR(isert_np->cm_id));
614 			isert_np->cm_id = NULL;
615 		}
616 		break;
617 	default:
618 		isert_err("isert np %p Unexpected event %d\n",
619 			  isert_np, event);
620 	}
621 
622 	return -1;
623 }
624 
625 static int
626 isert_disconnected_handler(struct rdma_cm_id *cma_id,
627 			   enum rdma_cm_event_type event)
628 {
629 	struct isert_conn *isert_conn = cma_id->qp->qp_context;
630 
631 	mutex_lock(&isert_conn->mutex);
632 	switch (isert_conn->state) {
633 	case ISER_CONN_TERMINATING:
634 		break;
635 	case ISER_CONN_UP:
636 		isert_conn_terminate(isert_conn);
637 		ib_drain_qp(isert_conn->qp);
638 		isert_handle_unbound_conn(isert_conn);
639 		break;
640 	case ISER_CONN_BOUND:
641 	case ISER_CONN_FULL_FEATURE: /* FALLTHRU */
642 		iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
643 		break;
644 	default:
645 		isert_warn("conn %p terminating in state %d\n",
646 			   isert_conn, isert_conn->state);
647 	}
648 	mutex_unlock(&isert_conn->mutex);
649 
650 	return 0;
651 }
652 
653 static int
654 isert_connect_error(struct rdma_cm_id *cma_id)
655 {
656 	struct isert_conn *isert_conn = cma_id->qp->qp_context;
657 	struct isert_np *isert_np = cma_id->context;
658 
659 	ib_drain_qp(isert_conn->qp);
660 
661 	mutex_lock(&isert_np->mutex);
662 	list_del_init(&isert_conn->node);
663 	mutex_unlock(&isert_np->mutex);
664 	isert_conn->cm_id = NULL;
665 	isert_put_conn(isert_conn);
666 
667 	return -1;
668 }
669 
670 static int
671 isert_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
672 {
673 	struct isert_np *isert_np = cma_id->context;
674 	struct isert_conn *isert_conn;
675 	int ret = 0;
676 
677 	isert_info("%s (%d): status %d id %p np %p\n",
678 		   rdma_event_msg(event->event), event->event,
679 		   event->status, cma_id, cma_id->context);
680 
681 	if (isert_np->cm_id == cma_id)
682 		return isert_np_cma_handler(cma_id->context, event->event);
683 
684 	switch (event->event) {
685 	case RDMA_CM_EVENT_CONNECT_REQUEST:
686 		ret = isert_connect_request(cma_id, event);
687 		if (ret)
688 			isert_err("failed handle connect request %d\n", ret);
689 		break;
690 	case RDMA_CM_EVENT_ESTABLISHED:
691 		isert_connected_handler(cma_id);
692 		break;
693 	case RDMA_CM_EVENT_ADDR_CHANGE:
694 	case RDMA_CM_EVENT_DISCONNECTED:
695 	case RDMA_CM_EVENT_TIMEWAIT_EXIT:  /* FALLTHRU */
696 		ret = isert_disconnected_handler(cma_id, event->event);
697 		break;
698 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
699 		isert_conn = cma_id->qp->qp_context;
700 		isert_conn->dev_removed = true;
701 		isert_disconnected_handler(cma_id, event->event);
702 		wait_event_interruptible(isert_conn->rem_wait,
703 					 isert_conn->state == ISER_CONN_DOWN);
704 		kfree(isert_conn);
705 		/*
706 		 * return non-zero from the callback to destroy
707 		 * the rdma cm id
708 		 */
709 		return 1;
710 	case RDMA_CM_EVENT_REJECTED:
711 		isert_info("Connection rejected: %s\n",
712 			   rdma_reject_msg(cma_id, event->status));
713 		fallthrough;
714 	case RDMA_CM_EVENT_UNREACHABLE:
715 	case RDMA_CM_EVENT_CONNECT_ERROR:
716 		ret = isert_connect_error(cma_id);
717 		break;
718 	default:
719 		isert_err("Unhandled RDMA CMA event: %d\n", event->event);
720 		break;
721 	}
722 
723 	return ret;
724 }
725 
726 static int
727 isert_post_recvm(struct isert_conn *isert_conn, u32 count)
728 {
729 	struct ib_recv_wr *rx_wr;
730 	int i, ret;
731 	struct iser_rx_desc *rx_desc;
732 
733 	for (rx_wr = isert_conn->rx_wr, i = 0; i < count; i++, rx_wr++) {
734 		rx_desc = &isert_conn->rx_descs[i];
735 
736 		rx_wr->wr_cqe = &rx_desc->rx_cqe;
737 		rx_wr->sg_list = &rx_desc->rx_sg;
738 		rx_wr->num_sge = 1;
739 		rx_wr->next = rx_wr + 1;
740 		rx_desc->in_use = false;
741 	}
742 	rx_wr--;
743 	rx_wr->next = NULL; /* mark end of work requests list */
744 
745 	ret = ib_post_recv(isert_conn->qp, isert_conn->rx_wr, NULL);
746 	if (ret)
747 		isert_err("ib_post_recv() failed with ret: %d\n", ret);
748 
749 	return ret;
750 }
751 
752 static int
753 isert_post_recv(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc)
754 {
755 	struct ib_recv_wr rx_wr;
756 	int ret;
757 
758 	if (!rx_desc->in_use) {
759 		/*
760 		 * if the descriptor is not in-use we already reposted it
761 		 * for recv, so just silently return
762 		 */
763 		return 0;
764 	}
765 
766 	rx_desc->in_use = false;
767 	rx_wr.wr_cqe = &rx_desc->rx_cqe;
768 	rx_wr.sg_list = &rx_desc->rx_sg;
769 	rx_wr.num_sge = 1;
770 	rx_wr.next = NULL;
771 
772 	ret = ib_post_recv(isert_conn->qp, &rx_wr, NULL);
773 	if (ret)
774 		isert_err("ib_post_recv() failed with ret: %d\n", ret);
775 
776 	return ret;
777 }
778 
779 static int
780 isert_login_post_send(struct isert_conn *isert_conn, struct iser_tx_desc *tx_desc)
781 {
782 	struct ib_device *ib_dev = isert_conn->cm_id->device;
783 	struct ib_send_wr send_wr;
784 	int ret;
785 
786 	ib_dma_sync_single_for_device(ib_dev, tx_desc->dma_addr,
787 				      ISER_HEADERS_LEN, DMA_TO_DEVICE);
788 
789 	tx_desc->tx_cqe.done = isert_login_send_done;
790 
791 	send_wr.next	= NULL;
792 	send_wr.wr_cqe	= &tx_desc->tx_cqe;
793 	send_wr.sg_list	= tx_desc->tx_sg;
794 	send_wr.num_sge	= tx_desc->num_sge;
795 	send_wr.opcode	= IB_WR_SEND;
796 	send_wr.send_flags = IB_SEND_SIGNALED;
797 
798 	ret = ib_post_send(isert_conn->qp, &send_wr, NULL);
799 	if (ret)
800 		isert_err("ib_post_send() failed, ret: %d\n", ret);
801 
802 	return ret;
803 }
804 
805 static void
806 __isert_create_send_desc(struct isert_device *device,
807 			 struct iser_tx_desc *tx_desc)
808 {
809 
810 	memset(&tx_desc->iser_header, 0, sizeof(struct iser_ctrl));
811 	tx_desc->iser_header.flags = ISCSI_CTRL;
812 
813 	tx_desc->num_sge = 1;
814 
815 	if (tx_desc->tx_sg[0].lkey != device->pd->local_dma_lkey) {
816 		tx_desc->tx_sg[0].lkey = device->pd->local_dma_lkey;
817 		isert_dbg("tx_desc %p lkey mismatch, fixing\n", tx_desc);
818 	}
819 }
820 
821 static void
822 isert_create_send_desc(struct isert_conn *isert_conn,
823 		       struct isert_cmd *isert_cmd,
824 		       struct iser_tx_desc *tx_desc)
825 {
826 	struct isert_device *device = isert_conn->device;
827 	struct ib_device *ib_dev = device->ib_device;
828 
829 	ib_dma_sync_single_for_cpu(ib_dev, tx_desc->dma_addr,
830 				   ISER_HEADERS_LEN, DMA_TO_DEVICE);
831 
832 	__isert_create_send_desc(device, tx_desc);
833 }
834 
835 static int
836 isert_init_tx_hdrs(struct isert_conn *isert_conn,
837 		   struct iser_tx_desc *tx_desc)
838 {
839 	struct isert_device *device = isert_conn->device;
840 	struct ib_device *ib_dev = device->ib_device;
841 	u64 dma_addr;
842 
843 	dma_addr = ib_dma_map_single(ib_dev, (void *)tx_desc,
844 			ISER_HEADERS_LEN, DMA_TO_DEVICE);
845 	if (ib_dma_mapping_error(ib_dev, dma_addr)) {
846 		isert_err("ib_dma_mapping_error() failed\n");
847 		return -ENOMEM;
848 	}
849 
850 	tx_desc->dma_addr = dma_addr;
851 	tx_desc->tx_sg[0].addr	= tx_desc->dma_addr;
852 	tx_desc->tx_sg[0].length = ISER_HEADERS_LEN;
853 	tx_desc->tx_sg[0].lkey = device->pd->local_dma_lkey;
854 
855 	isert_dbg("Setup tx_sg[0].addr: 0x%llx length: %u lkey: 0x%x\n",
856 		  tx_desc->tx_sg[0].addr, tx_desc->tx_sg[0].length,
857 		  tx_desc->tx_sg[0].lkey);
858 
859 	return 0;
860 }
861 
862 static void
863 isert_init_send_wr(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
864 		   struct ib_send_wr *send_wr)
865 {
866 	struct iser_tx_desc *tx_desc = &isert_cmd->tx_desc;
867 
868 	tx_desc->tx_cqe.done = isert_send_done;
869 	send_wr->wr_cqe = &tx_desc->tx_cqe;
870 
871 	if (isert_conn->snd_w_inv && isert_cmd->inv_rkey) {
872 		send_wr->opcode  = IB_WR_SEND_WITH_INV;
873 		send_wr->ex.invalidate_rkey = isert_cmd->inv_rkey;
874 	} else {
875 		send_wr->opcode = IB_WR_SEND;
876 	}
877 
878 	send_wr->sg_list = &tx_desc->tx_sg[0];
879 	send_wr->num_sge = isert_cmd->tx_desc.num_sge;
880 	send_wr->send_flags = IB_SEND_SIGNALED;
881 }
882 
883 static int
884 isert_login_post_recv(struct isert_conn *isert_conn)
885 {
886 	struct ib_recv_wr rx_wr;
887 	struct ib_sge sge;
888 	int ret;
889 
890 	memset(&sge, 0, sizeof(struct ib_sge));
891 	sge.addr = isert_conn->login_desc->dma_addr +
892 		isert_get_hdr_offset(isert_conn->login_desc);
893 	sge.length = ISER_RX_PAYLOAD_SIZE;
894 	sge.lkey = isert_conn->device->pd->local_dma_lkey;
895 
896 	isert_dbg("Setup sge: addr: %llx length: %d 0x%08x\n",
897 		sge.addr, sge.length, sge.lkey);
898 
899 	isert_conn->login_desc->rx_cqe.done = isert_login_recv_done;
900 
901 	memset(&rx_wr, 0, sizeof(struct ib_recv_wr));
902 	rx_wr.wr_cqe = &isert_conn->login_desc->rx_cqe;
903 	rx_wr.sg_list = &sge;
904 	rx_wr.num_sge = 1;
905 
906 	ret = ib_post_recv(isert_conn->qp, &rx_wr, NULL);
907 	if (ret)
908 		isert_err("ib_post_recv() failed: %d\n", ret);
909 
910 	return ret;
911 }
912 
913 static int
914 isert_put_login_tx(struct iscsit_conn *conn, struct iscsi_login *login,
915 		   u32 length)
916 {
917 	struct isert_conn *isert_conn = conn->context;
918 	struct isert_device *device = isert_conn->device;
919 	struct ib_device *ib_dev = device->ib_device;
920 	struct iser_tx_desc *tx_desc = &isert_conn->login_tx_desc;
921 	int ret;
922 
923 	__isert_create_send_desc(device, tx_desc);
924 
925 	memcpy(&tx_desc->iscsi_header, &login->rsp[0],
926 	       sizeof(struct iscsi_hdr));
927 
928 	isert_init_tx_hdrs(isert_conn, tx_desc);
929 
930 	if (length > 0) {
931 		struct ib_sge *tx_dsg = &tx_desc->tx_sg[1];
932 
933 		ib_dma_sync_single_for_cpu(ib_dev, isert_conn->login_rsp_dma,
934 					   length, DMA_TO_DEVICE);
935 
936 		memcpy(isert_conn->login_rsp_buf, login->rsp_buf, length);
937 
938 		ib_dma_sync_single_for_device(ib_dev, isert_conn->login_rsp_dma,
939 					      length, DMA_TO_DEVICE);
940 
941 		tx_dsg->addr	= isert_conn->login_rsp_dma;
942 		tx_dsg->length	= length;
943 		tx_dsg->lkey	= isert_conn->device->pd->local_dma_lkey;
944 		tx_desc->num_sge = 2;
945 	}
946 	if (!login->login_failed) {
947 		if (login->login_complete) {
948 			ret = isert_alloc_rx_descriptors(isert_conn);
949 			if (ret)
950 				return ret;
951 
952 			ret = isert_post_recvm(isert_conn,
953 					       ISERT_QP_MAX_RECV_DTOS);
954 			if (ret)
955 				return ret;
956 
957 			/* Now we are in FULL_FEATURE phase */
958 			mutex_lock(&isert_conn->mutex);
959 			isert_conn->state = ISER_CONN_FULL_FEATURE;
960 			mutex_unlock(&isert_conn->mutex);
961 			goto post_send;
962 		}
963 
964 		ret = isert_login_post_recv(isert_conn);
965 		if (ret)
966 			return ret;
967 	}
968 post_send:
969 	ret = isert_login_post_send(isert_conn, tx_desc);
970 	if (ret)
971 		return ret;
972 
973 	return 0;
974 }
975 
976 static void
977 isert_rx_login_req(struct isert_conn *isert_conn)
978 {
979 	struct iser_rx_desc *rx_desc = isert_conn->login_desc;
980 	int rx_buflen = isert_conn->login_req_len;
981 	struct iscsit_conn *conn = isert_conn->conn;
982 	struct iscsi_login *login = conn->conn_login;
983 	int size;
984 
985 	isert_info("conn %p\n", isert_conn);
986 
987 	WARN_ON_ONCE(!login);
988 
989 	if (login->first_request) {
990 		struct iscsi_login_req *login_req =
991 			(struct iscsi_login_req *)isert_get_iscsi_hdr(rx_desc);
992 		/*
993 		 * Setup the initial iscsi_login values from the leading
994 		 * login request PDU.
995 		 */
996 		login->leading_connection = (!login_req->tsih) ? 1 : 0;
997 		login->current_stage = ISCSI_LOGIN_CURRENT_STAGE(
998 				login_req->flags);
999 		login->version_min	= login_req->min_version;
1000 		login->version_max	= login_req->max_version;
1001 		memcpy(login->isid, login_req->isid, 6);
1002 		login->cmd_sn		= be32_to_cpu(login_req->cmdsn);
1003 		login->init_task_tag	= login_req->itt;
1004 		login->initial_exp_statsn = be32_to_cpu(login_req->exp_statsn);
1005 		login->cid		= be16_to_cpu(login_req->cid);
1006 		login->tsih		= be16_to_cpu(login_req->tsih);
1007 	}
1008 
1009 	memcpy(&login->req[0], isert_get_iscsi_hdr(rx_desc), ISCSI_HDR_LEN);
1010 
1011 	size = min(rx_buflen, MAX_KEY_VALUE_PAIRS);
1012 	isert_dbg("Using login payload size: %d, rx_buflen: %d "
1013 		  "MAX_KEY_VALUE_PAIRS: %d\n", size, rx_buflen,
1014 		  MAX_KEY_VALUE_PAIRS);
1015 	memcpy(login->req_buf, isert_get_data(rx_desc), size);
1016 
1017 	if (login->first_request) {
1018 		complete(&isert_conn->login_comp);
1019 		return;
1020 	}
1021 	queue_delayed_work(isert_login_wq, &conn->login_work, 0);
1022 }
1023 
1024 static struct iscsit_cmd
1025 *isert_allocate_cmd(struct iscsit_conn *conn, struct iser_rx_desc *rx_desc)
1026 {
1027 	struct isert_conn *isert_conn = conn->context;
1028 	struct isert_cmd *isert_cmd;
1029 	struct iscsit_cmd *cmd;
1030 
1031 	cmd = iscsit_allocate_cmd(conn, TASK_INTERRUPTIBLE);
1032 	if (!cmd) {
1033 		isert_err("Unable to allocate iscsit_cmd + isert_cmd\n");
1034 		return NULL;
1035 	}
1036 	isert_cmd = iscsit_priv_cmd(cmd);
1037 	isert_cmd->conn = isert_conn;
1038 	isert_cmd->iscsit_cmd = cmd;
1039 	isert_cmd->rx_desc = rx_desc;
1040 
1041 	return cmd;
1042 }
1043 
1044 static int
1045 isert_handle_scsi_cmd(struct isert_conn *isert_conn,
1046 		      struct isert_cmd *isert_cmd, struct iscsit_cmd *cmd,
1047 		      struct iser_rx_desc *rx_desc, unsigned char *buf)
1048 {
1049 	struct iscsit_conn *conn = isert_conn->conn;
1050 	struct iscsi_scsi_req *hdr = (struct iscsi_scsi_req *)buf;
1051 	int imm_data, imm_data_len, unsol_data, sg_nents, rc;
1052 	bool dump_payload = false;
1053 	unsigned int data_len;
1054 
1055 	rc = iscsit_setup_scsi_cmd(conn, cmd, buf);
1056 	if (rc < 0)
1057 		return rc;
1058 
1059 	imm_data = cmd->immediate_data;
1060 	imm_data_len = cmd->first_burst_len;
1061 	unsol_data = cmd->unsolicited_data;
1062 	data_len = cmd->se_cmd.data_length;
1063 
1064 	if (imm_data && imm_data_len == data_len)
1065 		cmd->se_cmd.se_cmd_flags |= SCF_PASSTHROUGH_SG_TO_MEM_NOALLOC;
1066 	rc = iscsit_process_scsi_cmd(conn, cmd, hdr);
1067 	if (rc < 0) {
1068 		return 0;
1069 	} else if (rc > 0) {
1070 		dump_payload = true;
1071 		goto sequence_cmd;
1072 	}
1073 
1074 	if (!imm_data)
1075 		return 0;
1076 
1077 	if (imm_data_len != data_len) {
1078 		sg_nents = max(1UL, DIV_ROUND_UP(imm_data_len, PAGE_SIZE));
1079 		sg_copy_from_buffer(cmd->se_cmd.t_data_sg, sg_nents,
1080 				    isert_get_data(rx_desc), imm_data_len);
1081 		isert_dbg("Copy Immediate sg_nents: %u imm_data_len: %d\n",
1082 			  sg_nents, imm_data_len);
1083 	} else {
1084 		sg_init_table(&isert_cmd->sg, 1);
1085 		cmd->se_cmd.t_data_sg = &isert_cmd->sg;
1086 		cmd->se_cmd.t_data_nents = 1;
1087 		sg_set_buf(&isert_cmd->sg, isert_get_data(rx_desc),
1088 				imm_data_len);
1089 		isert_dbg("Transfer Immediate imm_data_len: %d\n",
1090 			  imm_data_len);
1091 	}
1092 
1093 	cmd->write_data_done += imm_data_len;
1094 
1095 	if (cmd->write_data_done == cmd->se_cmd.data_length) {
1096 		spin_lock_bh(&cmd->istate_lock);
1097 		cmd->cmd_flags |= ICF_GOT_LAST_DATAOUT;
1098 		cmd->i_state = ISTATE_RECEIVED_LAST_DATAOUT;
1099 		spin_unlock_bh(&cmd->istate_lock);
1100 	}
1101 
1102 sequence_cmd:
1103 	rc = iscsit_sequence_cmd(conn, cmd, buf, hdr->cmdsn);
1104 
1105 	if (!rc && !dump_payload && unsol_data)
1106 		iscsit_set_unsolicited_dataout(cmd);
1107 	else if (dump_payload && imm_data)
1108 		target_put_sess_cmd(&cmd->se_cmd);
1109 
1110 	return 0;
1111 }
1112 
1113 static int
1114 isert_handle_iscsi_dataout(struct isert_conn *isert_conn,
1115 			   struct iser_rx_desc *rx_desc, unsigned char *buf)
1116 {
1117 	struct scatterlist *sg_start;
1118 	struct iscsit_conn *conn = isert_conn->conn;
1119 	struct iscsit_cmd *cmd = NULL;
1120 	struct iscsi_data *hdr = (struct iscsi_data *)buf;
1121 	u32 unsol_data_len = ntoh24(hdr->dlength);
1122 	int rc, sg_nents, sg_off, page_off;
1123 
1124 	rc = iscsit_check_dataout_hdr(conn, buf, &cmd);
1125 	if (rc < 0)
1126 		return rc;
1127 	else if (!cmd)
1128 		return 0;
1129 	/*
1130 	 * FIXME: Unexpected unsolicited_data out
1131 	 */
1132 	if (!cmd->unsolicited_data) {
1133 		isert_err("Received unexpected solicited data payload\n");
1134 		dump_stack();
1135 		return -1;
1136 	}
1137 
1138 	isert_dbg("Unsolicited DataOut unsol_data_len: %u, "
1139 		  "write_data_done: %u, data_length: %u\n",
1140 		  unsol_data_len,  cmd->write_data_done,
1141 		  cmd->se_cmd.data_length);
1142 
1143 	sg_off = cmd->write_data_done / PAGE_SIZE;
1144 	sg_start = &cmd->se_cmd.t_data_sg[sg_off];
1145 	sg_nents = max(1UL, DIV_ROUND_UP(unsol_data_len, PAGE_SIZE));
1146 	page_off = cmd->write_data_done % PAGE_SIZE;
1147 	/*
1148 	 * FIXME: Non page-aligned unsolicited_data out
1149 	 */
1150 	if (page_off) {
1151 		isert_err("unexpected non-page aligned data payload\n");
1152 		dump_stack();
1153 		return -1;
1154 	}
1155 	isert_dbg("Copying DataOut: sg_start: %p, sg_off: %u "
1156 		  "sg_nents: %u from %p %u\n", sg_start, sg_off,
1157 		  sg_nents, isert_get_data(rx_desc), unsol_data_len);
1158 
1159 	sg_copy_from_buffer(sg_start, sg_nents, isert_get_data(rx_desc),
1160 			    unsol_data_len);
1161 
1162 	rc = iscsit_check_dataout_payload(cmd, hdr, false);
1163 	if (rc < 0)
1164 		return rc;
1165 
1166 	/*
1167 	 * multiple data-outs on the same command can arrive -
1168 	 * so post the buffer before hand
1169 	 */
1170 	return isert_post_recv(isert_conn, rx_desc);
1171 }
1172 
1173 static int
1174 isert_handle_nop_out(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
1175 		     struct iscsit_cmd *cmd, struct iser_rx_desc *rx_desc,
1176 		     unsigned char *buf)
1177 {
1178 	struct iscsit_conn *conn = isert_conn->conn;
1179 	struct iscsi_nopout *hdr = (struct iscsi_nopout *)buf;
1180 	int rc;
1181 
1182 	rc = iscsit_setup_nop_out(conn, cmd, hdr);
1183 	if (rc < 0)
1184 		return rc;
1185 	/*
1186 	 * FIXME: Add support for NOPOUT payload using unsolicited RDMA payload
1187 	 */
1188 
1189 	return iscsit_process_nop_out(conn, cmd, hdr);
1190 }
1191 
1192 static int
1193 isert_handle_text_cmd(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
1194 		      struct iscsit_cmd *cmd, struct iser_rx_desc *rx_desc,
1195 		      struct iscsi_text *hdr)
1196 {
1197 	struct iscsit_conn *conn = isert_conn->conn;
1198 	u32 payload_length = ntoh24(hdr->dlength);
1199 	int rc;
1200 	unsigned char *text_in = NULL;
1201 
1202 	rc = iscsit_setup_text_cmd(conn, cmd, hdr);
1203 	if (rc < 0)
1204 		return rc;
1205 
1206 	if (payload_length) {
1207 		text_in = kzalloc(payload_length, GFP_KERNEL);
1208 		if (!text_in)
1209 			return -ENOMEM;
1210 	}
1211 	cmd->text_in_ptr = text_in;
1212 
1213 	memcpy(cmd->text_in_ptr, isert_get_data(rx_desc), payload_length);
1214 
1215 	return iscsit_process_text_cmd(conn, cmd, hdr);
1216 }
1217 
1218 static int
1219 isert_rx_opcode(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc,
1220 		uint32_t read_stag, uint64_t read_va,
1221 		uint32_t write_stag, uint64_t write_va)
1222 {
1223 	struct iscsi_hdr *hdr = isert_get_iscsi_hdr(rx_desc);
1224 	struct iscsit_conn *conn = isert_conn->conn;
1225 	struct iscsit_cmd *cmd;
1226 	struct isert_cmd *isert_cmd;
1227 	int ret = -EINVAL;
1228 	u8 opcode = (hdr->opcode & ISCSI_OPCODE_MASK);
1229 
1230 	if (conn->sess->sess_ops->SessionType &&
1231 	   (!(opcode & ISCSI_OP_TEXT) || !(opcode & ISCSI_OP_LOGOUT))) {
1232 		isert_err("Got illegal opcode: 0x%02x in SessionType=Discovery,"
1233 			  " ignoring\n", opcode);
1234 		return 0;
1235 	}
1236 
1237 	switch (opcode) {
1238 	case ISCSI_OP_SCSI_CMD:
1239 		cmd = isert_allocate_cmd(conn, rx_desc);
1240 		if (!cmd)
1241 			break;
1242 
1243 		isert_cmd = iscsit_priv_cmd(cmd);
1244 		isert_cmd->read_stag = read_stag;
1245 		isert_cmd->read_va = read_va;
1246 		isert_cmd->write_stag = write_stag;
1247 		isert_cmd->write_va = write_va;
1248 		isert_cmd->inv_rkey = read_stag ? read_stag : write_stag;
1249 
1250 		ret = isert_handle_scsi_cmd(isert_conn, isert_cmd, cmd,
1251 					rx_desc, (unsigned char *)hdr);
1252 		break;
1253 	case ISCSI_OP_NOOP_OUT:
1254 		cmd = isert_allocate_cmd(conn, rx_desc);
1255 		if (!cmd)
1256 			break;
1257 
1258 		isert_cmd = iscsit_priv_cmd(cmd);
1259 		ret = isert_handle_nop_out(isert_conn, isert_cmd, cmd,
1260 					   rx_desc, (unsigned char *)hdr);
1261 		break;
1262 	case ISCSI_OP_SCSI_DATA_OUT:
1263 		ret = isert_handle_iscsi_dataout(isert_conn, rx_desc,
1264 						(unsigned char *)hdr);
1265 		break;
1266 	case ISCSI_OP_SCSI_TMFUNC:
1267 		cmd = isert_allocate_cmd(conn, rx_desc);
1268 		if (!cmd)
1269 			break;
1270 
1271 		ret = iscsit_handle_task_mgt_cmd(conn, cmd,
1272 						(unsigned char *)hdr);
1273 		break;
1274 	case ISCSI_OP_LOGOUT:
1275 		cmd = isert_allocate_cmd(conn, rx_desc);
1276 		if (!cmd)
1277 			break;
1278 
1279 		ret = iscsit_handle_logout_cmd(conn, cmd, (unsigned char *)hdr);
1280 		break;
1281 	case ISCSI_OP_TEXT:
1282 		if (be32_to_cpu(hdr->ttt) != 0xFFFFFFFF)
1283 			cmd = iscsit_find_cmd_from_itt(conn, hdr->itt);
1284 		else
1285 			cmd = isert_allocate_cmd(conn, rx_desc);
1286 
1287 		if (!cmd)
1288 			break;
1289 
1290 		isert_cmd = iscsit_priv_cmd(cmd);
1291 		ret = isert_handle_text_cmd(isert_conn, isert_cmd, cmd,
1292 					    rx_desc, (struct iscsi_text *)hdr);
1293 		break;
1294 	default:
1295 		isert_err("Got unknown iSCSI OpCode: 0x%02x\n", opcode);
1296 		dump_stack();
1297 		break;
1298 	}
1299 
1300 	return ret;
1301 }
1302 
1303 static void
1304 isert_print_wc(struct ib_wc *wc, const char *type)
1305 {
1306 	if (wc->status != IB_WC_WR_FLUSH_ERR)
1307 		isert_err("%s failure: %s (%d) vend_err %x\n", type,
1308 			  ib_wc_status_msg(wc->status), wc->status,
1309 			  wc->vendor_err);
1310 	else
1311 		isert_dbg("%s failure: %s (%d)\n", type,
1312 			  ib_wc_status_msg(wc->status), wc->status);
1313 }
1314 
1315 static void
1316 isert_recv_done(struct ib_cq *cq, struct ib_wc *wc)
1317 {
1318 	struct isert_conn *isert_conn = wc->qp->qp_context;
1319 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1320 	struct iser_rx_desc *rx_desc = cqe_to_rx_desc(wc->wr_cqe);
1321 	struct iscsi_hdr *hdr = isert_get_iscsi_hdr(rx_desc);
1322 	struct iser_ctrl *iser_ctrl = isert_get_iser_hdr(rx_desc);
1323 	uint64_t read_va = 0, write_va = 0;
1324 	uint32_t read_stag = 0, write_stag = 0;
1325 
1326 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1327 		isert_print_wc(wc, "recv");
1328 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1329 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1330 		return;
1331 	}
1332 
1333 	rx_desc->in_use = true;
1334 
1335 	ib_dma_sync_single_for_cpu(ib_dev, rx_desc->dma_addr,
1336 			ISER_RX_SIZE, DMA_FROM_DEVICE);
1337 
1338 	isert_dbg("DMA: 0x%llx, iSCSI opcode: 0x%02x, ITT: 0x%08x, flags: 0x%02x dlen: %d\n",
1339 		 rx_desc->dma_addr, hdr->opcode, hdr->itt, hdr->flags,
1340 		 (int)(wc->byte_len - ISER_HEADERS_LEN));
1341 
1342 	switch (iser_ctrl->flags & 0xF0) {
1343 	case ISCSI_CTRL:
1344 		if (iser_ctrl->flags & ISER_RSV) {
1345 			read_stag = be32_to_cpu(iser_ctrl->read_stag);
1346 			read_va = be64_to_cpu(iser_ctrl->read_va);
1347 			isert_dbg("ISER_RSV: read_stag: 0x%x read_va: 0x%llx\n",
1348 				  read_stag, (unsigned long long)read_va);
1349 		}
1350 		if (iser_ctrl->flags & ISER_WSV) {
1351 			write_stag = be32_to_cpu(iser_ctrl->write_stag);
1352 			write_va = be64_to_cpu(iser_ctrl->write_va);
1353 			isert_dbg("ISER_WSV: write_stag: 0x%x write_va: 0x%llx\n",
1354 				  write_stag, (unsigned long long)write_va);
1355 		}
1356 
1357 		isert_dbg("ISER ISCSI_CTRL PDU\n");
1358 		break;
1359 	case ISER_HELLO:
1360 		isert_err("iSER Hello message\n");
1361 		break;
1362 	default:
1363 		isert_warn("Unknown iSER hdr flags: 0x%02x\n", iser_ctrl->flags);
1364 		break;
1365 	}
1366 
1367 	isert_rx_opcode(isert_conn, rx_desc,
1368 			read_stag, read_va, write_stag, write_va);
1369 
1370 	ib_dma_sync_single_for_device(ib_dev, rx_desc->dma_addr,
1371 			ISER_RX_SIZE, DMA_FROM_DEVICE);
1372 }
1373 
1374 static void
1375 isert_login_recv_done(struct ib_cq *cq, struct ib_wc *wc)
1376 {
1377 	struct isert_conn *isert_conn = wc->qp->qp_context;
1378 	struct ib_device *ib_dev = isert_conn->device->ib_device;
1379 
1380 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1381 		isert_print_wc(wc, "login recv");
1382 		return;
1383 	}
1384 
1385 	ib_dma_sync_single_for_cpu(ib_dev, isert_conn->login_desc->dma_addr,
1386 			ISER_RX_SIZE, DMA_FROM_DEVICE);
1387 
1388 	isert_conn->login_req_len = wc->byte_len - ISER_HEADERS_LEN;
1389 
1390 	if (isert_conn->conn) {
1391 		struct iscsi_login *login = isert_conn->conn->conn_login;
1392 
1393 		if (login && !login->first_request)
1394 			isert_rx_login_req(isert_conn);
1395 	}
1396 
1397 	mutex_lock(&isert_conn->mutex);
1398 	complete(&isert_conn->login_req_comp);
1399 	mutex_unlock(&isert_conn->mutex);
1400 
1401 	ib_dma_sync_single_for_device(ib_dev, isert_conn->login_desc->dma_addr,
1402 				ISER_RX_SIZE, DMA_FROM_DEVICE);
1403 }
1404 
1405 static void
1406 isert_rdma_rw_ctx_destroy(struct isert_cmd *cmd, struct isert_conn *conn)
1407 {
1408 	struct se_cmd *se_cmd = &cmd->iscsit_cmd->se_cmd;
1409 	enum dma_data_direction dir = target_reverse_dma_direction(se_cmd);
1410 
1411 	if (!cmd->rw.nr_ops)
1412 		return;
1413 
1414 	if (isert_prot_cmd(conn, se_cmd)) {
1415 		rdma_rw_ctx_destroy_signature(&cmd->rw, conn->qp,
1416 				conn->cm_id->port_num, se_cmd->t_data_sg,
1417 				se_cmd->t_data_nents, se_cmd->t_prot_sg,
1418 				se_cmd->t_prot_nents, dir);
1419 	} else {
1420 		rdma_rw_ctx_destroy(&cmd->rw, conn->qp, conn->cm_id->port_num,
1421 				se_cmd->t_data_sg, se_cmd->t_data_nents, dir);
1422 	}
1423 
1424 	cmd->rw.nr_ops = 0;
1425 }
1426 
1427 static void
1428 isert_put_cmd(struct isert_cmd *isert_cmd, bool comp_err)
1429 {
1430 	struct iscsit_cmd *cmd = isert_cmd->iscsit_cmd;
1431 	struct isert_conn *isert_conn = isert_cmd->conn;
1432 	struct iscsit_conn *conn = isert_conn->conn;
1433 	struct iscsi_text_rsp *hdr;
1434 
1435 	isert_dbg("Cmd %p\n", isert_cmd);
1436 
1437 	switch (cmd->iscsi_opcode) {
1438 	case ISCSI_OP_SCSI_CMD:
1439 		spin_lock_bh(&conn->cmd_lock);
1440 		if (!list_empty(&cmd->i_conn_node))
1441 			list_del_init(&cmd->i_conn_node);
1442 		spin_unlock_bh(&conn->cmd_lock);
1443 
1444 		if (cmd->data_direction == DMA_TO_DEVICE) {
1445 			iscsit_stop_dataout_timer(cmd);
1446 			/*
1447 			 * Check for special case during comp_err where
1448 			 * WRITE_PENDING has been handed off from core,
1449 			 * but requires an extra target_put_sess_cmd()
1450 			 * before transport_generic_free_cmd() below.
1451 			 */
1452 			if (comp_err &&
1453 			    cmd->se_cmd.t_state == TRANSPORT_WRITE_PENDING) {
1454 				struct se_cmd *se_cmd = &cmd->se_cmd;
1455 
1456 				target_put_sess_cmd(se_cmd);
1457 			}
1458 		}
1459 
1460 		isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1461 		transport_generic_free_cmd(&cmd->se_cmd, 0);
1462 		break;
1463 	case ISCSI_OP_SCSI_TMFUNC:
1464 		spin_lock_bh(&conn->cmd_lock);
1465 		if (!list_empty(&cmd->i_conn_node))
1466 			list_del_init(&cmd->i_conn_node);
1467 		spin_unlock_bh(&conn->cmd_lock);
1468 
1469 		transport_generic_free_cmd(&cmd->se_cmd, 0);
1470 		break;
1471 	case ISCSI_OP_REJECT:
1472 	case ISCSI_OP_NOOP_OUT:
1473 	case ISCSI_OP_TEXT:
1474 		hdr = (struct iscsi_text_rsp *)&isert_cmd->tx_desc.iscsi_header;
1475 		/* If the continue bit is on, keep the command alive */
1476 		if (hdr->flags & ISCSI_FLAG_TEXT_CONTINUE)
1477 			break;
1478 
1479 		spin_lock_bh(&conn->cmd_lock);
1480 		if (!list_empty(&cmd->i_conn_node))
1481 			list_del_init(&cmd->i_conn_node);
1482 		spin_unlock_bh(&conn->cmd_lock);
1483 
1484 		/*
1485 		 * Handle special case for REJECT when iscsi_add_reject*() has
1486 		 * overwritten the original iscsi_opcode assignment, and the
1487 		 * associated cmd->se_cmd needs to be released.
1488 		 */
1489 		if (cmd->se_cmd.se_tfo != NULL) {
1490 			isert_dbg("Calling transport_generic_free_cmd for 0x%02x\n",
1491 				 cmd->iscsi_opcode);
1492 			transport_generic_free_cmd(&cmd->se_cmd, 0);
1493 			break;
1494 		}
1495 		fallthrough;
1496 	default:
1497 		iscsit_release_cmd(cmd);
1498 		break;
1499 	}
1500 }
1501 
1502 static void
1503 isert_unmap_tx_desc(struct iser_tx_desc *tx_desc, struct ib_device *ib_dev)
1504 {
1505 	if (tx_desc->dma_addr != 0) {
1506 		isert_dbg("unmap single for tx_desc->dma_addr\n");
1507 		ib_dma_unmap_single(ib_dev, tx_desc->dma_addr,
1508 				    ISER_HEADERS_LEN, DMA_TO_DEVICE);
1509 		tx_desc->dma_addr = 0;
1510 	}
1511 }
1512 
1513 static void
1514 isert_completion_put(struct iser_tx_desc *tx_desc, struct isert_cmd *isert_cmd,
1515 		     struct ib_device *ib_dev, bool comp_err)
1516 {
1517 	if (isert_cmd->pdu_buf_dma != 0) {
1518 		isert_dbg("unmap single for isert_cmd->pdu_buf_dma\n");
1519 		ib_dma_unmap_single(ib_dev, isert_cmd->pdu_buf_dma,
1520 				    isert_cmd->pdu_buf_len, DMA_TO_DEVICE);
1521 		isert_cmd->pdu_buf_dma = 0;
1522 	}
1523 
1524 	isert_unmap_tx_desc(tx_desc, ib_dev);
1525 	isert_put_cmd(isert_cmd, comp_err);
1526 }
1527 
1528 static int
1529 isert_check_pi_status(struct se_cmd *se_cmd, struct ib_mr *sig_mr)
1530 {
1531 	struct ib_mr_status mr_status;
1532 	int ret;
1533 
1534 	ret = ib_check_mr_status(sig_mr, IB_MR_CHECK_SIG_STATUS, &mr_status);
1535 	if (ret) {
1536 		isert_err("ib_check_mr_status failed, ret %d\n", ret);
1537 		goto fail_mr_status;
1538 	}
1539 
1540 	if (mr_status.fail_status & IB_MR_CHECK_SIG_STATUS) {
1541 		u64 sec_offset_err;
1542 		u32 block_size = se_cmd->se_dev->dev_attrib.block_size + 8;
1543 
1544 		switch (mr_status.sig_err.err_type) {
1545 		case IB_SIG_BAD_GUARD:
1546 			se_cmd->pi_err = TCM_LOGICAL_BLOCK_GUARD_CHECK_FAILED;
1547 			break;
1548 		case IB_SIG_BAD_REFTAG:
1549 			se_cmd->pi_err = TCM_LOGICAL_BLOCK_REF_TAG_CHECK_FAILED;
1550 			break;
1551 		case IB_SIG_BAD_APPTAG:
1552 			se_cmd->pi_err = TCM_LOGICAL_BLOCK_APP_TAG_CHECK_FAILED;
1553 			break;
1554 		}
1555 		sec_offset_err = mr_status.sig_err.sig_err_offset;
1556 		do_div(sec_offset_err, block_size);
1557 		se_cmd->sense_info = sec_offset_err + se_cmd->t_task_lba;
1558 
1559 		isert_err("PI error found type %d at sector 0x%llx "
1560 			  "expected 0x%x vs actual 0x%x\n",
1561 			  mr_status.sig_err.err_type,
1562 			  (unsigned long long)se_cmd->sense_info,
1563 			  mr_status.sig_err.expected,
1564 			  mr_status.sig_err.actual);
1565 		ret = 1;
1566 	}
1567 
1568 fail_mr_status:
1569 	return ret;
1570 }
1571 
1572 static void
1573 isert_rdma_write_done(struct ib_cq *cq, struct ib_wc *wc)
1574 {
1575 	struct isert_conn *isert_conn = wc->qp->qp_context;
1576 	struct isert_device *device = isert_conn->device;
1577 	struct iser_tx_desc *desc = cqe_to_tx_desc(wc->wr_cqe);
1578 	struct isert_cmd *isert_cmd = tx_desc_to_cmd(desc);
1579 	struct se_cmd *cmd = &isert_cmd->iscsit_cmd->se_cmd;
1580 	int ret = 0;
1581 
1582 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1583 		isert_print_wc(wc, "rdma write");
1584 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1585 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1586 		isert_completion_put(desc, isert_cmd, device->ib_device, true);
1587 		return;
1588 	}
1589 
1590 	isert_dbg("Cmd %p\n", isert_cmd);
1591 
1592 	ret = isert_check_pi_status(cmd, isert_cmd->rw.reg->mr);
1593 	isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1594 
1595 	if (ret) {
1596 		/*
1597 		 * transport_generic_request_failure() expects to have
1598 		 * plus two references to handle queue-full, so re-add
1599 		 * one here as target-core will have already dropped
1600 		 * it after the first isert_put_datain() callback.
1601 		 */
1602 		kref_get(&cmd->cmd_kref);
1603 		transport_generic_request_failure(cmd, cmd->pi_err);
1604 	} else {
1605 		/*
1606 		 * XXX: isert_put_response() failure is not retried.
1607 		 */
1608 		ret = isert_put_response(isert_conn->conn, isert_cmd->iscsit_cmd);
1609 		if (ret)
1610 			pr_warn_ratelimited("isert_put_response() ret: %d\n", ret);
1611 	}
1612 }
1613 
1614 static void
1615 isert_rdma_read_done(struct ib_cq *cq, struct ib_wc *wc)
1616 {
1617 	struct isert_conn *isert_conn = wc->qp->qp_context;
1618 	struct isert_device *device = isert_conn->device;
1619 	struct iser_tx_desc *desc = cqe_to_tx_desc(wc->wr_cqe);
1620 	struct isert_cmd *isert_cmd = tx_desc_to_cmd(desc);
1621 	struct iscsit_cmd *cmd = isert_cmd->iscsit_cmd;
1622 	struct se_cmd *se_cmd = &cmd->se_cmd;
1623 	int ret = 0;
1624 
1625 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1626 		isert_print_wc(wc, "rdma read");
1627 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1628 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1629 		isert_completion_put(desc, isert_cmd, device->ib_device, true);
1630 		return;
1631 	}
1632 
1633 	isert_dbg("Cmd %p\n", isert_cmd);
1634 
1635 	iscsit_stop_dataout_timer(cmd);
1636 
1637 	if (isert_prot_cmd(isert_conn, se_cmd))
1638 		ret = isert_check_pi_status(se_cmd, isert_cmd->rw.reg->mr);
1639 	isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1640 	cmd->write_data_done = 0;
1641 
1642 	isert_dbg("Cmd: %p RDMA_READ comp calling execute_cmd\n", isert_cmd);
1643 	spin_lock_bh(&cmd->istate_lock);
1644 	cmd->cmd_flags |= ICF_GOT_LAST_DATAOUT;
1645 	cmd->i_state = ISTATE_RECEIVED_LAST_DATAOUT;
1646 	spin_unlock_bh(&cmd->istate_lock);
1647 
1648 	/*
1649 	 * transport_generic_request_failure() will drop the extra
1650 	 * se_cmd->cmd_kref reference after T10-PI error, and handle
1651 	 * any non-zero ->queue_status() callback error retries.
1652 	 */
1653 	if (ret)
1654 		transport_generic_request_failure(se_cmd, se_cmd->pi_err);
1655 	else
1656 		target_execute_cmd(se_cmd);
1657 }
1658 
1659 static void
1660 isert_do_control_comp(struct work_struct *work)
1661 {
1662 	struct isert_cmd *isert_cmd = container_of(work,
1663 			struct isert_cmd, comp_work);
1664 	struct isert_conn *isert_conn = isert_cmd->conn;
1665 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1666 	struct iscsit_cmd *cmd = isert_cmd->iscsit_cmd;
1667 
1668 	isert_dbg("Cmd %p i_state %d\n", isert_cmd, cmd->i_state);
1669 
1670 	switch (cmd->i_state) {
1671 	case ISTATE_SEND_TASKMGTRSP:
1672 		iscsit_tmr_post_handler(cmd, cmd->conn);
1673 		fallthrough;
1674 	case ISTATE_SEND_REJECT:
1675 	case ISTATE_SEND_TEXTRSP:
1676 		cmd->i_state = ISTATE_SENT_STATUS;
1677 		isert_completion_put(&isert_cmd->tx_desc, isert_cmd,
1678 				     ib_dev, false);
1679 		break;
1680 	case ISTATE_SEND_LOGOUTRSP:
1681 		iscsit_logout_post_handler(cmd, cmd->conn);
1682 		break;
1683 	default:
1684 		isert_err("Unknown i_state %d\n", cmd->i_state);
1685 		dump_stack();
1686 		break;
1687 	}
1688 }
1689 
1690 static void
1691 isert_login_send_done(struct ib_cq *cq, struct ib_wc *wc)
1692 {
1693 	struct isert_conn *isert_conn = wc->qp->qp_context;
1694 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1695 	struct iser_tx_desc *tx_desc = cqe_to_tx_desc(wc->wr_cqe);
1696 
1697 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1698 		isert_print_wc(wc, "login send");
1699 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1700 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1701 	}
1702 
1703 	isert_unmap_tx_desc(tx_desc, ib_dev);
1704 }
1705 
1706 static void
1707 isert_send_done(struct ib_cq *cq, struct ib_wc *wc)
1708 {
1709 	struct isert_conn *isert_conn = wc->qp->qp_context;
1710 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1711 	struct iser_tx_desc *tx_desc = cqe_to_tx_desc(wc->wr_cqe);
1712 	struct isert_cmd *isert_cmd = tx_desc_to_cmd(tx_desc);
1713 
1714 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1715 		isert_print_wc(wc, "send");
1716 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1717 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1718 		isert_completion_put(tx_desc, isert_cmd, ib_dev, true);
1719 		return;
1720 	}
1721 
1722 	isert_dbg("Cmd %p\n", isert_cmd);
1723 
1724 	switch (isert_cmd->iscsit_cmd->i_state) {
1725 	case ISTATE_SEND_TASKMGTRSP:
1726 	case ISTATE_SEND_LOGOUTRSP:
1727 	case ISTATE_SEND_REJECT:
1728 	case ISTATE_SEND_TEXTRSP:
1729 		isert_unmap_tx_desc(tx_desc, ib_dev);
1730 
1731 		INIT_WORK(&isert_cmd->comp_work, isert_do_control_comp);
1732 		queue_work(isert_comp_wq, &isert_cmd->comp_work);
1733 		return;
1734 	default:
1735 		isert_cmd->iscsit_cmd->i_state = ISTATE_SENT_STATUS;
1736 		isert_completion_put(tx_desc, isert_cmd, ib_dev, false);
1737 		break;
1738 	}
1739 }
1740 
1741 static int
1742 isert_post_response(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd)
1743 {
1744 	int ret;
1745 
1746 	ret = isert_post_recv(isert_conn, isert_cmd->rx_desc);
1747 	if (ret)
1748 		return ret;
1749 
1750 	ret = ib_post_send(isert_conn->qp, &isert_cmd->tx_desc.send_wr, NULL);
1751 	if (ret) {
1752 		isert_err("ib_post_send failed with %d\n", ret);
1753 		return ret;
1754 	}
1755 	return ret;
1756 }
1757 
1758 static int
1759 isert_put_response(struct iscsit_conn *conn, struct iscsit_cmd *cmd)
1760 {
1761 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1762 	struct isert_conn *isert_conn = conn->context;
1763 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1764 	struct iscsi_scsi_rsp *hdr = (struct iscsi_scsi_rsp *)
1765 				&isert_cmd->tx_desc.iscsi_header;
1766 
1767 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1768 	iscsit_build_rsp_pdu(cmd, conn, true, hdr);
1769 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1770 	/*
1771 	 * Attach SENSE DATA payload to iSCSI Response PDU
1772 	 */
1773 	if (cmd->se_cmd.sense_buffer &&
1774 	    ((cmd->se_cmd.se_cmd_flags & SCF_TRANSPORT_TASK_SENSE) ||
1775 	    (cmd->se_cmd.se_cmd_flags & SCF_EMULATED_TASK_SENSE))) {
1776 		struct isert_device *device = isert_conn->device;
1777 		struct ib_device *ib_dev = device->ib_device;
1778 		struct ib_sge *tx_dsg = &isert_cmd->tx_desc.tx_sg[1];
1779 		u32 padding, pdu_len;
1780 
1781 		put_unaligned_be16(cmd->se_cmd.scsi_sense_length,
1782 				   cmd->sense_buffer);
1783 		cmd->se_cmd.scsi_sense_length += sizeof(__be16);
1784 
1785 		padding = -(cmd->se_cmd.scsi_sense_length) & 3;
1786 		hton24(hdr->dlength, (u32)cmd->se_cmd.scsi_sense_length);
1787 		pdu_len = cmd->se_cmd.scsi_sense_length + padding;
1788 
1789 		isert_cmd->pdu_buf_dma = ib_dma_map_single(ib_dev,
1790 				(void *)cmd->sense_buffer, pdu_len,
1791 				DMA_TO_DEVICE);
1792 		if (ib_dma_mapping_error(ib_dev, isert_cmd->pdu_buf_dma))
1793 			return -ENOMEM;
1794 
1795 		isert_cmd->pdu_buf_len = pdu_len;
1796 		tx_dsg->addr	= isert_cmd->pdu_buf_dma;
1797 		tx_dsg->length	= pdu_len;
1798 		tx_dsg->lkey	= device->pd->local_dma_lkey;
1799 		isert_cmd->tx_desc.num_sge = 2;
1800 	}
1801 
1802 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1803 
1804 	isert_dbg("Posting SCSI Response\n");
1805 
1806 	return isert_post_response(isert_conn, isert_cmd);
1807 }
1808 
1809 static void
1810 isert_aborted_task(struct iscsit_conn *conn, struct iscsit_cmd *cmd)
1811 {
1812 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1813 	struct isert_conn *isert_conn = conn->context;
1814 
1815 	spin_lock_bh(&conn->cmd_lock);
1816 	if (!list_empty(&cmd->i_conn_node))
1817 		list_del_init(&cmd->i_conn_node);
1818 	spin_unlock_bh(&conn->cmd_lock);
1819 
1820 	if (cmd->data_direction == DMA_TO_DEVICE)
1821 		iscsit_stop_dataout_timer(cmd);
1822 	isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1823 }
1824 
1825 static enum target_prot_op
1826 isert_get_sup_prot_ops(struct iscsit_conn *conn)
1827 {
1828 	struct isert_conn *isert_conn = conn->context;
1829 	struct isert_device *device = isert_conn->device;
1830 
1831 	if (conn->tpg->tpg_attrib.t10_pi) {
1832 		if (device->pi_capable) {
1833 			isert_info("conn %p PI offload enabled\n", isert_conn);
1834 			isert_conn->pi_support = true;
1835 			return TARGET_PROT_ALL;
1836 		}
1837 	}
1838 
1839 	isert_info("conn %p PI offload disabled\n", isert_conn);
1840 	isert_conn->pi_support = false;
1841 
1842 	return TARGET_PROT_NORMAL;
1843 }
1844 
1845 static int
1846 isert_put_nopin(struct iscsit_cmd *cmd, struct iscsit_conn *conn,
1847 		bool nopout_response)
1848 {
1849 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1850 	struct isert_conn *isert_conn = conn->context;
1851 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1852 
1853 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1854 	iscsit_build_nopin_rsp(cmd, conn, (struct iscsi_nopin *)
1855 			       &isert_cmd->tx_desc.iscsi_header,
1856 			       nopout_response);
1857 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1858 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1859 
1860 	isert_dbg("conn %p Posting NOPIN Response\n", isert_conn);
1861 
1862 	return isert_post_response(isert_conn, isert_cmd);
1863 }
1864 
1865 static int
1866 isert_put_logout_rsp(struct iscsit_cmd *cmd, struct iscsit_conn *conn)
1867 {
1868 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1869 	struct isert_conn *isert_conn = conn->context;
1870 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1871 
1872 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1873 	iscsit_build_logout_rsp(cmd, conn, (struct iscsi_logout_rsp *)
1874 				&isert_cmd->tx_desc.iscsi_header);
1875 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1876 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1877 
1878 	isert_dbg("conn %p Posting Logout Response\n", isert_conn);
1879 
1880 	return isert_post_response(isert_conn, isert_cmd);
1881 }
1882 
1883 static int
1884 isert_put_tm_rsp(struct iscsit_cmd *cmd, struct iscsit_conn *conn)
1885 {
1886 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1887 	struct isert_conn *isert_conn = conn->context;
1888 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1889 
1890 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1891 	iscsit_build_task_mgt_rsp(cmd, conn, (struct iscsi_tm_rsp *)
1892 				  &isert_cmd->tx_desc.iscsi_header);
1893 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1894 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1895 
1896 	isert_dbg("conn %p Posting Task Management Response\n", isert_conn);
1897 
1898 	return isert_post_response(isert_conn, isert_cmd);
1899 }
1900 
1901 static int
1902 isert_put_reject(struct iscsit_cmd *cmd, struct iscsit_conn *conn)
1903 {
1904 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1905 	struct isert_conn *isert_conn = conn->context;
1906 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1907 	struct isert_device *device = isert_conn->device;
1908 	struct ib_device *ib_dev = device->ib_device;
1909 	struct ib_sge *tx_dsg = &isert_cmd->tx_desc.tx_sg[1];
1910 	struct iscsi_reject *hdr =
1911 		(struct iscsi_reject *)&isert_cmd->tx_desc.iscsi_header;
1912 
1913 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1914 	iscsit_build_reject(cmd, conn, hdr);
1915 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1916 
1917 	hton24(hdr->dlength, ISCSI_HDR_LEN);
1918 	isert_cmd->pdu_buf_dma = ib_dma_map_single(ib_dev,
1919 			(void *)cmd->buf_ptr, ISCSI_HDR_LEN,
1920 			DMA_TO_DEVICE);
1921 	if (ib_dma_mapping_error(ib_dev, isert_cmd->pdu_buf_dma))
1922 		return -ENOMEM;
1923 	isert_cmd->pdu_buf_len = ISCSI_HDR_LEN;
1924 	tx_dsg->addr	= isert_cmd->pdu_buf_dma;
1925 	tx_dsg->length	= ISCSI_HDR_LEN;
1926 	tx_dsg->lkey	= device->pd->local_dma_lkey;
1927 	isert_cmd->tx_desc.num_sge = 2;
1928 
1929 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1930 
1931 	isert_dbg("conn %p Posting Reject\n", isert_conn);
1932 
1933 	return isert_post_response(isert_conn, isert_cmd);
1934 }
1935 
1936 static int
1937 isert_put_text_rsp(struct iscsit_cmd *cmd, struct iscsit_conn *conn)
1938 {
1939 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1940 	struct isert_conn *isert_conn = conn->context;
1941 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1942 	struct iscsi_text_rsp *hdr =
1943 		(struct iscsi_text_rsp *)&isert_cmd->tx_desc.iscsi_header;
1944 	u32 txt_rsp_len;
1945 	int rc;
1946 
1947 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1948 	rc = iscsit_build_text_rsp(cmd, conn, hdr, ISCSI_INFINIBAND);
1949 	if (rc < 0)
1950 		return rc;
1951 
1952 	txt_rsp_len = rc;
1953 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1954 
1955 	if (txt_rsp_len) {
1956 		struct isert_device *device = isert_conn->device;
1957 		struct ib_device *ib_dev = device->ib_device;
1958 		struct ib_sge *tx_dsg = &isert_cmd->tx_desc.tx_sg[1];
1959 		void *txt_rsp_buf = cmd->buf_ptr;
1960 
1961 		isert_cmd->pdu_buf_dma = ib_dma_map_single(ib_dev,
1962 				txt_rsp_buf, txt_rsp_len, DMA_TO_DEVICE);
1963 		if (ib_dma_mapping_error(ib_dev, isert_cmd->pdu_buf_dma))
1964 			return -ENOMEM;
1965 
1966 		isert_cmd->pdu_buf_len = txt_rsp_len;
1967 		tx_dsg->addr	= isert_cmd->pdu_buf_dma;
1968 		tx_dsg->length	= txt_rsp_len;
1969 		tx_dsg->lkey	= device->pd->local_dma_lkey;
1970 		isert_cmd->tx_desc.num_sge = 2;
1971 	}
1972 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1973 
1974 	isert_dbg("conn %p Text Response\n", isert_conn);
1975 
1976 	return isert_post_response(isert_conn, isert_cmd);
1977 }
1978 
1979 static inline void
1980 isert_set_dif_domain(struct se_cmd *se_cmd, struct ib_sig_domain *domain)
1981 {
1982 	domain->sig_type = IB_SIG_TYPE_T10_DIF;
1983 	domain->sig.dif.bg_type = IB_T10DIF_CRC;
1984 	domain->sig.dif.pi_interval = se_cmd->se_dev->dev_attrib.block_size;
1985 	domain->sig.dif.ref_tag = se_cmd->reftag_seed;
1986 	/*
1987 	 * At the moment we hard code those, but if in the future
1988 	 * the target core would like to use it, we will take it
1989 	 * from se_cmd.
1990 	 */
1991 	domain->sig.dif.apptag_check_mask = 0xffff;
1992 	domain->sig.dif.app_escape = true;
1993 	domain->sig.dif.ref_escape = true;
1994 	if (se_cmd->prot_type == TARGET_DIF_TYPE1_PROT ||
1995 	    se_cmd->prot_type == TARGET_DIF_TYPE2_PROT)
1996 		domain->sig.dif.ref_remap = true;
1997 }
1998 
1999 static int
2000 isert_set_sig_attrs(struct se_cmd *se_cmd, struct ib_sig_attrs *sig_attrs)
2001 {
2002 	memset(sig_attrs, 0, sizeof(*sig_attrs));
2003 
2004 	switch (se_cmd->prot_op) {
2005 	case TARGET_PROT_DIN_INSERT:
2006 	case TARGET_PROT_DOUT_STRIP:
2007 		sig_attrs->mem.sig_type = IB_SIG_TYPE_NONE;
2008 		isert_set_dif_domain(se_cmd, &sig_attrs->wire);
2009 		break;
2010 	case TARGET_PROT_DOUT_INSERT:
2011 	case TARGET_PROT_DIN_STRIP:
2012 		sig_attrs->wire.sig_type = IB_SIG_TYPE_NONE;
2013 		isert_set_dif_domain(se_cmd, &sig_attrs->mem);
2014 		break;
2015 	case TARGET_PROT_DIN_PASS:
2016 	case TARGET_PROT_DOUT_PASS:
2017 		isert_set_dif_domain(se_cmd, &sig_attrs->wire);
2018 		isert_set_dif_domain(se_cmd, &sig_attrs->mem);
2019 		break;
2020 	default:
2021 		isert_err("Unsupported PI operation %d\n", se_cmd->prot_op);
2022 		return -EINVAL;
2023 	}
2024 
2025 	if (se_cmd->prot_checks & TARGET_DIF_CHECK_GUARD)
2026 		sig_attrs->check_mask |= IB_SIG_CHECK_GUARD;
2027 	if (se_cmd->prot_checks & TARGET_DIF_CHECK_APPTAG)
2028 		sig_attrs->check_mask |= IB_SIG_CHECK_APPTAG;
2029 	if (se_cmd->prot_checks & TARGET_DIF_CHECK_REFTAG)
2030 		sig_attrs->check_mask |= IB_SIG_CHECK_REFTAG;
2031 
2032 	return 0;
2033 }
2034 
2035 static int
2036 isert_rdma_rw_ctx_post(struct isert_cmd *cmd, struct isert_conn *conn,
2037 		struct ib_cqe *cqe, struct ib_send_wr *chain_wr)
2038 {
2039 	struct se_cmd *se_cmd = &cmd->iscsit_cmd->se_cmd;
2040 	enum dma_data_direction dir = target_reverse_dma_direction(se_cmd);
2041 	u8 port_num = conn->cm_id->port_num;
2042 	u64 addr;
2043 	u32 rkey, offset;
2044 	int ret;
2045 
2046 	if (cmd->ctx_init_done)
2047 		goto rdma_ctx_post;
2048 
2049 	if (dir == DMA_FROM_DEVICE) {
2050 		addr = cmd->write_va;
2051 		rkey = cmd->write_stag;
2052 		offset = cmd->iscsit_cmd->write_data_done;
2053 	} else {
2054 		addr = cmd->read_va;
2055 		rkey = cmd->read_stag;
2056 		offset = 0;
2057 	}
2058 
2059 	if (isert_prot_cmd(conn, se_cmd)) {
2060 		struct ib_sig_attrs sig_attrs;
2061 
2062 		ret = isert_set_sig_attrs(se_cmd, &sig_attrs);
2063 		if (ret)
2064 			return ret;
2065 
2066 		WARN_ON_ONCE(offset);
2067 		ret = rdma_rw_ctx_signature_init(&cmd->rw, conn->qp, port_num,
2068 				se_cmd->t_data_sg, se_cmd->t_data_nents,
2069 				se_cmd->t_prot_sg, se_cmd->t_prot_nents,
2070 				&sig_attrs, addr, rkey, dir);
2071 	} else {
2072 		ret = rdma_rw_ctx_init(&cmd->rw, conn->qp, port_num,
2073 				se_cmd->t_data_sg, se_cmd->t_data_nents,
2074 				offset, addr, rkey, dir);
2075 	}
2076 
2077 	if (ret < 0) {
2078 		isert_err("Cmd: %p failed to prepare RDMA res\n", cmd);
2079 		return ret;
2080 	}
2081 
2082 	cmd->ctx_init_done = true;
2083 
2084 rdma_ctx_post:
2085 	ret = rdma_rw_ctx_post(&cmd->rw, conn->qp, port_num, cqe, chain_wr);
2086 	if (ret < 0)
2087 		isert_err("Cmd: %p failed to post RDMA res\n", cmd);
2088 	return ret;
2089 }
2090 
2091 static int
2092 isert_put_datain(struct iscsit_conn *conn, struct iscsit_cmd *cmd)
2093 {
2094 	struct se_cmd *se_cmd = &cmd->se_cmd;
2095 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2096 	struct isert_conn *isert_conn = conn->context;
2097 	struct ib_cqe *cqe = NULL;
2098 	struct ib_send_wr *chain_wr = NULL;
2099 	int rc;
2100 
2101 	isert_dbg("Cmd: %p RDMA_WRITE data_length: %u\n",
2102 		 isert_cmd, se_cmd->data_length);
2103 
2104 	if (isert_prot_cmd(isert_conn, se_cmd)) {
2105 		isert_cmd->tx_desc.tx_cqe.done = isert_rdma_write_done;
2106 		cqe = &isert_cmd->tx_desc.tx_cqe;
2107 	} else {
2108 		/*
2109 		 * Build isert_conn->tx_desc for iSCSI response PDU and attach
2110 		 */
2111 		isert_create_send_desc(isert_conn, isert_cmd,
2112 				       &isert_cmd->tx_desc);
2113 		iscsit_build_rsp_pdu(cmd, conn, true, (struct iscsi_scsi_rsp *)
2114 				     &isert_cmd->tx_desc.iscsi_header);
2115 		isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
2116 		isert_init_send_wr(isert_conn, isert_cmd,
2117 				   &isert_cmd->tx_desc.send_wr);
2118 
2119 		rc = isert_post_recv(isert_conn, isert_cmd->rx_desc);
2120 		if (rc)
2121 			return rc;
2122 
2123 		chain_wr = &isert_cmd->tx_desc.send_wr;
2124 	}
2125 
2126 	rc = isert_rdma_rw_ctx_post(isert_cmd, isert_conn, cqe, chain_wr);
2127 	isert_dbg("Cmd: %p posted RDMA_WRITE for iSER Data READ rc: %d\n",
2128 		  isert_cmd, rc);
2129 	return rc;
2130 }
2131 
2132 static int
2133 isert_get_dataout(struct iscsit_conn *conn, struct iscsit_cmd *cmd, bool recovery)
2134 {
2135 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2136 	int ret;
2137 
2138 	isert_dbg("Cmd: %p RDMA_READ data_length: %u write_data_done: %u\n",
2139 		 isert_cmd, cmd->se_cmd.data_length, cmd->write_data_done);
2140 
2141 	isert_cmd->tx_desc.tx_cqe.done = isert_rdma_read_done;
2142 	ret = isert_rdma_rw_ctx_post(isert_cmd, conn->context,
2143 				     &isert_cmd->tx_desc.tx_cqe, NULL);
2144 
2145 	isert_dbg("Cmd: %p posted RDMA_READ memory for ISER Data WRITE rc: %d\n",
2146 		 isert_cmd, ret);
2147 	return ret;
2148 }
2149 
2150 static int
2151 isert_immediate_queue(struct iscsit_conn *conn, struct iscsit_cmd *cmd, int state)
2152 {
2153 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2154 	int ret = 0;
2155 
2156 	switch (state) {
2157 	case ISTATE_REMOVE:
2158 		spin_lock_bh(&conn->cmd_lock);
2159 		list_del_init(&cmd->i_conn_node);
2160 		spin_unlock_bh(&conn->cmd_lock);
2161 		isert_put_cmd(isert_cmd, true);
2162 		break;
2163 	case ISTATE_SEND_NOPIN_WANT_RESPONSE:
2164 		ret = isert_put_nopin(cmd, conn, false);
2165 		break;
2166 	default:
2167 		isert_err("Unknown immediate state: 0x%02x\n", state);
2168 		ret = -EINVAL;
2169 		break;
2170 	}
2171 
2172 	return ret;
2173 }
2174 
2175 static int
2176 isert_response_queue(struct iscsit_conn *conn, struct iscsit_cmd *cmd, int state)
2177 {
2178 	struct isert_conn *isert_conn = conn->context;
2179 	int ret;
2180 
2181 	switch (state) {
2182 	case ISTATE_SEND_LOGOUTRSP:
2183 		ret = isert_put_logout_rsp(cmd, conn);
2184 		if (!ret)
2185 			isert_conn->logout_posted = true;
2186 		break;
2187 	case ISTATE_SEND_NOPIN:
2188 		ret = isert_put_nopin(cmd, conn, true);
2189 		break;
2190 	case ISTATE_SEND_TASKMGTRSP:
2191 		ret = isert_put_tm_rsp(cmd, conn);
2192 		break;
2193 	case ISTATE_SEND_REJECT:
2194 		ret = isert_put_reject(cmd, conn);
2195 		break;
2196 	case ISTATE_SEND_TEXTRSP:
2197 		ret = isert_put_text_rsp(cmd, conn);
2198 		break;
2199 	case ISTATE_SEND_STATUS:
2200 		/*
2201 		 * Special case for sending non GOOD SCSI status from TX thread
2202 		 * context during pre se_cmd excecution failure.
2203 		 */
2204 		ret = isert_put_response(conn, cmd);
2205 		break;
2206 	default:
2207 		isert_err("Unknown response state: 0x%02x\n", state);
2208 		ret = -EINVAL;
2209 		break;
2210 	}
2211 
2212 	return ret;
2213 }
2214 
2215 struct rdma_cm_id *
2216 isert_setup_id(struct isert_np *isert_np)
2217 {
2218 	struct iscsi_np *np = isert_np->np;
2219 	struct rdma_cm_id *id;
2220 	struct sockaddr *sa;
2221 	int ret;
2222 
2223 	sa = (struct sockaddr *)&np->np_sockaddr;
2224 	isert_dbg("ksockaddr: %p, sa: %p\n", &np->np_sockaddr, sa);
2225 
2226 	id = rdma_create_id(&init_net, isert_cma_handler, isert_np,
2227 			    RDMA_PS_TCP, IB_QPT_RC);
2228 	if (IS_ERR(id)) {
2229 		isert_err("rdma_create_id() failed: %ld\n", PTR_ERR(id));
2230 		ret = PTR_ERR(id);
2231 		goto out;
2232 	}
2233 	isert_dbg("id %p context %p\n", id, id->context);
2234 
2235 	/*
2236 	 * Allow both IPv4 and IPv6 sockets to bind a single port
2237 	 * at the same time.
2238 	 */
2239 	ret = rdma_set_afonly(id, 1);
2240 	if (ret) {
2241 		isert_err("rdma_set_afonly() failed: %d\n", ret);
2242 		goto out_id;
2243 	}
2244 
2245 	ret = rdma_bind_addr(id, sa);
2246 	if (ret) {
2247 		isert_err("rdma_bind_addr() failed: %d\n", ret);
2248 		goto out_id;
2249 	}
2250 
2251 	ret = rdma_listen(id, 0);
2252 	if (ret) {
2253 		isert_err("rdma_listen() failed: %d\n", ret);
2254 		goto out_id;
2255 	}
2256 
2257 	return id;
2258 out_id:
2259 	rdma_destroy_id(id);
2260 out:
2261 	return ERR_PTR(ret);
2262 }
2263 
2264 static int
2265 isert_setup_np(struct iscsi_np *np,
2266 	       struct sockaddr_storage *ksockaddr)
2267 {
2268 	struct isert_np *isert_np;
2269 	struct rdma_cm_id *isert_lid;
2270 	int ret;
2271 
2272 	isert_np = kzalloc(sizeof(struct isert_np), GFP_KERNEL);
2273 	if (!isert_np)
2274 		return -ENOMEM;
2275 
2276 	sema_init(&isert_np->sem, 0);
2277 	mutex_init(&isert_np->mutex);
2278 	INIT_LIST_HEAD(&isert_np->accepted);
2279 	INIT_LIST_HEAD(&isert_np->pending);
2280 	isert_np->np = np;
2281 
2282 	/*
2283 	 * Setup the np->np_sockaddr from the passed sockaddr setup
2284 	 * in iscsi_target_configfs.c code..
2285 	 */
2286 	memcpy(&np->np_sockaddr, ksockaddr,
2287 	       sizeof(struct sockaddr_storage));
2288 
2289 	isert_lid = isert_setup_id(isert_np);
2290 	if (IS_ERR(isert_lid)) {
2291 		ret = PTR_ERR(isert_lid);
2292 		goto out;
2293 	}
2294 
2295 	isert_np->cm_id = isert_lid;
2296 	np->np_context = isert_np;
2297 
2298 	return 0;
2299 
2300 out:
2301 	kfree(isert_np);
2302 
2303 	return ret;
2304 }
2305 
2306 static int
2307 isert_rdma_accept(struct isert_conn *isert_conn)
2308 {
2309 	struct rdma_cm_id *cm_id = isert_conn->cm_id;
2310 	struct rdma_conn_param cp;
2311 	int ret;
2312 	struct iser_cm_hdr rsp_hdr;
2313 
2314 	memset(&cp, 0, sizeof(struct rdma_conn_param));
2315 	cp.initiator_depth = isert_conn->initiator_depth;
2316 	cp.retry_count = 7;
2317 	cp.rnr_retry_count = 7;
2318 
2319 	memset(&rsp_hdr, 0, sizeof(rsp_hdr));
2320 	rsp_hdr.flags = ISERT_ZBVA_NOT_USED;
2321 	if (!isert_conn->snd_w_inv)
2322 		rsp_hdr.flags = rsp_hdr.flags | ISERT_SEND_W_INV_NOT_USED;
2323 	cp.private_data = (void *)&rsp_hdr;
2324 	cp.private_data_len = sizeof(rsp_hdr);
2325 
2326 	ret = rdma_accept(cm_id, &cp);
2327 	if (ret) {
2328 		isert_err("rdma_accept() failed with: %d\n", ret);
2329 		return ret;
2330 	}
2331 
2332 	return 0;
2333 }
2334 
2335 static int
2336 isert_get_login_rx(struct iscsit_conn *conn, struct iscsi_login *login)
2337 {
2338 	struct isert_conn *isert_conn = conn->context;
2339 	int ret;
2340 
2341 	isert_info("before login_req comp conn: %p\n", isert_conn);
2342 	ret = wait_for_completion_interruptible(&isert_conn->login_req_comp);
2343 	if (ret) {
2344 		isert_err("isert_conn %p interrupted before got login req\n",
2345 			  isert_conn);
2346 		return ret;
2347 	}
2348 	reinit_completion(&isert_conn->login_req_comp);
2349 
2350 	/*
2351 	 * For login requests after the first PDU, isert_rx_login_req() will
2352 	 * kick queue_delayed_work(isert_login_wq, &conn->login_work) as
2353 	 * the packet is received, which turns this callback from
2354 	 * iscsi_target_do_login_rx() into a NOP.
2355 	 */
2356 	if (!login->first_request)
2357 		return 0;
2358 
2359 	isert_rx_login_req(isert_conn);
2360 
2361 	isert_info("before login_comp conn: %p\n", conn);
2362 	ret = wait_for_completion_interruptible(&isert_conn->login_comp);
2363 	if (ret)
2364 		return ret;
2365 
2366 	isert_info("processing login->req: %p\n", login->req);
2367 
2368 	return 0;
2369 }
2370 
2371 static void
2372 isert_set_conn_info(struct iscsi_np *np, struct iscsit_conn *conn,
2373 		    struct isert_conn *isert_conn)
2374 {
2375 	struct rdma_cm_id *cm_id = isert_conn->cm_id;
2376 	struct rdma_route *cm_route = &cm_id->route;
2377 
2378 	conn->login_family = np->np_sockaddr.ss_family;
2379 
2380 	conn->login_sockaddr = cm_route->addr.dst_addr;
2381 	conn->local_sockaddr = cm_route->addr.src_addr;
2382 }
2383 
2384 static int
2385 isert_accept_np(struct iscsi_np *np, struct iscsit_conn *conn)
2386 {
2387 	struct isert_np *isert_np = np->np_context;
2388 	struct isert_conn *isert_conn;
2389 	int ret;
2390 
2391 accept_wait:
2392 	ret = down_interruptible(&isert_np->sem);
2393 	if (ret)
2394 		return -ENODEV;
2395 
2396 	spin_lock_bh(&np->np_thread_lock);
2397 	if (np->np_thread_state >= ISCSI_NP_THREAD_RESET) {
2398 		spin_unlock_bh(&np->np_thread_lock);
2399 		isert_dbg("np_thread_state %d\n",
2400 			 np->np_thread_state);
2401 		/*
2402 		 * No point in stalling here when np_thread
2403 		 * is in state RESET/SHUTDOWN/EXIT - bail
2404 		 */
2405 		return -ENODEV;
2406 	}
2407 	spin_unlock_bh(&np->np_thread_lock);
2408 
2409 	mutex_lock(&isert_np->mutex);
2410 	if (list_empty(&isert_np->pending)) {
2411 		mutex_unlock(&isert_np->mutex);
2412 		goto accept_wait;
2413 	}
2414 	isert_conn = list_first_entry(&isert_np->pending,
2415 			struct isert_conn, node);
2416 	list_del_init(&isert_conn->node);
2417 	mutex_unlock(&isert_np->mutex);
2418 
2419 	conn->context = isert_conn;
2420 	isert_conn->conn = conn;
2421 	isert_conn->state = ISER_CONN_BOUND;
2422 
2423 	isert_set_conn_info(np, conn, isert_conn);
2424 
2425 	isert_dbg("Processing isert_conn: %p\n", isert_conn);
2426 
2427 	return 0;
2428 }
2429 
2430 static void
2431 isert_free_np(struct iscsi_np *np)
2432 {
2433 	struct isert_np *isert_np = np->np_context;
2434 	struct isert_conn *isert_conn, *n;
2435 	LIST_HEAD(drop_conn_list);
2436 
2437 	if (isert_np->cm_id)
2438 		rdma_destroy_id(isert_np->cm_id);
2439 
2440 	/*
2441 	 * FIXME: At this point we don't have a good way to insure
2442 	 * that at this point we don't have hanging connections that
2443 	 * completed RDMA establishment but didn't start iscsi login
2444 	 * process. So work-around this by cleaning up what ever piled
2445 	 * up in accepted and pending lists.
2446 	 */
2447 	mutex_lock(&isert_np->mutex);
2448 	if (!list_empty(&isert_np->pending)) {
2449 		isert_info("Still have isert pending connections\n");
2450 		list_for_each_entry_safe(isert_conn, n,
2451 					 &isert_np->pending,
2452 					 node) {
2453 			isert_info("cleaning isert_conn %p state (%d)\n",
2454 				   isert_conn, isert_conn->state);
2455 			list_move_tail(&isert_conn->node, &drop_conn_list);
2456 		}
2457 	}
2458 
2459 	if (!list_empty(&isert_np->accepted)) {
2460 		isert_info("Still have isert accepted connections\n");
2461 		list_for_each_entry_safe(isert_conn, n,
2462 					 &isert_np->accepted,
2463 					 node) {
2464 			isert_info("cleaning isert_conn %p state (%d)\n",
2465 				   isert_conn, isert_conn->state);
2466 			list_move_tail(&isert_conn->node, &drop_conn_list);
2467 		}
2468 	}
2469 	mutex_unlock(&isert_np->mutex);
2470 
2471 	list_for_each_entry_safe(isert_conn, n, &drop_conn_list, node) {
2472 		list_del_init(&isert_conn->node);
2473 		isert_connect_release(isert_conn);
2474 	}
2475 
2476 	np->np_context = NULL;
2477 	kfree(isert_np);
2478 }
2479 
2480 static void isert_release_work(struct work_struct *work)
2481 {
2482 	struct isert_conn *isert_conn = container_of(work,
2483 						     struct isert_conn,
2484 						     release_work);
2485 
2486 	isert_info("Starting release conn %p\n", isert_conn);
2487 
2488 	mutex_lock(&isert_conn->mutex);
2489 	isert_conn->state = ISER_CONN_DOWN;
2490 	mutex_unlock(&isert_conn->mutex);
2491 
2492 	isert_info("Destroying conn %p\n", isert_conn);
2493 	isert_put_conn(isert_conn);
2494 }
2495 
2496 static void
2497 isert_wait4logout(struct isert_conn *isert_conn)
2498 {
2499 	struct iscsit_conn *conn = isert_conn->conn;
2500 
2501 	isert_info("conn %p\n", isert_conn);
2502 
2503 	if (isert_conn->logout_posted) {
2504 		isert_info("conn %p wait for conn_logout_comp\n", isert_conn);
2505 		wait_for_completion_timeout(&conn->conn_logout_comp,
2506 					    SECONDS_FOR_LOGOUT_COMP * HZ);
2507 	}
2508 }
2509 
2510 static void
2511 isert_wait4cmds(struct iscsit_conn *conn)
2512 {
2513 	isert_info("iscsit_conn %p\n", conn);
2514 
2515 	if (conn->sess) {
2516 		target_stop_cmd_counter(conn->cmd_cnt);
2517 		target_wait_for_cmds(conn->cmd_cnt);
2518 	}
2519 }
2520 
2521 /**
2522  * isert_put_unsol_pending_cmds() - Drop commands waiting for
2523  *     unsolicitate dataout
2524  * @conn:    iscsi connection
2525  *
2526  * We might still have commands that are waiting for unsolicited
2527  * dataouts messages. We must put the extra reference on those
2528  * before blocking on the target_wait_for_session_cmds
2529  */
2530 static void
2531 isert_put_unsol_pending_cmds(struct iscsit_conn *conn)
2532 {
2533 	struct iscsit_cmd *cmd, *tmp;
2534 	static LIST_HEAD(drop_cmd_list);
2535 
2536 	spin_lock_bh(&conn->cmd_lock);
2537 	list_for_each_entry_safe(cmd, tmp, &conn->conn_cmd_list, i_conn_node) {
2538 		if ((cmd->cmd_flags & ICF_NON_IMMEDIATE_UNSOLICITED_DATA) &&
2539 		    (cmd->write_data_done < conn->sess->sess_ops->FirstBurstLength) &&
2540 		    (cmd->write_data_done < cmd->se_cmd.data_length))
2541 			list_move_tail(&cmd->i_conn_node, &drop_cmd_list);
2542 	}
2543 	spin_unlock_bh(&conn->cmd_lock);
2544 
2545 	list_for_each_entry_safe(cmd, tmp, &drop_cmd_list, i_conn_node) {
2546 		list_del_init(&cmd->i_conn_node);
2547 		if (cmd->i_state != ISTATE_REMOVE) {
2548 			struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2549 
2550 			isert_info("conn %p dropping cmd %p\n", conn, cmd);
2551 			isert_put_cmd(isert_cmd, true);
2552 		}
2553 	}
2554 }
2555 
2556 static void isert_wait_conn(struct iscsit_conn *conn)
2557 {
2558 	struct isert_conn *isert_conn = conn->context;
2559 
2560 	isert_info("Starting conn %p\n", isert_conn);
2561 
2562 	mutex_lock(&isert_conn->mutex);
2563 	isert_conn_terminate(isert_conn);
2564 	mutex_unlock(&isert_conn->mutex);
2565 
2566 	ib_drain_qp(isert_conn->qp);
2567 	isert_put_unsol_pending_cmds(conn);
2568 	isert_wait4cmds(conn);
2569 	isert_wait4logout(isert_conn);
2570 
2571 	queue_work(isert_release_wq, &isert_conn->release_work);
2572 }
2573 
2574 static void isert_free_conn(struct iscsit_conn *conn)
2575 {
2576 	struct isert_conn *isert_conn = conn->context;
2577 
2578 	ib_drain_qp(isert_conn->qp);
2579 	isert_put_conn(isert_conn);
2580 }
2581 
2582 static void isert_get_rx_pdu(struct iscsit_conn *conn)
2583 {
2584 	struct completion comp;
2585 
2586 	init_completion(&comp);
2587 
2588 	wait_for_completion_interruptible(&comp);
2589 }
2590 
2591 static struct iscsit_transport iser_target_transport = {
2592 	.name			= "IB/iSER",
2593 	.transport_type		= ISCSI_INFINIBAND,
2594 	.rdma_shutdown		= true,
2595 	.priv_size		= sizeof(struct isert_cmd),
2596 	.owner			= THIS_MODULE,
2597 	.iscsit_setup_np	= isert_setup_np,
2598 	.iscsit_accept_np	= isert_accept_np,
2599 	.iscsit_free_np		= isert_free_np,
2600 	.iscsit_wait_conn	= isert_wait_conn,
2601 	.iscsit_free_conn	= isert_free_conn,
2602 	.iscsit_get_login_rx	= isert_get_login_rx,
2603 	.iscsit_put_login_tx	= isert_put_login_tx,
2604 	.iscsit_immediate_queue	= isert_immediate_queue,
2605 	.iscsit_response_queue	= isert_response_queue,
2606 	.iscsit_get_dataout	= isert_get_dataout,
2607 	.iscsit_queue_data_in	= isert_put_datain,
2608 	.iscsit_queue_status	= isert_put_response,
2609 	.iscsit_aborted_task	= isert_aborted_task,
2610 	.iscsit_get_rx_pdu	= isert_get_rx_pdu,
2611 	.iscsit_get_sup_prot_ops = isert_get_sup_prot_ops,
2612 };
2613 
2614 static int __init isert_init(void)
2615 {
2616 	isert_login_wq = alloc_workqueue("isert_login_wq", 0, 0);
2617 	if (!isert_login_wq) {
2618 		isert_err("Unable to allocate isert_login_wq\n");
2619 		return -ENOMEM;
2620 	}
2621 
2622 	isert_comp_wq = alloc_workqueue("isert_comp_wq",
2623 					WQ_UNBOUND | WQ_HIGHPRI, 0);
2624 	if (!isert_comp_wq) {
2625 		isert_err("Unable to allocate isert_comp_wq\n");
2626 		goto destroy_login_wq;
2627 	}
2628 
2629 	isert_release_wq = alloc_workqueue("isert_release_wq", WQ_UNBOUND,
2630 					WQ_UNBOUND_MAX_ACTIVE);
2631 	if (!isert_release_wq) {
2632 		isert_err("Unable to allocate isert_release_wq\n");
2633 		goto destroy_comp_wq;
2634 	}
2635 
2636 	iscsit_register_transport(&iser_target_transport);
2637 	isert_info("iSER_TARGET[0] - Loaded iser_target_transport\n");
2638 
2639 	return 0;
2640 
2641 destroy_comp_wq:
2642 	destroy_workqueue(isert_comp_wq);
2643 destroy_login_wq:
2644 	destroy_workqueue(isert_login_wq);
2645 
2646 	return -ENOMEM;
2647 }
2648 
2649 static void __exit isert_exit(void)
2650 {
2651 	flush_workqueue(isert_login_wq);
2652 	destroy_workqueue(isert_release_wq);
2653 	destroy_workqueue(isert_comp_wq);
2654 	iscsit_unregister_transport(&iser_target_transport);
2655 	isert_info("iSER_TARGET[0] - Released iser_target_transport\n");
2656 	destroy_workqueue(isert_login_wq);
2657 }
2658 
2659 MODULE_DESCRIPTION("iSER-Target for mainline target infrastructure");
2660 MODULE_AUTHOR("nab@Linux-iSCSI.org");
2661 MODULE_LICENSE("GPL");
2662 
2663 module_init(isert_init);
2664 module_exit(isert_exit);
2665