xref: /linux/drivers/infiniband/ulp/isert/ib_isert.c (revision fd7d598270724cc787982ea48bbe17ad383a8b7f)
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 	case IB_EVENT_QP_LAST_WQE_REACHED:
95 		isert_warn("Reached TX IB_EVENT_QP_LAST_WQE_REACHED\n");
96 		break;
97 	default:
98 		break;
99 	}
100 }
101 
102 static struct ib_qp *
103 isert_create_qp(struct isert_conn *isert_conn,
104 		struct rdma_cm_id *cma_id)
105 {
106 	u32 cq_size = ISERT_QP_MAX_REQ_DTOS + ISERT_QP_MAX_RECV_DTOS + 2;
107 	struct isert_device *device = isert_conn->device;
108 	struct ib_device *ib_dev = device->ib_device;
109 	struct ib_qp_init_attr attr;
110 	int ret, factor;
111 
112 	isert_conn->cq = ib_cq_pool_get(ib_dev, cq_size, -1, IB_POLL_WORKQUEUE);
113 	if (IS_ERR(isert_conn->cq)) {
114 		isert_err("Unable to allocate cq\n");
115 		ret = PTR_ERR(isert_conn->cq);
116 		return ERR_PTR(ret);
117 	}
118 	isert_conn->cq_size = cq_size;
119 
120 	memset(&attr, 0, sizeof(struct ib_qp_init_attr));
121 	attr.event_handler = isert_qp_event_callback;
122 	attr.qp_context = isert_conn;
123 	attr.send_cq = isert_conn->cq;
124 	attr.recv_cq = isert_conn->cq;
125 	attr.cap.max_send_wr = ISERT_QP_MAX_REQ_DTOS + 1;
126 	attr.cap.max_recv_wr = ISERT_QP_MAX_RECV_DTOS + 1;
127 	factor = rdma_rw_mr_factor(device->ib_device, cma_id->port_num,
128 				   isert_sg_tablesize);
129 	attr.cap.max_rdma_ctxs = ISCSI_DEF_XMIT_CMDS_MAX * factor;
130 	attr.cap.max_send_sge = device->ib_device->attrs.max_send_sge;
131 	attr.cap.max_recv_sge = 1;
132 	attr.sq_sig_type = IB_SIGNAL_REQ_WR;
133 	attr.qp_type = IB_QPT_RC;
134 	if (device->pi_capable)
135 		attr.create_flags |= IB_QP_CREATE_INTEGRITY_EN;
136 
137 	ret = rdma_create_qp(cma_id, device->pd, &attr);
138 	if (ret) {
139 		isert_err("rdma_create_qp failed for cma_id %d\n", ret);
140 		ib_cq_pool_put(isert_conn->cq, isert_conn->cq_size);
141 
142 		return ERR_PTR(ret);
143 	}
144 
145 	return cma_id->qp;
146 }
147 
148 static int
149 isert_alloc_rx_descriptors(struct isert_conn *isert_conn)
150 {
151 	struct isert_device *device = isert_conn->device;
152 	struct ib_device *ib_dev = device->ib_device;
153 	struct iser_rx_desc *rx_desc;
154 	struct ib_sge *rx_sg;
155 	u64 dma_addr;
156 	int i, j;
157 
158 	isert_conn->rx_descs = kcalloc(ISERT_QP_MAX_RECV_DTOS,
159 				       sizeof(struct iser_rx_desc),
160 				       GFP_KERNEL);
161 	if (!isert_conn->rx_descs)
162 		return -ENOMEM;
163 
164 	rx_desc = isert_conn->rx_descs;
165 
166 	for (i = 0; i < ISERT_QP_MAX_RECV_DTOS; i++, rx_desc++)  {
167 		dma_addr = ib_dma_map_single(ib_dev, rx_desc->buf,
168 					ISER_RX_SIZE, DMA_FROM_DEVICE);
169 		if (ib_dma_mapping_error(ib_dev, dma_addr))
170 			goto dma_map_fail;
171 
172 		rx_desc->dma_addr = dma_addr;
173 
174 		rx_sg = &rx_desc->rx_sg;
175 		rx_sg->addr = rx_desc->dma_addr + isert_get_hdr_offset(rx_desc);
176 		rx_sg->length = ISER_RX_PAYLOAD_SIZE;
177 		rx_sg->lkey = device->pd->local_dma_lkey;
178 		rx_desc->rx_cqe.done = isert_recv_done;
179 	}
180 
181 	return 0;
182 
183 dma_map_fail:
184 	rx_desc = isert_conn->rx_descs;
185 	for (j = 0; j < i; j++, rx_desc++) {
186 		ib_dma_unmap_single(ib_dev, rx_desc->dma_addr,
187 				    ISER_RX_SIZE, DMA_FROM_DEVICE);
188 	}
189 	kfree(isert_conn->rx_descs);
190 	isert_conn->rx_descs = NULL;
191 	isert_err("conn %p failed to allocate rx descriptors\n", isert_conn);
192 	return -ENOMEM;
193 }
194 
195 static void
196 isert_free_rx_descriptors(struct isert_conn *isert_conn)
197 {
198 	struct ib_device *ib_dev = isert_conn->device->ib_device;
199 	struct iser_rx_desc *rx_desc;
200 	int i;
201 
202 	if (!isert_conn->rx_descs)
203 		return;
204 
205 	rx_desc = isert_conn->rx_descs;
206 	for (i = 0; i < ISERT_QP_MAX_RECV_DTOS; i++, rx_desc++)  {
207 		ib_dma_unmap_single(ib_dev, rx_desc->dma_addr,
208 				    ISER_RX_SIZE, DMA_FROM_DEVICE);
209 	}
210 
211 	kfree(isert_conn->rx_descs);
212 	isert_conn->rx_descs = NULL;
213 }
214 
215 static int
216 isert_create_device_ib_res(struct isert_device *device)
217 {
218 	struct ib_device *ib_dev = device->ib_device;
219 	int ret;
220 
221 	isert_dbg("devattr->max_send_sge: %d devattr->max_recv_sge %d\n",
222 		  ib_dev->attrs.max_send_sge, ib_dev->attrs.max_recv_sge);
223 	isert_dbg("devattr->max_sge_rd: %d\n", ib_dev->attrs.max_sge_rd);
224 
225 	device->pd = ib_alloc_pd(ib_dev, 0);
226 	if (IS_ERR(device->pd)) {
227 		ret = PTR_ERR(device->pd);
228 		isert_err("failed to allocate pd, device %p, ret=%d\n",
229 			  device, ret);
230 		return ret;
231 	}
232 
233 	/* Check signature cap */
234 	if (ib_dev->attrs.kernel_cap_flags & IBK_INTEGRITY_HANDOVER)
235 		device->pi_capable = true;
236 	else
237 		device->pi_capable = false;
238 
239 	return 0;
240 }
241 
242 static void
243 isert_free_device_ib_res(struct isert_device *device)
244 {
245 	isert_info("device %p\n", device);
246 
247 	ib_dealloc_pd(device->pd);
248 }
249 
250 static void
251 isert_device_put(struct isert_device *device)
252 {
253 	mutex_lock(&device_list_mutex);
254 	device->refcount--;
255 	isert_info("device %p refcount %d\n", device, device->refcount);
256 	if (!device->refcount) {
257 		isert_free_device_ib_res(device);
258 		list_del(&device->dev_node);
259 		kfree(device);
260 	}
261 	mutex_unlock(&device_list_mutex);
262 }
263 
264 static struct isert_device *
265 isert_device_get(struct rdma_cm_id *cma_id)
266 {
267 	struct isert_device *device;
268 	int ret;
269 
270 	mutex_lock(&device_list_mutex);
271 	list_for_each_entry(device, &device_list, dev_node) {
272 		if (device->ib_device->node_guid == cma_id->device->node_guid) {
273 			device->refcount++;
274 			isert_info("Found iser device %p refcount %d\n",
275 				   device, device->refcount);
276 			mutex_unlock(&device_list_mutex);
277 			return device;
278 		}
279 	}
280 
281 	device = kzalloc(sizeof(struct isert_device), GFP_KERNEL);
282 	if (!device) {
283 		mutex_unlock(&device_list_mutex);
284 		return ERR_PTR(-ENOMEM);
285 	}
286 
287 	INIT_LIST_HEAD(&device->dev_node);
288 
289 	device->ib_device = cma_id->device;
290 	ret = isert_create_device_ib_res(device);
291 	if (ret) {
292 		kfree(device);
293 		mutex_unlock(&device_list_mutex);
294 		return ERR_PTR(ret);
295 	}
296 
297 	device->refcount++;
298 	list_add_tail(&device->dev_node, &device_list);
299 	isert_info("Created a new iser device %p refcount %d\n",
300 		   device, device->refcount);
301 	mutex_unlock(&device_list_mutex);
302 
303 	return device;
304 }
305 
306 static void
307 isert_init_conn(struct isert_conn *isert_conn)
308 {
309 	isert_conn->state = ISER_CONN_INIT;
310 	INIT_LIST_HEAD(&isert_conn->node);
311 	init_completion(&isert_conn->login_comp);
312 	init_completion(&isert_conn->login_req_comp);
313 	init_waitqueue_head(&isert_conn->rem_wait);
314 	kref_init(&isert_conn->kref);
315 	mutex_init(&isert_conn->mutex);
316 	INIT_WORK(&isert_conn->release_work, isert_release_work);
317 }
318 
319 static void
320 isert_free_login_buf(struct isert_conn *isert_conn)
321 {
322 	struct ib_device *ib_dev = isert_conn->device->ib_device;
323 
324 	ib_dma_unmap_single(ib_dev, isert_conn->login_rsp_dma,
325 			    ISER_RX_PAYLOAD_SIZE, DMA_TO_DEVICE);
326 	kfree(isert_conn->login_rsp_buf);
327 
328 	ib_dma_unmap_single(ib_dev, isert_conn->login_desc->dma_addr,
329 			    ISER_RX_SIZE, DMA_FROM_DEVICE);
330 	kfree(isert_conn->login_desc);
331 }
332 
333 static int
334 isert_alloc_login_buf(struct isert_conn *isert_conn,
335 		      struct ib_device *ib_dev)
336 {
337 	int ret;
338 
339 	isert_conn->login_desc = kzalloc(sizeof(*isert_conn->login_desc),
340 			GFP_KERNEL);
341 	if (!isert_conn->login_desc)
342 		return -ENOMEM;
343 
344 	isert_conn->login_desc->dma_addr = ib_dma_map_single(ib_dev,
345 				isert_conn->login_desc->buf,
346 				ISER_RX_SIZE, DMA_FROM_DEVICE);
347 	ret = ib_dma_mapping_error(ib_dev, isert_conn->login_desc->dma_addr);
348 	if (ret) {
349 		isert_err("login_desc dma mapping error: %d\n", ret);
350 		isert_conn->login_desc->dma_addr = 0;
351 		goto out_free_login_desc;
352 	}
353 
354 	isert_conn->login_rsp_buf = kzalloc(ISER_RX_PAYLOAD_SIZE, GFP_KERNEL);
355 	if (!isert_conn->login_rsp_buf) {
356 		ret = -ENOMEM;
357 		goto out_unmap_login_desc;
358 	}
359 
360 	isert_conn->login_rsp_dma = ib_dma_map_single(ib_dev,
361 					isert_conn->login_rsp_buf,
362 					ISER_RX_PAYLOAD_SIZE, DMA_TO_DEVICE);
363 	ret = ib_dma_mapping_error(ib_dev, isert_conn->login_rsp_dma);
364 	if (ret) {
365 		isert_err("login_rsp_dma mapping error: %d\n", ret);
366 		isert_conn->login_rsp_dma = 0;
367 		goto out_free_login_rsp_buf;
368 	}
369 
370 	return 0;
371 
372 out_free_login_rsp_buf:
373 	kfree(isert_conn->login_rsp_buf);
374 out_unmap_login_desc:
375 	ib_dma_unmap_single(ib_dev, isert_conn->login_desc->dma_addr,
376 			    ISER_RX_SIZE, DMA_FROM_DEVICE);
377 out_free_login_desc:
378 	kfree(isert_conn->login_desc);
379 	return ret;
380 }
381 
382 static void
383 isert_set_nego_params(struct isert_conn *isert_conn,
384 		      struct rdma_conn_param *param)
385 {
386 	struct ib_device_attr *attr = &isert_conn->device->ib_device->attrs;
387 
388 	/* Set max inflight RDMA READ requests */
389 	isert_conn->initiator_depth = min_t(u8, param->initiator_depth,
390 				attr->max_qp_init_rd_atom);
391 	isert_dbg("Using initiator_depth: %u\n", isert_conn->initiator_depth);
392 
393 	if (param->private_data) {
394 		u8 flags = *(u8 *)param->private_data;
395 
396 		/*
397 		 * use remote invalidation if the both initiator
398 		 * and the HCA support it
399 		 */
400 		isert_conn->snd_w_inv = !(flags & ISER_SEND_W_INV_NOT_SUP) &&
401 					  (attr->device_cap_flags &
402 					   IB_DEVICE_MEM_MGT_EXTENSIONS);
403 		if (isert_conn->snd_w_inv)
404 			isert_info("Using remote invalidation\n");
405 	}
406 }
407 
408 static void
409 isert_destroy_qp(struct isert_conn *isert_conn)
410 {
411 	ib_destroy_qp(isert_conn->qp);
412 	ib_cq_pool_put(isert_conn->cq, isert_conn->cq_size);
413 }
414 
415 static int
416 isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
417 {
418 	struct isert_np *isert_np = cma_id->context;
419 	struct iscsi_np *np = isert_np->np;
420 	struct isert_conn *isert_conn;
421 	struct isert_device *device;
422 	int ret = 0;
423 
424 	spin_lock_bh(&np->np_thread_lock);
425 	if (!np->enabled) {
426 		spin_unlock_bh(&np->np_thread_lock);
427 		isert_dbg("iscsi_np is not enabled, reject connect request\n");
428 		return rdma_reject(cma_id, NULL, 0, IB_CM_REJ_CONSUMER_DEFINED);
429 	}
430 	spin_unlock_bh(&np->np_thread_lock);
431 
432 	isert_dbg("cma_id: %p, portal: %p\n",
433 		 cma_id, cma_id->context);
434 
435 	isert_conn = kzalloc(sizeof(struct isert_conn), GFP_KERNEL);
436 	if (!isert_conn)
437 		return -ENOMEM;
438 
439 	isert_init_conn(isert_conn);
440 	isert_conn->cm_id = cma_id;
441 
442 	device = isert_device_get(cma_id);
443 	if (IS_ERR(device)) {
444 		ret = PTR_ERR(device);
445 		goto out;
446 	}
447 	isert_conn->device = device;
448 
449 	ret = isert_alloc_login_buf(isert_conn, cma_id->device);
450 	if (ret)
451 		goto out_conn_dev;
452 
453 	isert_set_nego_params(isert_conn, &event->param.conn);
454 
455 	isert_conn->qp = isert_create_qp(isert_conn, cma_id);
456 	if (IS_ERR(isert_conn->qp)) {
457 		ret = PTR_ERR(isert_conn->qp);
458 		goto out_rsp_dma_map;
459 	}
460 
461 	ret = isert_login_post_recv(isert_conn);
462 	if (ret)
463 		goto out_destroy_qp;
464 
465 	ret = isert_rdma_accept(isert_conn);
466 	if (ret)
467 		goto out_destroy_qp;
468 
469 	mutex_lock(&isert_np->mutex);
470 	list_add_tail(&isert_conn->node, &isert_np->accepted);
471 	mutex_unlock(&isert_np->mutex);
472 
473 	return 0;
474 
475 out_destroy_qp:
476 	isert_destroy_qp(isert_conn);
477 out_rsp_dma_map:
478 	isert_free_login_buf(isert_conn);
479 out_conn_dev:
480 	isert_device_put(device);
481 out:
482 	kfree(isert_conn);
483 	rdma_reject(cma_id, NULL, 0, IB_CM_REJ_CONSUMER_DEFINED);
484 	return ret;
485 }
486 
487 static void
488 isert_connect_release(struct isert_conn *isert_conn)
489 {
490 	struct isert_device *device = isert_conn->device;
491 
492 	isert_dbg("conn %p\n", isert_conn);
493 
494 	BUG_ON(!device);
495 
496 	isert_free_rx_descriptors(isert_conn);
497 	if (isert_conn->cm_id &&
498 	    !isert_conn->dev_removed)
499 		rdma_destroy_id(isert_conn->cm_id);
500 
501 	if (isert_conn->qp)
502 		isert_destroy_qp(isert_conn);
503 
504 	if (isert_conn->login_desc)
505 		isert_free_login_buf(isert_conn);
506 
507 	isert_device_put(device);
508 
509 	if (isert_conn->dev_removed)
510 		wake_up_interruptible(&isert_conn->rem_wait);
511 	else
512 		kfree(isert_conn);
513 }
514 
515 static void
516 isert_connected_handler(struct rdma_cm_id *cma_id)
517 {
518 	struct isert_conn *isert_conn = cma_id->qp->qp_context;
519 	struct isert_np *isert_np = cma_id->context;
520 
521 	isert_info("conn %p\n", isert_conn);
522 
523 	mutex_lock(&isert_conn->mutex);
524 	isert_conn->state = ISER_CONN_UP;
525 	kref_get(&isert_conn->kref);
526 	mutex_unlock(&isert_conn->mutex);
527 
528 	mutex_lock(&isert_np->mutex);
529 	list_move_tail(&isert_conn->node, &isert_np->pending);
530 	mutex_unlock(&isert_np->mutex);
531 
532 	isert_info("np %p: Allow accept_np to continue\n", isert_np);
533 	up(&isert_np->sem);
534 }
535 
536 static void
537 isert_release_kref(struct kref *kref)
538 {
539 	struct isert_conn *isert_conn = container_of(kref,
540 				struct isert_conn, kref);
541 
542 	isert_info("conn %p final kref %s/%d\n", isert_conn, current->comm,
543 		   current->pid);
544 
545 	isert_connect_release(isert_conn);
546 }
547 
548 static void
549 isert_put_conn(struct isert_conn *isert_conn)
550 {
551 	kref_put(&isert_conn->kref, isert_release_kref);
552 }
553 
554 static void
555 isert_handle_unbound_conn(struct isert_conn *isert_conn)
556 {
557 	struct isert_np *isert_np = isert_conn->cm_id->context;
558 
559 	mutex_lock(&isert_np->mutex);
560 	if (!list_empty(&isert_conn->node)) {
561 		/*
562 		 * This means iscsi doesn't know this connection
563 		 * so schedule a cleanup ourselves
564 		 */
565 		list_del_init(&isert_conn->node);
566 		isert_put_conn(isert_conn);
567 		queue_work(isert_release_wq, &isert_conn->release_work);
568 	}
569 	mutex_unlock(&isert_np->mutex);
570 }
571 
572 /**
573  * isert_conn_terminate() - Initiate connection termination
574  * @isert_conn: isert connection struct
575  *
576  * Notes:
577  * In case the connection state is BOUND, move state
578  * to TEMINATING and start teardown sequence (rdma_disconnect).
579  * In case the connection state is UP, complete flush as well.
580  *
581  * This routine must be called with mutex held. Thus it is
582  * safe to call multiple times.
583  */
584 static void
585 isert_conn_terminate(struct isert_conn *isert_conn)
586 {
587 	int err;
588 
589 	if (isert_conn->state >= ISER_CONN_TERMINATING)
590 		return;
591 
592 	isert_info("Terminating conn %p state %d\n",
593 		   isert_conn, isert_conn->state);
594 	isert_conn->state = ISER_CONN_TERMINATING;
595 	err = rdma_disconnect(isert_conn->cm_id);
596 	if (err)
597 		isert_warn("Failed rdma_disconnect isert_conn %p\n",
598 			   isert_conn);
599 }
600 
601 static int
602 isert_np_cma_handler(struct isert_np *isert_np,
603 		     enum rdma_cm_event_type event)
604 {
605 	isert_dbg("%s (%d): isert np %p\n",
606 		  rdma_event_msg(event), event, isert_np);
607 
608 	switch (event) {
609 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
610 		isert_np->cm_id = NULL;
611 		break;
612 	case RDMA_CM_EVENT_ADDR_CHANGE:
613 		isert_np->cm_id = isert_setup_id(isert_np);
614 		if (IS_ERR(isert_np->cm_id)) {
615 			isert_err("isert np %p setup id failed: %ld\n",
616 				  isert_np, PTR_ERR(isert_np->cm_id));
617 			isert_np->cm_id = NULL;
618 		}
619 		break;
620 	default:
621 		isert_err("isert np %p Unexpected event %d\n",
622 			  isert_np, event);
623 	}
624 
625 	return -1;
626 }
627 
628 static int
629 isert_disconnected_handler(struct rdma_cm_id *cma_id,
630 			   enum rdma_cm_event_type event)
631 {
632 	struct isert_conn *isert_conn = cma_id->qp->qp_context;
633 
634 	mutex_lock(&isert_conn->mutex);
635 	switch (isert_conn->state) {
636 	case ISER_CONN_TERMINATING:
637 		break;
638 	case ISER_CONN_UP:
639 		isert_conn_terminate(isert_conn);
640 		ib_drain_qp(isert_conn->qp);
641 		isert_handle_unbound_conn(isert_conn);
642 		break;
643 	case ISER_CONN_BOUND:
644 	case ISER_CONN_FULL_FEATURE: /* FALLTHRU */
645 		iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
646 		break;
647 	default:
648 		isert_warn("conn %p terminating in state %d\n",
649 			   isert_conn, isert_conn->state);
650 	}
651 	mutex_unlock(&isert_conn->mutex);
652 
653 	return 0;
654 }
655 
656 static int
657 isert_connect_error(struct rdma_cm_id *cma_id)
658 {
659 	struct isert_conn *isert_conn = cma_id->qp->qp_context;
660 	struct isert_np *isert_np = cma_id->context;
661 
662 	ib_drain_qp(isert_conn->qp);
663 
664 	mutex_lock(&isert_np->mutex);
665 	list_del_init(&isert_conn->node);
666 	mutex_unlock(&isert_np->mutex);
667 	isert_conn->cm_id = NULL;
668 	isert_put_conn(isert_conn);
669 
670 	return -1;
671 }
672 
673 static int
674 isert_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
675 {
676 	struct isert_np *isert_np = cma_id->context;
677 	struct isert_conn *isert_conn;
678 	int ret = 0;
679 
680 	isert_info("%s (%d): status %d id %p np %p\n",
681 		   rdma_event_msg(event->event), event->event,
682 		   event->status, cma_id, cma_id->context);
683 
684 	if (isert_np->cm_id == cma_id)
685 		return isert_np_cma_handler(cma_id->context, event->event);
686 
687 	switch (event->event) {
688 	case RDMA_CM_EVENT_CONNECT_REQUEST:
689 		ret = isert_connect_request(cma_id, event);
690 		if (ret)
691 			isert_err("failed handle connect request %d\n", ret);
692 		break;
693 	case RDMA_CM_EVENT_ESTABLISHED:
694 		isert_connected_handler(cma_id);
695 		break;
696 	case RDMA_CM_EVENT_ADDR_CHANGE:
697 	case RDMA_CM_EVENT_DISCONNECTED:
698 	case RDMA_CM_EVENT_TIMEWAIT_EXIT:  /* FALLTHRU */
699 		ret = isert_disconnected_handler(cma_id, event->event);
700 		break;
701 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
702 		isert_conn = cma_id->qp->qp_context;
703 		isert_conn->dev_removed = true;
704 		isert_disconnected_handler(cma_id, event->event);
705 		wait_event_interruptible(isert_conn->rem_wait,
706 					 isert_conn->state == ISER_CONN_DOWN);
707 		kfree(isert_conn);
708 		/*
709 		 * return non-zero from the callback to destroy
710 		 * the rdma cm id
711 		 */
712 		return 1;
713 	case RDMA_CM_EVENT_REJECTED:
714 		isert_info("Connection rejected: %s\n",
715 			   rdma_reject_msg(cma_id, event->status));
716 		fallthrough;
717 	case RDMA_CM_EVENT_UNREACHABLE:
718 	case RDMA_CM_EVENT_CONNECT_ERROR:
719 		ret = isert_connect_error(cma_id);
720 		break;
721 	default:
722 		isert_err("Unhandled RDMA CMA event: %d\n", event->event);
723 		break;
724 	}
725 
726 	return ret;
727 }
728 
729 static int
730 isert_post_recvm(struct isert_conn *isert_conn, u32 count)
731 {
732 	struct ib_recv_wr *rx_wr;
733 	int i, ret;
734 	struct iser_rx_desc *rx_desc;
735 
736 	for (rx_wr = isert_conn->rx_wr, i = 0; i < count; i++, rx_wr++) {
737 		rx_desc = &isert_conn->rx_descs[i];
738 
739 		rx_wr->wr_cqe = &rx_desc->rx_cqe;
740 		rx_wr->sg_list = &rx_desc->rx_sg;
741 		rx_wr->num_sge = 1;
742 		rx_wr->next = rx_wr + 1;
743 		rx_desc->in_use = false;
744 	}
745 	rx_wr--;
746 	rx_wr->next = NULL; /* mark end of work requests list */
747 
748 	ret = ib_post_recv(isert_conn->qp, isert_conn->rx_wr, NULL);
749 	if (ret)
750 		isert_err("ib_post_recv() failed with ret: %d\n", ret);
751 
752 	return ret;
753 }
754 
755 static int
756 isert_post_recv(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc)
757 {
758 	struct ib_recv_wr rx_wr;
759 	int ret;
760 
761 	if (!rx_desc->in_use) {
762 		/*
763 		 * if the descriptor is not in-use we already reposted it
764 		 * for recv, so just silently return
765 		 */
766 		return 0;
767 	}
768 
769 	rx_desc->in_use = false;
770 	rx_wr.wr_cqe = &rx_desc->rx_cqe;
771 	rx_wr.sg_list = &rx_desc->rx_sg;
772 	rx_wr.num_sge = 1;
773 	rx_wr.next = NULL;
774 
775 	ret = ib_post_recv(isert_conn->qp, &rx_wr, NULL);
776 	if (ret)
777 		isert_err("ib_post_recv() failed with ret: %d\n", ret);
778 
779 	return ret;
780 }
781 
782 static int
783 isert_login_post_send(struct isert_conn *isert_conn, struct iser_tx_desc *tx_desc)
784 {
785 	struct ib_device *ib_dev = isert_conn->cm_id->device;
786 	struct ib_send_wr send_wr;
787 	int ret;
788 
789 	ib_dma_sync_single_for_device(ib_dev, tx_desc->dma_addr,
790 				      ISER_HEADERS_LEN, DMA_TO_DEVICE);
791 
792 	tx_desc->tx_cqe.done = isert_login_send_done;
793 
794 	send_wr.next	= NULL;
795 	send_wr.wr_cqe	= &tx_desc->tx_cqe;
796 	send_wr.sg_list	= tx_desc->tx_sg;
797 	send_wr.num_sge	= tx_desc->num_sge;
798 	send_wr.opcode	= IB_WR_SEND;
799 	send_wr.send_flags = IB_SEND_SIGNALED;
800 
801 	ret = ib_post_send(isert_conn->qp, &send_wr, NULL);
802 	if (ret)
803 		isert_err("ib_post_send() failed, ret: %d\n", ret);
804 
805 	return ret;
806 }
807 
808 static void
809 __isert_create_send_desc(struct isert_device *device,
810 			 struct iser_tx_desc *tx_desc)
811 {
812 
813 	memset(&tx_desc->iser_header, 0, sizeof(struct iser_ctrl));
814 	tx_desc->iser_header.flags = ISCSI_CTRL;
815 
816 	tx_desc->num_sge = 1;
817 
818 	if (tx_desc->tx_sg[0].lkey != device->pd->local_dma_lkey) {
819 		tx_desc->tx_sg[0].lkey = device->pd->local_dma_lkey;
820 		isert_dbg("tx_desc %p lkey mismatch, fixing\n", tx_desc);
821 	}
822 }
823 
824 static void
825 isert_create_send_desc(struct isert_conn *isert_conn,
826 		       struct isert_cmd *isert_cmd,
827 		       struct iser_tx_desc *tx_desc)
828 {
829 	struct isert_device *device = isert_conn->device;
830 	struct ib_device *ib_dev = device->ib_device;
831 
832 	ib_dma_sync_single_for_cpu(ib_dev, tx_desc->dma_addr,
833 				   ISER_HEADERS_LEN, DMA_TO_DEVICE);
834 
835 	__isert_create_send_desc(device, tx_desc);
836 }
837 
838 static int
839 isert_init_tx_hdrs(struct isert_conn *isert_conn,
840 		   struct iser_tx_desc *tx_desc)
841 {
842 	struct isert_device *device = isert_conn->device;
843 	struct ib_device *ib_dev = device->ib_device;
844 	u64 dma_addr;
845 
846 	dma_addr = ib_dma_map_single(ib_dev, (void *)tx_desc,
847 			ISER_HEADERS_LEN, DMA_TO_DEVICE);
848 	if (ib_dma_mapping_error(ib_dev, dma_addr)) {
849 		isert_err("ib_dma_mapping_error() failed\n");
850 		return -ENOMEM;
851 	}
852 
853 	tx_desc->dma_addr = dma_addr;
854 	tx_desc->tx_sg[0].addr	= tx_desc->dma_addr;
855 	tx_desc->tx_sg[0].length = ISER_HEADERS_LEN;
856 	tx_desc->tx_sg[0].lkey = device->pd->local_dma_lkey;
857 
858 	isert_dbg("Setup tx_sg[0].addr: 0x%llx length: %u lkey: 0x%x\n",
859 		  tx_desc->tx_sg[0].addr, tx_desc->tx_sg[0].length,
860 		  tx_desc->tx_sg[0].lkey);
861 
862 	return 0;
863 }
864 
865 static void
866 isert_init_send_wr(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
867 		   struct ib_send_wr *send_wr)
868 {
869 	struct iser_tx_desc *tx_desc = &isert_cmd->tx_desc;
870 
871 	tx_desc->tx_cqe.done = isert_send_done;
872 	send_wr->wr_cqe = &tx_desc->tx_cqe;
873 
874 	if (isert_conn->snd_w_inv && isert_cmd->inv_rkey) {
875 		send_wr->opcode  = IB_WR_SEND_WITH_INV;
876 		send_wr->ex.invalidate_rkey = isert_cmd->inv_rkey;
877 	} else {
878 		send_wr->opcode = IB_WR_SEND;
879 	}
880 
881 	send_wr->sg_list = &tx_desc->tx_sg[0];
882 	send_wr->num_sge = isert_cmd->tx_desc.num_sge;
883 	send_wr->send_flags = IB_SEND_SIGNALED;
884 }
885 
886 static int
887 isert_login_post_recv(struct isert_conn *isert_conn)
888 {
889 	struct ib_recv_wr rx_wr;
890 	struct ib_sge sge;
891 	int ret;
892 
893 	memset(&sge, 0, sizeof(struct ib_sge));
894 	sge.addr = isert_conn->login_desc->dma_addr +
895 		isert_get_hdr_offset(isert_conn->login_desc);
896 	sge.length = ISER_RX_PAYLOAD_SIZE;
897 	sge.lkey = isert_conn->device->pd->local_dma_lkey;
898 
899 	isert_dbg("Setup sge: addr: %llx length: %d 0x%08x\n",
900 		sge.addr, sge.length, sge.lkey);
901 
902 	isert_conn->login_desc->rx_cqe.done = isert_login_recv_done;
903 
904 	memset(&rx_wr, 0, sizeof(struct ib_recv_wr));
905 	rx_wr.wr_cqe = &isert_conn->login_desc->rx_cqe;
906 	rx_wr.sg_list = &sge;
907 	rx_wr.num_sge = 1;
908 
909 	ret = ib_post_recv(isert_conn->qp, &rx_wr, NULL);
910 	if (ret)
911 		isert_err("ib_post_recv() failed: %d\n", ret);
912 
913 	return ret;
914 }
915 
916 static int
917 isert_put_login_tx(struct iscsit_conn *conn, struct iscsi_login *login,
918 		   u32 length)
919 {
920 	struct isert_conn *isert_conn = conn->context;
921 	struct isert_device *device = isert_conn->device;
922 	struct ib_device *ib_dev = device->ib_device;
923 	struct iser_tx_desc *tx_desc = &isert_conn->login_tx_desc;
924 	int ret;
925 
926 	__isert_create_send_desc(device, tx_desc);
927 
928 	memcpy(&tx_desc->iscsi_header, &login->rsp[0],
929 	       sizeof(struct iscsi_hdr));
930 
931 	isert_init_tx_hdrs(isert_conn, tx_desc);
932 
933 	if (length > 0) {
934 		struct ib_sge *tx_dsg = &tx_desc->tx_sg[1];
935 
936 		ib_dma_sync_single_for_cpu(ib_dev, isert_conn->login_rsp_dma,
937 					   length, DMA_TO_DEVICE);
938 
939 		memcpy(isert_conn->login_rsp_buf, login->rsp_buf, length);
940 
941 		ib_dma_sync_single_for_device(ib_dev, isert_conn->login_rsp_dma,
942 					      length, DMA_TO_DEVICE);
943 
944 		tx_dsg->addr	= isert_conn->login_rsp_dma;
945 		tx_dsg->length	= length;
946 		tx_dsg->lkey	= isert_conn->device->pd->local_dma_lkey;
947 		tx_desc->num_sge = 2;
948 	}
949 	if (!login->login_failed) {
950 		if (login->login_complete) {
951 			ret = isert_alloc_rx_descriptors(isert_conn);
952 			if (ret)
953 				return ret;
954 
955 			ret = isert_post_recvm(isert_conn,
956 					       ISERT_QP_MAX_RECV_DTOS);
957 			if (ret)
958 				return ret;
959 
960 			/* Now we are in FULL_FEATURE phase */
961 			mutex_lock(&isert_conn->mutex);
962 			isert_conn->state = ISER_CONN_FULL_FEATURE;
963 			mutex_unlock(&isert_conn->mutex);
964 			goto post_send;
965 		}
966 
967 		ret = isert_login_post_recv(isert_conn);
968 		if (ret)
969 			return ret;
970 	}
971 post_send:
972 	ret = isert_login_post_send(isert_conn, tx_desc);
973 	if (ret)
974 		return ret;
975 
976 	return 0;
977 }
978 
979 static void
980 isert_rx_login_req(struct isert_conn *isert_conn)
981 {
982 	struct iser_rx_desc *rx_desc = isert_conn->login_desc;
983 	int rx_buflen = isert_conn->login_req_len;
984 	struct iscsit_conn *conn = isert_conn->conn;
985 	struct iscsi_login *login = conn->conn_login;
986 	int size;
987 
988 	isert_info("conn %p\n", isert_conn);
989 
990 	WARN_ON_ONCE(!login);
991 
992 	if (login->first_request) {
993 		struct iscsi_login_req *login_req =
994 			(struct iscsi_login_req *)isert_get_iscsi_hdr(rx_desc);
995 		/*
996 		 * Setup the initial iscsi_login values from the leading
997 		 * login request PDU.
998 		 */
999 		login->leading_connection = (!login_req->tsih) ? 1 : 0;
1000 		login->current_stage = ISCSI_LOGIN_CURRENT_STAGE(
1001 				login_req->flags);
1002 		login->version_min	= login_req->min_version;
1003 		login->version_max	= login_req->max_version;
1004 		memcpy(login->isid, login_req->isid, 6);
1005 		login->cmd_sn		= be32_to_cpu(login_req->cmdsn);
1006 		login->init_task_tag	= login_req->itt;
1007 		login->initial_exp_statsn = be32_to_cpu(login_req->exp_statsn);
1008 		login->cid		= be16_to_cpu(login_req->cid);
1009 		login->tsih		= be16_to_cpu(login_req->tsih);
1010 	}
1011 
1012 	memcpy(&login->req[0], isert_get_iscsi_hdr(rx_desc), ISCSI_HDR_LEN);
1013 
1014 	size = min(rx_buflen, MAX_KEY_VALUE_PAIRS);
1015 	isert_dbg("Using login payload size: %d, rx_buflen: %d "
1016 		  "MAX_KEY_VALUE_PAIRS: %d\n", size, rx_buflen,
1017 		  MAX_KEY_VALUE_PAIRS);
1018 	memcpy(login->req_buf, isert_get_data(rx_desc), size);
1019 
1020 	if (login->first_request) {
1021 		complete(&isert_conn->login_comp);
1022 		return;
1023 	}
1024 	queue_delayed_work(isert_login_wq, &conn->login_work, 0);
1025 }
1026 
1027 static struct iscsit_cmd
1028 *isert_allocate_cmd(struct iscsit_conn *conn, struct iser_rx_desc *rx_desc)
1029 {
1030 	struct isert_conn *isert_conn = conn->context;
1031 	struct isert_cmd *isert_cmd;
1032 	struct iscsit_cmd *cmd;
1033 
1034 	cmd = iscsit_allocate_cmd(conn, TASK_INTERRUPTIBLE);
1035 	if (!cmd) {
1036 		isert_err("Unable to allocate iscsit_cmd + isert_cmd\n");
1037 		return NULL;
1038 	}
1039 	isert_cmd = iscsit_priv_cmd(cmd);
1040 	isert_cmd->conn = isert_conn;
1041 	isert_cmd->iscsit_cmd = cmd;
1042 	isert_cmd->rx_desc = rx_desc;
1043 
1044 	return cmd;
1045 }
1046 
1047 static int
1048 isert_handle_scsi_cmd(struct isert_conn *isert_conn,
1049 		      struct isert_cmd *isert_cmd, struct iscsit_cmd *cmd,
1050 		      struct iser_rx_desc *rx_desc, unsigned char *buf)
1051 {
1052 	struct iscsit_conn *conn = isert_conn->conn;
1053 	struct iscsi_scsi_req *hdr = (struct iscsi_scsi_req *)buf;
1054 	int imm_data, imm_data_len, unsol_data, sg_nents, rc;
1055 	bool dump_payload = false;
1056 	unsigned int data_len;
1057 
1058 	rc = iscsit_setup_scsi_cmd(conn, cmd, buf);
1059 	if (rc < 0)
1060 		return rc;
1061 
1062 	imm_data = cmd->immediate_data;
1063 	imm_data_len = cmd->first_burst_len;
1064 	unsol_data = cmd->unsolicited_data;
1065 	data_len = cmd->se_cmd.data_length;
1066 
1067 	if (imm_data && imm_data_len == data_len)
1068 		cmd->se_cmd.se_cmd_flags |= SCF_PASSTHROUGH_SG_TO_MEM_NOALLOC;
1069 	rc = iscsit_process_scsi_cmd(conn, cmd, hdr);
1070 	if (rc < 0) {
1071 		return 0;
1072 	} else if (rc > 0) {
1073 		dump_payload = true;
1074 		goto sequence_cmd;
1075 	}
1076 
1077 	if (!imm_data)
1078 		return 0;
1079 
1080 	if (imm_data_len != data_len) {
1081 		sg_nents = max(1UL, DIV_ROUND_UP(imm_data_len, PAGE_SIZE));
1082 		sg_copy_from_buffer(cmd->se_cmd.t_data_sg, sg_nents,
1083 				    isert_get_data(rx_desc), imm_data_len);
1084 		isert_dbg("Copy Immediate sg_nents: %u imm_data_len: %d\n",
1085 			  sg_nents, imm_data_len);
1086 	} else {
1087 		sg_init_table(&isert_cmd->sg, 1);
1088 		cmd->se_cmd.t_data_sg = &isert_cmd->sg;
1089 		cmd->se_cmd.t_data_nents = 1;
1090 		sg_set_buf(&isert_cmd->sg, isert_get_data(rx_desc),
1091 				imm_data_len);
1092 		isert_dbg("Transfer Immediate imm_data_len: %d\n",
1093 			  imm_data_len);
1094 	}
1095 
1096 	cmd->write_data_done += imm_data_len;
1097 
1098 	if (cmd->write_data_done == cmd->se_cmd.data_length) {
1099 		spin_lock_bh(&cmd->istate_lock);
1100 		cmd->cmd_flags |= ICF_GOT_LAST_DATAOUT;
1101 		cmd->i_state = ISTATE_RECEIVED_LAST_DATAOUT;
1102 		spin_unlock_bh(&cmd->istate_lock);
1103 	}
1104 
1105 sequence_cmd:
1106 	rc = iscsit_sequence_cmd(conn, cmd, buf, hdr->cmdsn);
1107 
1108 	if (!rc && !dump_payload && unsol_data)
1109 		iscsit_set_unsolicited_dataout(cmd);
1110 	else if (dump_payload && imm_data)
1111 		target_put_sess_cmd(&cmd->se_cmd);
1112 
1113 	return 0;
1114 }
1115 
1116 static int
1117 isert_handle_iscsi_dataout(struct isert_conn *isert_conn,
1118 			   struct iser_rx_desc *rx_desc, unsigned char *buf)
1119 {
1120 	struct scatterlist *sg_start;
1121 	struct iscsit_conn *conn = isert_conn->conn;
1122 	struct iscsit_cmd *cmd = NULL;
1123 	struct iscsi_data *hdr = (struct iscsi_data *)buf;
1124 	u32 unsol_data_len = ntoh24(hdr->dlength);
1125 	int rc, sg_nents, sg_off, page_off;
1126 
1127 	rc = iscsit_check_dataout_hdr(conn, buf, &cmd);
1128 	if (rc < 0)
1129 		return rc;
1130 	else if (!cmd)
1131 		return 0;
1132 	/*
1133 	 * FIXME: Unexpected unsolicited_data out
1134 	 */
1135 	if (!cmd->unsolicited_data) {
1136 		isert_err("Received unexpected solicited data payload\n");
1137 		dump_stack();
1138 		return -1;
1139 	}
1140 
1141 	isert_dbg("Unsolicited DataOut unsol_data_len: %u, "
1142 		  "write_data_done: %u, data_length: %u\n",
1143 		  unsol_data_len,  cmd->write_data_done,
1144 		  cmd->se_cmd.data_length);
1145 
1146 	sg_off = cmd->write_data_done / PAGE_SIZE;
1147 	sg_start = &cmd->se_cmd.t_data_sg[sg_off];
1148 	sg_nents = max(1UL, DIV_ROUND_UP(unsol_data_len, PAGE_SIZE));
1149 	page_off = cmd->write_data_done % PAGE_SIZE;
1150 	/*
1151 	 * FIXME: Non page-aligned unsolicited_data out
1152 	 */
1153 	if (page_off) {
1154 		isert_err("unexpected non-page aligned data payload\n");
1155 		dump_stack();
1156 		return -1;
1157 	}
1158 	isert_dbg("Copying DataOut: sg_start: %p, sg_off: %u "
1159 		  "sg_nents: %u from %p %u\n", sg_start, sg_off,
1160 		  sg_nents, isert_get_data(rx_desc), unsol_data_len);
1161 
1162 	sg_copy_from_buffer(sg_start, sg_nents, isert_get_data(rx_desc),
1163 			    unsol_data_len);
1164 
1165 	rc = iscsit_check_dataout_payload(cmd, hdr, false);
1166 	if (rc < 0)
1167 		return rc;
1168 
1169 	/*
1170 	 * multiple data-outs on the same command can arrive -
1171 	 * so post the buffer before hand
1172 	 */
1173 	return isert_post_recv(isert_conn, rx_desc);
1174 }
1175 
1176 static int
1177 isert_handle_nop_out(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
1178 		     struct iscsit_cmd *cmd, struct iser_rx_desc *rx_desc,
1179 		     unsigned char *buf)
1180 {
1181 	struct iscsit_conn *conn = isert_conn->conn;
1182 	struct iscsi_nopout *hdr = (struct iscsi_nopout *)buf;
1183 	int rc;
1184 
1185 	rc = iscsit_setup_nop_out(conn, cmd, hdr);
1186 	if (rc < 0)
1187 		return rc;
1188 	/*
1189 	 * FIXME: Add support for NOPOUT payload using unsolicited RDMA payload
1190 	 */
1191 
1192 	return iscsit_process_nop_out(conn, cmd, hdr);
1193 }
1194 
1195 static int
1196 isert_handle_text_cmd(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
1197 		      struct iscsit_cmd *cmd, struct iser_rx_desc *rx_desc,
1198 		      struct iscsi_text *hdr)
1199 {
1200 	struct iscsit_conn *conn = isert_conn->conn;
1201 	u32 payload_length = ntoh24(hdr->dlength);
1202 	int rc;
1203 	unsigned char *text_in = NULL;
1204 
1205 	rc = iscsit_setup_text_cmd(conn, cmd, hdr);
1206 	if (rc < 0)
1207 		return rc;
1208 
1209 	if (payload_length) {
1210 		text_in = kzalloc(payload_length, GFP_KERNEL);
1211 		if (!text_in)
1212 			return -ENOMEM;
1213 	}
1214 	cmd->text_in_ptr = text_in;
1215 
1216 	memcpy(cmd->text_in_ptr, isert_get_data(rx_desc), payload_length);
1217 
1218 	return iscsit_process_text_cmd(conn, cmd, hdr);
1219 }
1220 
1221 static int
1222 isert_rx_opcode(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc,
1223 		uint32_t read_stag, uint64_t read_va,
1224 		uint32_t write_stag, uint64_t write_va)
1225 {
1226 	struct iscsi_hdr *hdr = isert_get_iscsi_hdr(rx_desc);
1227 	struct iscsit_conn *conn = isert_conn->conn;
1228 	struct iscsit_cmd *cmd;
1229 	struct isert_cmd *isert_cmd;
1230 	int ret = -EINVAL;
1231 	u8 opcode = (hdr->opcode & ISCSI_OPCODE_MASK);
1232 
1233 	if (conn->sess->sess_ops->SessionType &&
1234 	   (!(opcode & ISCSI_OP_TEXT) || !(opcode & ISCSI_OP_LOGOUT))) {
1235 		isert_err("Got illegal opcode: 0x%02x in SessionType=Discovery,"
1236 			  " ignoring\n", opcode);
1237 		return 0;
1238 	}
1239 
1240 	switch (opcode) {
1241 	case ISCSI_OP_SCSI_CMD:
1242 		cmd = isert_allocate_cmd(conn, rx_desc);
1243 		if (!cmd)
1244 			break;
1245 
1246 		isert_cmd = iscsit_priv_cmd(cmd);
1247 		isert_cmd->read_stag = read_stag;
1248 		isert_cmd->read_va = read_va;
1249 		isert_cmd->write_stag = write_stag;
1250 		isert_cmd->write_va = write_va;
1251 		isert_cmd->inv_rkey = read_stag ? read_stag : write_stag;
1252 
1253 		ret = isert_handle_scsi_cmd(isert_conn, isert_cmd, cmd,
1254 					rx_desc, (unsigned char *)hdr);
1255 		break;
1256 	case ISCSI_OP_NOOP_OUT:
1257 		cmd = isert_allocate_cmd(conn, rx_desc);
1258 		if (!cmd)
1259 			break;
1260 
1261 		isert_cmd = iscsit_priv_cmd(cmd);
1262 		ret = isert_handle_nop_out(isert_conn, isert_cmd, cmd,
1263 					   rx_desc, (unsigned char *)hdr);
1264 		break;
1265 	case ISCSI_OP_SCSI_DATA_OUT:
1266 		ret = isert_handle_iscsi_dataout(isert_conn, rx_desc,
1267 						(unsigned char *)hdr);
1268 		break;
1269 	case ISCSI_OP_SCSI_TMFUNC:
1270 		cmd = isert_allocate_cmd(conn, rx_desc);
1271 		if (!cmd)
1272 			break;
1273 
1274 		ret = iscsit_handle_task_mgt_cmd(conn, cmd,
1275 						(unsigned char *)hdr);
1276 		break;
1277 	case ISCSI_OP_LOGOUT:
1278 		cmd = isert_allocate_cmd(conn, rx_desc);
1279 		if (!cmd)
1280 			break;
1281 
1282 		ret = iscsit_handle_logout_cmd(conn, cmd, (unsigned char *)hdr);
1283 		break;
1284 	case ISCSI_OP_TEXT:
1285 		if (be32_to_cpu(hdr->ttt) != 0xFFFFFFFF)
1286 			cmd = iscsit_find_cmd_from_itt(conn, hdr->itt);
1287 		else
1288 			cmd = isert_allocate_cmd(conn, rx_desc);
1289 
1290 		if (!cmd)
1291 			break;
1292 
1293 		isert_cmd = iscsit_priv_cmd(cmd);
1294 		ret = isert_handle_text_cmd(isert_conn, isert_cmd, cmd,
1295 					    rx_desc, (struct iscsi_text *)hdr);
1296 		break;
1297 	default:
1298 		isert_err("Got unknown iSCSI OpCode: 0x%02x\n", opcode);
1299 		dump_stack();
1300 		break;
1301 	}
1302 
1303 	return ret;
1304 }
1305 
1306 static void
1307 isert_print_wc(struct ib_wc *wc, const char *type)
1308 {
1309 	if (wc->status != IB_WC_WR_FLUSH_ERR)
1310 		isert_err("%s failure: %s (%d) vend_err %x\n", type,
1311 			  ib_wc_status_msg(wc->status), wc->status,
1312 			  wc->vendor_err);
1313 	else
1314 		isert_dbg("%s failure: %s (%d)\n", type,
1315 			  ib_wc_status_msg(wc->status), wc->status);
1316 }
1317 
1318 static void
1319 isert_recv_done(struct ib_cq *cq, struct ib_wc *wc)
1320 {
1321 	struct isert_conn *isert_conn = wc->qp->qp_context;
1322 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1323 	struct iser_rx_desc *rx_desc = cqe_to_rx_desc(wc->wr_cqe);
1324 	struct iscsi_hdr *hdr = isert_get_iscsi_hdr(rx_desc);
1325 	struct iser_ctrl *iser_ctrl = isert_get_iser_hdr(rx_desc);
1326 	uint64_t read_va = 0, write_va = 0;
1327 	uint32_t read_stag = 0, write_stag = 0;
1328 
1329 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1330 		isert_print_wc(wc, "recv");
1331 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1332 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1333 		return;
1334 	}
1335 
1336 	rx_desc->in_use = true;
1337 
1338 	ib_dma_sync_single_for_cpu(ib_dev, rx_desc->dma_addr,
1339 			ISER_RX_SIZE, DMA_FROM_DEVICE);
1340 
1341 	isert_dbg("DMA: 0x%llx, iSCSI opcode: 0x%02x, ITT: 0x%08x, flags: 0x%02x dlen: %d\n",
1342 		 rx_desc->dma_addr, hdr->opcode, hdr->itt, hdr->flags,
1343 		 (int)(wc->byte_len - ISER_HEADERS_LEN));
1344 
1345 	switch (iser_ctrl->flags & 0xF0) {
1346 	case ISCSI_CTRL:
1347 		if (iser_ctrl->flags & ISER_RSV) {
1348 			read_stag = be32_to_cpu(iser_ctrl->read_stag);
1349 			read_va = be64_to_cpu(iser_ctrl->read_va);
1350 			isert_dbg("ISER_RSV: read_stag: 0x%x read_va: 0x%llx\n",
1351 				  read_stag, (unsigned long long)read_va);
1352 		}
1353 		if (iser_ctrl->flags & ISER_WSV) {
1354 			write_stag = be32_to_cpu(iser_ctrl->write_stag);
1355 			write_va = be64_to_cpu(iser_ctrl->write_va);
1356 			isert_dbg("ISER_WSV: write_stag: 0x%x write_va: 0x%llx\n",
1357 				  write_stag, (unsigned long long)write_va);
1358 		}
1359 
1360 		isert_dbg("ISER ISCSI_CTRL PDU\n");
1361 		break;
1362 	case ISER_HELLO:
1363 		isert_err("iSER Hello message\n");
1364 		break;
1365 	default:
1366 		isert_warn("Unknown iSER hdr flags: 0x%02x\n", iser_ctrl->flags);
1367 		break;
1368 	}
1369 
1370 	isert_rx_opcode(isert_conn, rx_desc,
1371 			read_stag, read_va, write_stag, write_va);
1372 
1373 	ib_dma_sync_single_for_device(ib_dev, rx_desc->dma_addr,
1374 			ISER_RX_SIZE, DMA_FROM_DEVICE);
1375 }
1376 
1377 static void
1378 isert_login_recv_done(struct ib_cq *cq, struct ib_wc *wc)
1379 {
1380 	struct isert_conn *isert_conn = wc->qp->qp_context;
1381 	struct ib_device *ib_dev = isert_conn->device->ib_device;
1382 
1383 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1384 		isert_print_wc(wc, "login recv");
1385 		return;
1386 	}
1387 
1388 	ib_dma_sync_single_for_cpu(ib_dev, isert_conn->login_desc->dma_addr,
1389 			ISER_RX_SIZE, DMA_FROM_DEVICE);
1390 
1391 	isert_conn->login_req_len = wc->byte_len - ISER_HEADERS_LEN;
1392 
1393 	if (isert_conn->conn) {
1394 		struct iscsi_login *login = isert_conn->conn->conn_login;
1395 
1396 		if (login && !login->first_request)
1397 			isert_rx_login_req(isert_conn);
1398 	}
1399 
1400 	mutex_lock(&isert_conn->mutex);
1401 	complete(&isert_conn->login_req_comp);
1402 	mutex_unlock(&isert_conn->mutex);
1403 
1404 	ib_dma_sync_single_for_device(ib_dev, isert_conn->login_desc->dma_addr,
1405 				ISER_RX_SIZE, DMA_FROM_DEVICE);
1406 }
1407 
1408 static void
1409 isert_rdma_rw_ctx_destroy(struct isert_cmd *cmd, struct isert_conn *conn)
1410 {
1411 	struct se_cmd *se_cmd = &cmd->iscsit_cmd->se_cmd;
1412 	enum dma_data_direction dir = target_reverse_dma_direction(se_cmd);
1413 
1414 	if (!cmd->rw.nr_ops)
1415 		return;
1416 
1417 	if (isert_prot_cmd(conn, se_cmd)) {
1418 		rdma_rw_ctx_destroy_signature(&cmd->rw, conn->qp,
1419 				conn->cm_id->port_num, se_cmd->t_data_sg,
1420 				se_cmd->t_data_nents, se_cmd->t_prot_sg,
1421 				se_cmd->t_prot_nents, dir);
1422 	} else {
1423 		rdma_rw_ctx_destroy(&cmd->rw, conn->qp, conn->cm_id->port_num,
1424 				se_cmd->t_data_sg, se_cmd->t_data_nents, dir);
1425 	}
1426 
1427 	cmd->rw.nr_ops = 0;
1428 }
1429 
1430 static void
1431 isert_put_cmd(struct isert_cmd *isert_cmd, bool comp_err)
1432 {
1433 	struct iscsit_cmd *cmd = isert_cmd->iscsit_cmd;
1434 	struct isert_conn *isert_conn = isert_cmd->conn;
1435 	struct iscsit_conn *conn = isert_conn->conn;
1436 	struct iscsi_text_rsp *hdr;
1437 
1438 	isert_dbg("Cmd %p\n", isert_cmd);
1439 
1440 	switch (cmd->iscsi_opcode) {
1441 	case ISCSI_OP_SCSI_CMD:
1442 		spin_lock_bh(&conn->cmd_lock);
1443 		if (!list_empty(&cmd->i_conn_node))
1444 			list_del_init(&cmd->i_conn_node);
1445 		spin_unlock_bh(&conn->cmd_lock);
1446 
1447 		if (cmd->data_direction == DMA_TO_DEVICE) {
1448 			iscsit_stop_dataout_timer(cmd);
1449 			/*
1450 			 * Check for special case during comp_err where
1451 			 * WRITE_PENDING has been handed off from core,
1452 			 * but requires an extra target_put_sess_cmd()
1453 			 * before transport_generic_free_cmd() below.
1454 			 */
1455 			if (comp_err &&
1456 			    cmd->se_cmd.t_state == TRANSPORT_WRITE_PENDING) {
1457 				struct se_cmd *se_cmd = &cmd->se_cmd;
1458 
1459 				target_put_sess_cmd(se_cmd);
1460 			}
1461 		}
1462 
1463 		isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1464 		transport_generic_free_cmd(&cmd->se_cmd, 0);
1465 		break;
1466 	case ISCSI_OP_SCSI_TMFUNC:
1467 		spin_lock_bh(&conn->cmd_lock);
1468 		if (!list_empty(&cmd->i_conn_node))
1469 			list_del_init(&cmd->i_conn_node);
1470 		spin_unlock_bh(&conn->cmd_lock);
1471 
1472 		transport_generic_free_cmd(&cmd->se_cmd, 0);
1473 		break;
1474 	case ISCSI_OP_REJECT:
1475 	case ISCSI_OP_NOOP_OUT:
1476 	case ISCSI_OP_TEXT:
1477 		hdr = (struct iscsi_text_rsp *)&isert_cmd->tx_desc.iscsi_header;
1478 		/* If the continue bit is on, keep the command alive */
1479 		if (hdr->flags & ISCSI_FLAG_TEXT_CONTINUE)
1480 			break;
1481 
1482 		spin_lock_bh(&conn->cmd_lock);
1483 		if (!list_empty(&cmd->i_conn_node))
1484 			list_del_init(&cmd->i_conn_node);
1485 		spin_unlock_bh(&conn->cmd_lock);
1486 
1487 		/*
1488 		 * Handle special case for REJECT when iscsi_add_reject*() has
1489 		 * overwritten the original iscsi_opcode assignment, and the
1490 		 * associated cmd->se_cmd needs to be released.
1491 		 */
1492 		if (cmd->se_cmd.se_tfo != NULL) {
1493 			isert_dbg("Calling transport_generic_free_cmd for 0x%02x\n",
1494 				 cmd->iscsi_opcode);
1495 			transport_generic_free_cmd(&cmd->se_cmd, 0);
1496 			break;
1497 		}
1498 		fallthrough;
1499 	default:
1500 		iscsit_release_cmd(cmd);
1501 		break;
1502 	}
1503 }
1504 
1505 static void
1506 isert_unmap_tx_desc(struct iser_tx_desc *tx_desc, struct ib_device *ib_dev)
1507 {
1508 	if (tx_desc->dma_addr != 0) {
1509 		isert_dbg("unmap single for tx_desc->dma_addr\n");
1510 		ib_dma_unmap_single(ib_dev, tx_desc->dma_addr,
1511 				    ISER_HEADERS_LEN, DMA_TO_DEVICE);
1512 		tx_desc->dma_addr = 0;
1513 	}
1514 }
1515 
1516 static void
1517 isert_completion_put(struct iser_tx_desc *tx_desc, struct isert_cmd *isert_cmd,
1518 		     struct ib_device *ib_dev, bool comp_err)
1519 {
1520 	if (isert_cmd->pdu_buf_dma != 0) {
1521 		isert_dbg("unmap single for isert_cmd->pdu_buf_dma\n");
1522 		ib_dma_unmap_single(ib_dev, isert_cmd->pdu_buf_dma,
1523 				    isert_cmd->pdu_buf_len, DMA_TO_DEVICE);
1524 		isert_cmd->pdu_buf_dma = 0;
1525 	}
1526 
1527 	isert_unmap_tx_desc(tx_desc, ib_dev);
1528 	isert_put_cmd(isert_cmd, comp_err);
1529 }
1530 
1531 static int
1532 isert_check_pi_status(struct se_cmd *se_cmd, struct ib_mr *sig_mr)
1533 {
1534 	struct ib_mr_status mr_status;
1535 	int ret;
1536 
1537 	ret = ib_check_mr_status(sig_mr, IB_MR_CHECK_SIG_STATUS, &mr_status);
1538 	if (ret) {
1539 		isert_err("ib_check_mr_status failed, ret %d\n", ret);
1540 		goto fail_mr_status;
1541 	}
1542 
1543 	if (mr_status.fail_status & IB_MR_CHECK_SIG_STATUS) {
1544 		u64 sec_offset_err;
1545 		u32 block_size = se_cmd->se_dev->dev_attrib.block_size + 8;
1546 
1547 		switch (mr_status.sig_err.err_type) {
1548 		case IB_SIG_BAD_GUARD:
1549 			se_cmd->pi_err = TCM_LOGICAL_BLOCK_GUARD_CHECK_FAILED;
1550 			break;
1551 		case IB_SIG_BAD_REFTAG:
1552 			se_cmd->pi_err = TCM_LOGICAL_BLOCK_REF_TAG_CHECK_FAILED;
1553 			break;
1554 		case IB_SIG_BAD_APPTAG:
1555 			se_cmd->pi_err = TCM_LOGICAL_BLOCK_APP_TAG_CHECK_FAILED;
1556 			break;
1557 		}
1558 		sec_offset_err = mr_status.sig_err.sig_err_offset;
1559 		do_div(sec_offset_err, block_size);
1560 		se_cmd->sense_info = sec_offset_err + se_cmd->t_task_lba;
1561 
1562 		isert_err("PI error found type %d at sector 0x%llx "
1563 			  "expected 0x%x vs actual 0x%x\n",
1564 			  mr_status.sig_err.err_type,
1565 			  (unsigned long long)se_cmd->sense_info,
1566 			  mr_status.sig_err.expected,
1567 			  mr_status.sig_err.actual);
1568 		ret = 1;
1569 	}
1570 
1571 fail_mr_status:
1572 	return ret;
1573 }
1574 
1575 static void
1576 isert_rdma_write_done(struct ib_cq *cq, struct ib_wc *wc)
1577 {
1578 	struct isert_conn *isert_conn = wc->qp->qp_context;
1579 	struct isert_device *device = isert_conn->device;
1580 	struct iser_tx_desc *desc = cqe_to_tx_desc(wc->wr_cqe);
1581 	struct isert_cmd *isert_cmd = tx_desc_to_cmd(desc);
1582 	struct se_cmd *cmd = &isert_cmd->iscsit_cmd->se_cmd;
1583 	int ret = 0;
1584 
1585 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1586 		isert_print_wc(wc, "rdma write");
1587 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1588 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1589 		isert_completion_put(desc, isert_cmd, device->ib_device, true);
1590 		return;
1591 	}
1592 
1593 	isert_dbg("Cmd %p\n", isert_cmd);
1594 
1595 	ret = isert_check_pi_status(cmd, isert_cmd->rw.reg->mr);
1596 	isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1597 
1598 	if (ret) {
1599 		/*
1600 		 * transport_generic_request_failure() expects to have
1601 		 * plus two references to handle queue-full, so re-add
1602 		 * one here as target-core will have already dropped
1603 		 * it after the first isert_put_datain() callback.
1604 		 */
1605 		kref_get(&cmd->cmd_kref);
1606 		transport_generic_request_failure(cmd, cmd->pi_err);
1607 	} else {
1608 		/*
1609 		 * XXX: isert_put_response() failure is not retried.
1610 		 */
1611 		ret = isert_put_response(isert_conn->conn, isert_cmd->iscsit_cmd);
1612 		if (ret)
1613 			pr_warn_ratelimited("isert_put_response() ret: %d\n", ret);
1614 	}
1615 }
1616 
1617 static void
1618 isert_rdma_read_done(struct ib_cq *cq, struct ib_wc *wc)
1619 {
1620 	struct isert_conn *isert_conn = wc->qp->qp_context;
1621 	struct isert_device *device = isert_conn->device;
1622 	struct iser_tx_desc *desc = cqe_to_tx_desc(wc->wr_cqe);
1623 	struct isert_cmd *isert_cmd = tx_desc_to_cmd(desc);
1624 	struct iscsit_cmd *cmd = isert_cmd->iscsit_cmd;
1625 	struct se_cmd *se_cmd = &cmd->se_cmd;
1626 	int ret = 0;
1627 
1628 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1629 		isert_print_wc(wc, "rdma read");
1630 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1631 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1632 		isert_completion_put(desc, isert_cmd, device->ib_device, true);
1633 		return;
1634 	}
1635 
1636 	isert_dbg("Cmd %p\n", isert_cmd);
1637 
1638 	iscsit_stop_dataout_timer(cmd);
1639 
1640 	if (isert_prot_cmd(isert_conn, se_cmd))
1641 		ret = isert_check_pi_status(se_cmd, isert_cmd->rw.reg->mr);
1642 	isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1643 	cmd->write_data_done = 0;
1644 
1645 	isert_dbg("Cmd: %p RDMA_READ comp calling execute_cmd\n", isert_cmd);
1646 	spin_lock_bh(&cmd->istate_lock);
1647 	cmd->cmd_flags |= ICF_GOT_LAST_DATAOUT;
1648 	cmd->i_state = ISTATE_RECEIVED_LAST_DATAOUT;
1649 	spin_unlock_bh(&cmd->istate_lock);
1650 
1651 	/*
1652 	 * transport_generic_request_failure() will drop the extra
1653 	 * se_cmd->cmd_kref reference after T10-PI error, and handle
1654 	 * any non-zero ->queue_status() callback error retries.
1655 	 */
1656 	if (ret)
1657 		transport_generic_request_failure(se_cmd, se_cmd->pi_err);
1658 	else
1659 		target_execute_cmd(se_cmd);
1660 }
1661 
1662 static void
1663 isert_do_control_comp(struct work_struct *work)
1664 {
1665 	struct isert_cmd *isert_cmd = container_of(work,
1666 			struct isert_cmd, comp_work);
1667 	struct isert_conn *isert_conn = isert_cmd->conn;
1668 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1669 	struct iscsit_cmd *cmd = isert_cmd->iscsit_cmd;
1670 
1671 	isert_dbg("Cmd %p i_state %d\n", isert_cmd, cmd->i_state);
1672 
1673 	switch (cmd->i_state) {
1674 	case ISTATE_SEND_TASKMGTRSP:
1675 		iscsit_tmr_post_handler(cmd, cmd->conn);
1676 		fallthrough;
1677 	case ISTATE_SEND_REJECT:
1678 	case ISTATE_SEND_TEXTRSP:
1679 		cmd->i_state = ISTATE_SENT_STATUS;
1680 		isert_completion_put(&isert_cmd->tx_desc, isert_cmd,
1681 				     ib_dev, false);
1682 		break;
1683 	case ISTATE_SEND_LOGOUTRSP:
1684 		iscsit_logout_post_handler(cmd, cmd->conn);
1685 		break;
1686 	default:
1687 		isert_err("Unknown i_state %d\n", cmd->i_state);
1688 		dump_stack();
1689 		break;
1690 	}
1691 }
1692 
1693 static void
1694 isert_login_send_done(struct ib_cq *cq, struct ib_wc *wc)
1695 {
1696 	struct isert_conn *isert_conn = wc->qp->qp_context;
1697 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1698 	struct iser_tx_desc *tx_desc = cqe_to_tx_desc(wc->wr_cqe);
1699 
1700 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1701 		isert_print_wc(wc, "login send");
1702 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1703 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1704 	}
1705 
1706 	isert_unmap_tx_desc(tx_desc, ib_dev);
1707 }
1708 
1709 static void
1710 isert_send_done(struct ib_cq *cq, struct ib_wc *wc)
1711 {
1712 	struct isert_conn *isert_conn = wc->qp->qp_context;
1713 	struct ib_device *ib_dev = isert_conn->cm_id->device;
1714 	struct iser_tx_desc *tx_desc = cqe_to_tx_desc(wc->wr_cqe);
1715 	struct isert_cmd *isert_cmd = tx_desc_to_cmd(tx_desc);
1716 
1717 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
1718 		isert_print_wc(wc, "send");
1719 		if (wc->status != IB_WC_WR_FLUSH_ERR)
1720 			iscsit_cause_connection_reinstatement(isert_conn->conn, 0);
1721 		isert_completion_put(tx_desc, isert_cmd, ib_dev, true);
1722 		return;
1723 	}
1724 
1725 	isert_dbg("Cmd %p\n", isert_cmd);
1726 
1727 	switch (isert_cmd->iscsit_cmd->i_state) {
1728 	case ISTATE_SEND_TASKMGTRSP:
1729 	case ISTATE_SEND_LOGOUTRSP:
1730 	case ISTATE_SEND_REJECT:
1731 	case ISTATE_SEND_TEXTRSP:
1732 		isert_unmap_tx_desc(tx_desc, ib_dev);
1733 
1734 		INIT_WORK(&isert_cmd->comp_work, isert_do_control_comp);
1735 		queue_work(isert_comp_wq, &isert_cmd->comp_work);
1736 		return;
1737 	default:
1738 		isert_cmd->iscsit_cmd->i_state = ISTATE_SENT_STATUS;
1739 		isert_completion_put(tx_desc, isert_cmd, ib_dev, false);
1740 		break;
1741 	}
1742 }
1743 
1744 static int
1745 isert_post_response(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd)
1746 {
1747 	int ret;
1748 
1749 	ret = isert_post_recv(isert_conn, isert_cmd->rx_desc);
1750 	if (ret)
1751 		return ret;
1752 
1753 	ret = ib_post_send(isert_conn->qp, &isert_cmd->tx_desc.send_wr, NULL);
1754 	if (ret) {
1755 		isert_err("ib_post_send failed with %d\n", ret);
1756 		return ret;
1757 	}
1758 	return ret;
1759 }
1760 
1761 static int
1762 isert_put_response(struct iscsit_conn *conn, struct iscsit_cmd *cmd)
1763 {
1764 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1765 	struct isert_conn *isert_conn = conn->context;
1766 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1767 	struct iscsi_scsi_rsp *hdr = (struct iscsi_scsi_rsp *)
1768 				&isert_cmd->tx_desc.iscsi_header;
1769 
1770 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1771 	iscsit_build_rsp_pdu(cmd, conn, true, hdr);
1772 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1773 	/*
1774 	 * Attach SENSE DATA payload to iSCSI Response PDU
1775 	 */
1776 	if (cmd->se_cmd.sense_buffer &&
1777 	    ((cmd->se_cmd.se_cmd_flags & SCF_TRANSPORT_TASK_SENSE) ||
1778 	    (cmd->se_cmd.se_cmd_flags & SCF_EMULATED_TASK_SENSE))) {
1779 		struct isert_device *device = isert_conn->device;
1780 		struct ib_device *ib_dev = device->ib_device;
1781 		struct ib_sge *tx_dsg = &isert_cmd->tx_desc.tx_sg[1];
1782 		u32 padding, pdu_len;
1783 
1784 		put_unaligned_be16(cmd->se_cmd.scsi_sense_length,
1785 				   cmd->sense_buffer);
1786 		cmd->se_cmd.scsi_sense_length += sizeof(__be16);
1787 
1788 		padding = -(cmd->se_cmd.scsi_sense_length) & 3;
1789 		hton24(hdr->dlength, (u32)cmd->se_cmd.scsi_sense_length);
1790 		pdu_len = cmd->se_cmd.scsi_sense_length + padding;
1791 
1792 		isert_cmd->pdu_buf_dma = ib_dma_map_single(ib_dev,
1793 				(void *)cmd->sense_buffer, pdu_len,
1794 				DMA_TO_DEVICE);
1795 		if (ib_dma_mapping_error(ib_dev, isert_cmd->pdu_buf_dma))
1796 			return -ENOMEM;
1797 
1798 		isert_cmd->pdu_buf_len = pdu_len;
1799 		tx_dsg->addr	= isert_cmd->pdu_buf_dma;
1800 		tx_dsg->length	= pdu_len;
1801 		tx_dsg->lkey	= device->pd->local_dma_lkey;
1802 		isert_cmd->tx_desc.num_sge = 2;
1803 	}
1804 
1805 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1806 
1807 	isert_dbg("Posting SCSI Response\n");
1808 
1809 	return isert_post_response(isert_conn, isert_cmd);
1810 }
1811 
1812 static void
1813 isert_aborted_task(struct iscsit_conn *conn, struct iscsit_cmd *cmd)
1814 {
1815 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1816 	struct isert_conn *isert_conn = conn->context;
1817 
1818 	spin_lock_bh(&conn->cmd_lock);
1819 	if (!list_empty(&cmd->i_conn_node))
1820 		list_del_init(&cmd->i_conn_node);
1821 	spin_unlock_bh(&conn->cmd_lock);
1822 
1823 	if (cmd->data_direction == DMA_TO_DEVICE)
1824 		iscsit_stop_dataout_timer(cmd);
1825 	isert_rdma_rw_ctx_destroy(isert_cmd, isert_conn);
1826 }
1827 
1828 static enum target_prot_op
1829 isert_get_sup_prot_ops(struct iscsit_conn *conn)
1830 {
1831 	struct isert_conn *isert_conn = conn->context;
1832 	struct isert_device *device = isert_conn->device;
1833 
1834 	if (conn->tpg->tpg_attrib.t10_pi) {
1835 		if (device->pi_capable) {
1836 			isert_info("conn %p PI offload enabled\n", isert_conn);
1837 			isert_conn->pi_support = true;
1838 			return TARGET_PROT_ALL;
1839 		}
1840 	}
1841 
1842 	isert_info("conn %p PI offload disabled\n", isert_conn);
1843 	isert_conn->pi_support = false;
1844 
1845 	return TARGET_PROT_NORMAL;
1846 }
1847 
1848 static int
1849 isert_put_nopin(struct iscsit_cmd *cmd, struct iscsit_conn *conn,
1850 		bool nopout_response)
1851 {
1852 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1853 	struct isert_conn *isert_conn = conn->context;
1854 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1855 
1856 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1857 	iscsit_build_nopin_rsp(cmd, conn, (struct iscsi_nopin *)
1858 			       &isert_cmd->tx_desc.iscsi_header,
1859 			       nopout_response);
1860 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1861 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1862 
1863 	isert_dbg("conn %p Posting NOPIN Response\n", isert_conn);
1864 
1865 	return isert_post_response(isert_conn, isert_cmd);
1866 }
1867 
1868 static int
1869 isert_put_logout_rsp(struct iscsit_cmd *cmd, struct iscsit_conn *conn)
1870 {
1871 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1872 	struct isert_conn *isert_conn = conn->context;
1873 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1874 
1875 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1876 	iscsit_build_logout_rsp(cmd, conn, (struct iscsi_logout_rsp *)
1877 				&isert_cmd->tx_desc.iscsi_header);
1878 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1879 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1880 
1881 	isert_dbg("conn %p Posting Logout Response\n", isert_conn);
1882 
1883 	return isert_post_response(isert_conn, isert_cmd);
1884 }
1885 
1886 static int
1887 isert_put_tm_rsp(struct iscsit_cmd *cmd, struct iscsit_conn *conn)
1888 {
1889 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1890 	struct isert_conn *isert_conn = conn->context;
1891 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1892 
1893 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1894 	iscsit_build_task_mgt_rsp(cmd, conn, (struct iscsi_tm_rsp *)
1895 				  &isert_cmd->tx_desc.iscsi_header);
1896 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1897 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1898 
1899 	isert_dbg("conn %p Posting Task Management Response\n", isert_conn);
1900 
1901 	return isert_post_response(isert_conn, isert_cmd);
1902 }
1903 
1904 static int
1905 isert_put_reject(struct iscsit_cmd *cmd, struct iscsit_conn *conn)
1906 {
1907 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1908 	struct isert_conn *isert_conn = conn->context;
1909 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1910 	struct isert_device *device = isert_conn->device;
1911 	struct ib_device *ib_dev = device->ib_device;
1912 	struct ib_sge *tx_dsg = &isert_cmd->tx_desc.tx_sg[1];
1913 	struct iscsi_reject *hdr =
1914 		(struct iscsi_reject *)&isert_cmd->tx_desc.iscsi_header;
1915 
1916 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1917 	iscsit_build_reject(cmd, conn, hdr);
1918 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1919 
1920 	hton24(hdr->dlength, ISCSI_HDR_LEN);
1921 	isert_cmd->pdu_buf_dma = ib_dma_map_single(ib_dev,
1922 			(void *)cmd->buf_ptr, ISCSI_HDR_LEN,
1923 			DMA_TO_DEVICE);
1924 	if (ib_dma_mapping_error(ib_dev, isert_cmd->pdu_buf_dma))
1925 		return -ENOMEM;
1926 	isert_cmd->pdu_buf_len = ISCSI_HDR_LEN;
1927 	tx_dsg->addr	= isert_cmd->pdu_buf_dma;
1928 	tx_dsg->length	= ISCSI_HDR_LEN;
1929 	tx_dsg->lkey	= device->pd->local_dma_lkey;
1930 	isert_cmd->tx_desc.num_sge = 2;
1931 
1932 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1933 
1934 	isert_dbg("conn %p Posting Reject\n", isert_conn);
1935 
1936 	return isert_post_response(isert_conn, isert_cmd);
1937 }
1938 
1939 static int
1940 isert_put_text_rsp(struct iscsit_cmd *cmd, struct iscsit_conn *conn)
1941 {
1942 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
1943 	struct isert_conn *isert_conn = conn->context;
1944 	struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr;
1945 	struct iscsi_text_rsp *hdr =
1946 		(struct iscsi_text_rsp *)&isert_cmd->tx_desc.iscsi_header;
1947 	u32 txt_rsp_len;
1948 	int rc;
1949 
1950 	isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc);
1951 	rc = iscsit_build_text_rsp(cmd, conn, hdr, ISCSI_INFINIBAND);
1952 	if (rc < 0)
1953 		return rc;
1954 
1955 	txt_rsp_len = rc;
1956 	isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
1957 
1958 	if (txt_rsp_len) {
1959 		struct isert_device *device = isert_conn->device;
1960 		struct ib_device *ib_dev = device->ib_device;
1961 		struct ib_sge *tx_dsg = &isert_cmd->tx_desc.tx_sg[1];
1962 		void *txt_rsp_buf = cmd->buf_ptr;
1963 
1964 		isert_cmd->pdu_buf_dma = ib_dma_map_single(ib_dev,
1965 				txt_rsp_buf, txt_rsp_len, DMA_TO_DEVICE);
1966 		if (ib_dma_mapping_error(ib_dev, isert_cmd->pdu_buf_dma))
1967 			return -ENOMEM;
1968 
1969 		isert_cmd->pdu_buf_len = txt_rsp_len;
1970 		tx_dsg->addr	= isert_cmd->pdu_buf_dma;
1971 		tx_dsg->length	= txt_rsp_len;
1972 		tx_dsg->lkey	= device->pd->local_dma_lkey;
1973 		isert_cmd->tx_desc.num_sge = 2;
1974 	}
1975 	isert_init_send_wr(isert_conn, isert_cmd, send_wr);
1976 
1977 	isert_dbg("conn %p Text Response\n", isert_conn);
1978 
1979 	return isert_post_response(isert_conn, isert_cmd);
1980 }
1981 
1982 static inline void
1983 isert_set_dif_domain(struct se_cmd *se_cmd, struct ib_sig_domain *domain)
1984 {
1985 	domain->sig_type = IB_SIG_TYPE_T10_DIF;
1986 	domain->sig.dif.bg_type = IB_T10DIF_CRC;
1987 	domain->sig.dif.pi_interval = se_cmd->se_dev->dev_attrib.block_size;
1988 	domain->sig.dif.ref_tag = se_cmd->reftag_seed;
1989 	/*
1990 	 * At the moment we hard code those, but if in the future
1991 	 * the target core would like to use it, we will take it
1992 	 * from se_cmd.
1993 	 */
1994 	domain->sig.dif.apptag_check_mask = 0xffff;
1995 	domain->sig.dif.app_escape = true;
1996 	domain->sig.dif.ref_escape = true;
1997 	if (se_cmd->prot_type == TARGET_DIF_TYPE1_PROT ||
1998 	    se_cmd->prot_type == TARGET_DIF_TYPE2_PROT)
1999 		domain->sig.dif.ref_remap = true;
2000 }
2001 
2002 static int
2003 isert_set_sig_attrs(struct se_cmd *se_cmd, struct ib_sig_attrs *sig_attrs)
2004 {
2005 	memset(sig_attrs, 0, sizeof(*sig_attrs));
2006 
2007 	switch (se_cmd->prot_op) {
2008 	case TARGET_PROT_DIN_INSERT:
2009 	case TARGET_PROT_DOUT_STRIP:
2010 		sig_attrs->mem.sig_type = IB_SIG_TYPE_NONE;
2011 		isert_set_dif_domain(se_cmd, &sig_attrs->wire);
2012 		break;
2013 	case TARGET_PROT_DOUT_INSERT:
2014 	case TARGET_PROT_DIN_STRIP:
2015 		sig_attrs->wire.sig_type = IB_SIG_TYPE_NONE;
2016 		isert_set_dif_domain(se_cmd, &sig_attrs->mem);
2017 		break;
2018 	case TARGET_PROT_DIN_PASS:
2019 	case TARGET_PROT_DOUT_PASS:
2020 		isert_set_dif_domain(se_cmd, &sig_attrs->wire);
2021 		isert_set_dif_domain(se_cmd, &sig_attrs->mem);
2022 		break;
2023 	default:
2024 		isert_err("Unsupported PI operation %d\n", se_cmd->prot_op);
2025 		return -EINVAL;
2026 	}
2027 
2028 	if (se_cmd->prot_checks & TARGET_DIF_CHECK_GUARD)
2029 		sig_attrs->check_mask |= IB_SIG_CHECK_GUARD;
2030 	if (se_cmd->prot_checks & TARGET_DIF_CHECK_APPTAG)
2031 		sig_attrs->check_mask |= IB_SIG_CHECK_APPTAG;
2032 	if (se_cmd->prot_checks & TARGET_DIF_CHECK_REFTAG)
2033 		sig_attrs->check_mask |= IB_SIG_CHECK_REFTAG;
2034 
2035 	return 0;
2036 }
2037 
2038 static int
2039 isert_rdma_rw_ctx_post(struct isert_cmd *cmd, struct isert_conn *conn,
2040 		struct ib_cqe *cqe, struct ib_send_wr *chain_wr)
2041 {
2042 	struct se_cmd *se_cmd = &cmd->iscsit_cmd->se_cmd;
2043 	enum dma_data_direction dir = target_reverse_dma_direction(se_cmd);
2044 	u8 port_num = conn->cm_id->port_num;
2045 	u64 addr;
2046 	u32 rkey, offset;
2047 	int ret;
2048 
2049 	if (cmd->ctx_init_done)
2050 		goto rdma_ctx_post;
2051 
2052 	if (dir == DMA_FROM_DEVICE) {
2053 		addr = cmd->write_va;
2054 		rkey = cmd->write_stag;
2055 		offset = cmd->iscsit_cmd->write_data_done;
2056 	} else {
2057 		addr = cmd->read_va;
2058 		rkey = cmd->read_stag;
2059 		offset = 0;
2060 	}
2061 
2062 	if (isert_prot_cmd(conn, se_cmd)) {
2063 		struct ib_sig_attrs sig_attrs;
2064 
2065 		ret = isert_set_sig_attrs(se_cmd, &sig_attrs);
2066 		if (ret)
2067 			return ret;
2068 
2069 		WARN_ON_ONCE(offset);
2070 		ret = rdma_rw_ctx_signature_init(&cmd->rw, conn->qp, port_num,
2071 				se_cmd->t_data_sg, se_cmd->t_data_nents,
2072 				se_cmd->t_prot_sg, se_cmd->t_prot_nents,
2073 				&sig_attrs, addr, rkey, dir);
2074 	} else {
2075 		ret = rdma_rw_ctx_init(&cmd->rw, conn->qp, port_num,
2076 				se_cmd->t_data_sg, se_cmd->t_data_nents,
2077 				offset, addr, rkey, dir);
2078 	}
2079 
2080 	if (ret < 0) {
2081 		isert_err("Cmd: %p failed to prepare RDMA res\n", cmd);
2082 		return ret;
2083 	}
2084 
2085 	cmd->ctx_init_done = true;
2086 
2087 rdma_ctx_post:
2088 	ret = rdma_rw_ctx_post(&cmd->rw, conn->qp, port_num, cqe, chain_wr);
2089 	if (ret < 0)
2090 		isert_err("Cmd: %p failed to post RDMA res\n", cmd);
2091 	return ret;
2092 }
2093 
2094 static int
2095 isert_put_datain(struct iscsit_conn *conn, struct iscsit_cmd *cmd)
2096 {
2097 	struct se_cmd *se_cmd = &cmd->se_cmd;
2098 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2099 	struct isert_conn *isert_conn = conn->context;
2100 	struct ib_cqe *cqe = NULL;
2101 	struct ib_send_wr *chain_wr = NULL;
2102 	int rc;
2103 
2104 	isert_dbg("Cmd: %p RDMA_WRITE data_length: %u\n",
2105 		 isert_cmd, se_cmd->data_length);
2106 
2107 	if (isert_prot_cmd(isert_conn, se_cmd)) {
2108 		isert_cmd->tx_desc.tx_cqe.done = isert_rdma_write_done;
2109 		cqe = &isert_cmd->tx_desc.tx_cqe;
2110 	} else {
2111 		/*
2112 		 * Build isert_conn->tx_desc for iSCSI response PDU and attach
2113 		 */
2114 		isert_create_send_desc(isert_conn, isert_cmd,
2115 				       &isert_cmd->tx_desc);
2116 		iscsit_build_rsp_pdu(cmd, conn, true, (struct iscsi_scsi_rsp *)
2117 				     &isert_cmd->tx_desc.iscsi_header);
2118 		isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
2119 		isert_init_send_wr(isert_conn, isert_cmd,
2120 				   &isert_cmd->tx_desc.send_wr);
2121 
2122 		rc = isert_post_recv(isert_conn, isert_cmd->rx_desc);
2123 		if (rc)
2124 			return rc;
2125 
2126 		chain_wr = &isert_cmd->tx_desc.send_wr;
2127 	}
2128 
2129 	rc = isert_rdma_rw_ctx_post(isert_cmd, isert_conn, cqe, chain_wr);
2130 	isert_dbg("Cmd: %p posted RDMA_WRITE for iSER Data READ rc: %d\n",
2131 		  isert_cmd, rc);
2132 	return rc;
2133 }
2134 
2135 static int
2136 isert_get_dataout(struct iscsit_conn *conn, struct iscsit_cmd *cmd, bool recovery)
2137 {
2138 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2139 	int ret;
2140 
2141 	isert_dbg("Cmd: %p RDMA_READ data_length: %u write_data_done: %u\n",
2142 		 isert_cmd, cmd->se_cmd.data_length, cmd->write_data_done);
2143 
2144 	isert_cmd->tx_desc.tx_cqe.done = isert_rdma_read_done;
2145 	ret = isert_rdma_rw_ctx_post(isert_cmd, conn->context,
2146 				     &isert_cmd->tx_desc.tx_cqe, NULL);
2147 
2148 	isert_dbg("Cmd: %p posted RDMA_READ memory for ISER Data WRITE rc: %d\n",
2149 		 isert_cmd, ret);
2150 	return ret;
2151 }
2152 
2153 static int
2154 isert_immediate_queue(struct iscsit_conn *conn, struct iscsit_cmd *cmd, int state)
2155 {
2156 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2157 	int ret = 0;
2158 
2159 	switch (state) {
2160 	case ISTATE_REMOVE:
2161 		spin_lock_bh(&conn->cmd_lock);
2162 		list_del_init(&cmd->i_conn_node);
2163 		spin_unlock_bh(&conn->cmd_lock);
2164 		isert_put_cmd(isert_cmd, true);
2165 		break;
2166 	case ISTATE_SEND_NOPIN_WANT_RESPONSE:
2167 		ret = isert_put_nopin(cmd, conn, false);
2168 		break;
2169 	default:
2170 		isert_err("Unknown immediate state: 0x%02x\n", state);
2171 		ret = -EINVAL;
2172 		break;
2173 	}
2174 
2175 	return ret;
2176 }
2177 
2178 static int
2179 isert_response_queue(struct iscsit_conn *conn, struct iscsit_cmd *cmd, int state)
2180 {
2181 	struct isert_conn *isert_conn = conn->context;
2182 	int ret;
2183 
2184 	switch (state) {
2185 	case ISTATE_SEND_LOGOUTRSP:
2186 		ret = isert_put_logout_rsp(cmd, conn);
2187 		if (!ret)
2188 			isert_conn->logout_posted = true;
2189 		break;
2190 	case ISTATE_SEND_NOPIN:
2191 		ret = isert_put_nopin(cmd, conn, true);
2192 		break;
2193 	case ISTATE_SEND_TASKMGTRSP:
2194 		ret = isert_put_tm_rsp(cmd, conn);
2195 		break;
2196 	case ISTATE_SEND_REJECT:
2197 		ret = isert_put_reject(cmd, conn);
2198 		break;
2199 	case ISTATE_SEND_TEXTRSP:
2200 		ret = isert_put_text_rsp(cmd, conn);
2201 		break;
2202 	case ISTATE_SEND_STATUS:
2203 		/*
2204 		 * Special case for sending non GOOD SCSI status from TX thread
2205 		 * context during pre se_cmd excecution failure.
2206 		 */
2207 		ret = isert_put_response(conn, cmd);
2208 		break;
2209 	default:
2210 		isert_err("Unknown response state: 0x%02x\n", state);
2211 		ret = -EINVAL;
2212 		break;
2213 	}
2214 
2215 	return ret;
2216 }
2217 
2218 struct rdma_cm_id *
2219 isert_setup_id(struct isert_np *isert_np)
2220 {
2221 	struct iscsi_np *np = isert_np->np;
2222 	struct rdma_cm_id *id;
2223 	struct sockaddr *sa;
2224 	int ret;
2225 
2226 	sa = (struct sockaddr *)&np->np_sockaddr;
2227 	isert_dbg("ksockaddr: %p, sa: %p\n", &np->np_sockaddr, sa);
2228 
2229 	id = rdma_create_id(&init_net, isert_cma_handler, isert_np,
2230 			    RDMA_PS_TCP, IB_QPT_RC);
2231 	if (IS_ERR(id)) {
2232 		isert_err("rdma_create_id() failed: %ld\n", PTR_ERR(id));
2233 		ret = PTR_ERR(id);
2234 		goto out;
2235 	}
2236 	isert_dbg("id %p context %p\n", id, id->context);
2237 
2238 	/*
2239 	 * Allow both IPv4 and IPv6 sockets to bind a single port
2240 	 * at the same time.
2241 	 */
2242 	ret = rdma_set_afonly(id, 1);
2243 	if (ret) {
2244 		isert_err("rdma_set_afonly() failed: %d\n", ret);
2245 		goto out_id;
2246 	}
2247 
2248 	ret = rdma_bind_addr(id, sa);
2249 	if (ret) {
2250 		isert_err("rdma_bind_addr() failed: %d\n", ret);
2251 		goto out_id;
2252 	}
2253 
2254 	ret = rdma_listen(id, 0);
2255 	if (ret) {
2256 		isert_err("rdma_listen() failed: %d\n", ret);
2257 		goto out_id;
2258 	}
2259 
2260 	return id;
2261 out_id:
2262 	rdma_destroy_id(id);
2263 out:
2264 	return ERR_PTR(ret);
2265 }
2266 
2267 static int
2268 isert_setup_np(struct iscsi_np *np,
2269 	       struct sockaddr_storage *ksockaddr)
2270 {
2271 	struct isert_np *isert_np;
2272 	struct rdma_cm_id *isert_lid;
2273 	int ret;
2274 
2275 	isert_np = kzalloc(sizeof(struct isert_np), GFP_KERNEL);
2276 	if (!isert_np)
2277 		return -ENOMEM;
2278 
2279 	sema_init(&isert_np->sem, 0);
2280 	mutex_init(&isert_np->mutex);
2281 	INIT_LIST_HEAD(&isert_np->accepted);
2282 	INIT_LIST_HEAD(&isert_np->pending);
2283 	isert_np->np = np;
2284 
2285 	/*
2286 	 * Setup the np->np_sockaddr from the passed sockaddr setup
2287 	 * in iscsi_target_configfs.c code..
2288 	 */
2289 	memcpy(&np->np_sockaddr, ksockaddr,
2290 	       sizeof(struct sockaddr_storage));
2291 
2292 	isert_lid = isert_setup_id(isert_np);
2293 	if (IS_ERR(isert_lid)) {
2294 		ret = PTR_ERR(isert_lid);
2295 		goto out;
2296 	}
2297 
2298 	isert_np->cm_id = isert_lid;
2299 	np->np_context = isert_np;
2300 
2301 	return 0;
2302 
2303 out:
2304 	kfree(isert_np);
2305 
2306 	return ret;
2307 }
2308 
2309 static int
2310 isert_rdma_accept(struct isert_conn *isert_conn)
2311 {
2312 	struct rdma_cm_id *cm_id = isert_conn->cm_id;
2313 	struct rdma_conn_param cp;
2314 	int ret;
2315 	struct iser_cm_hdr rsp_hdr;
2316 
2317 	memset(&cp, 0, sizeof(struct rdma_conn_param));
2318 	cp.initiator_depth = isert_conn->initiator_depth;
2319 	cp.retry_count = 7;
2320 	cp.rnr_retry_count = 7;
2321 
2322 	memset(&rsp_hdr, 0, sizeof(rsp_hdr));
2323 	rsp_hdr.flags = ISERT_ZBVA_NOT_USED;
2324 	if (!isert_conn->snd_w_inv)
2325 		rsp_hdr.flags = rsp_hdr.flags | ISERT_SEND_W_INV_NOT_USED;
2326 	cp.private_data = (void *)&rsp_hdr;
2327 	cp.private_data_len = sizeof(rsp_hdr);
2328 
2329 	ret = rdma_accept(cm_id, &cp);
2330 	if (ret) {
2331 		isert_err("rdma_accept() failed with: %d\n", ret);
2332 		return ret;
2333 	}
2334 
2335 	return 0;
2336 }
2337 
2338 static int
2339 isert_get_login_rx(struct iscsit_conn *conn, struct iscsi_login *login)
2340 {
2341 	struct isert_conn *isert_conn = conn->context;
2342 	int ret;
2343 
2344 	isert_info("before login_req comp conn: %p\n", isert_conn);
2345 	ret = wait_for_completion_interruptible(&isert_conn->login_req_comp);
2346 	if (ret) {
2347 		isert_err("isert_conn %p interrupted before got login req\n",
2348 			  isert_conn);
2349 		return ret;
2350 	}
2351 	reinit_completion(&isert_conn->login_req_comp);
2352 
2353 	/*
2354 	 * For login requests after the first PDU, isert_rx_login_req() will
2355 	 * kick queue_delayed_work(isert_login_wq, &conn->login_work) as
2356 	 * the packet is received, which turns this callback from
2357 	 * iscsi_target_do_login_rx() into a NOP.
2358 	 */
2359 	if (!login->first_request)
2360 		return 0;
2361 
2362 	isert_rx_login_req(isert_conn);
2363 
2364 	isert_info("before login_comp conn: %p\n", conn);
2365 	ret = wait_for_completion_interruptible(&isert_conn->login_comp);
2366 	if (ret)
2367 		return ret;
2368 
2369 	isert_info("processing login->req: %p\n", login->req);
2370 
2371 	return 0;
2372 }
2373 
2374 static void
2375 isert_set_conn_info(struct iscsi_np *np, struct iscsit_conn *conn,
2376 		    struct isert_conn *isert_conn)
2377 {
2378 	struct rdma_cm_id *cm_id = isert_conn->cm_id;
2379 	struct rdma_route *cm_route = &cm_id->route;
2380 
2381 	conn->login_family = np->np_sockaddr.ss_family;
2382 
2383 	conn->login_sockaddr = cm_route->addr.dst_addr;
2384 	conn->local_sockaddr = cm_route->addr.src_addr;
2385 }
2386 
2387 static int
2388 isert_accept_np(struct iscsi_np *np, struct iscsit_conn *conn)
2389 {
2390 	struct isert_np *isert_np = np->np_context;
2391 	struct isert_conn *isert_conn;
2392 	int ret;
2393 
2394 accept_wait:
2395 	ret = down_interruptible(&isert_np->sem);
2396 	if (ret)
2397 		return -ENODEV;
2398 
2399 	spin_lock_bh(&np->np_thread_lock);
2400 	if (np->np_thread_state >= ISCSI_NP_THREAD_RESET) {
2401 		spin_unlock_bh(&np->np_thread_lock);
2402 		isert_dbg("np_thread_state %d\n",
2403 			 np->np_thread_state);
2404 		/*
2405 		 * No point in stalling here when np_thread
2406 		 * is in state RESET/SHUTDOWN/EXIT - bail
2407 		 */
2408 		return -ENODEV;
2409 	}
2410 	spin_unlock_bh(&np->np_thread_lock);
2411 
2412 	mutex_lock(&isert_np->mutex);
2413 	if (list_empty(&isert_np->pending)) {
2414 		mutex_unlock(&isert_np->mutex);
2415 		goto accept_wait;
2416 	}
2417 	isert_conn = list_first_entry(&isert_np->pending,
2418 			struct isert_conn, node);
2419 	list_del_init(&isert_conn->node);
2420 	mutex_unlock(&isert_np->mutex);
2421 
2422 	conn->context = isert_conn;
2423 	isert_conn->conn = conn;
2424 	isert_conn->state = ISER_CONN_BOUND;
2425 
2426 	isert_set_conn_info(np, conn, isert_conn);
2427 
2428 	isert_dbg("Processing isert_conn: %p\n", isert_conn);
2429 
2430 	return 0;
2431 }
2432 
2433 static void
2434 isert_free_np(struct iscsi_np *np)
2435 {
2436 	struct isert_np *isert_np = np->np_context;
2437 	struct isert_conn *isert_conn, *n;
2438 	LIST_HEAD(drop_conn_list);
2439 
2440 	if (isert_np->cm_id)
2441 		rdma_destroy_id(isert_np->cm_id);
2442 
2443 	/*
2444 	 * FIXME: At this point we don't have a good way to insure
2445 	 * that at this point we don't have hanging connections that
2446 	 * completed RDMA establishment but didn't start iscsi login
2447 	 * process. So work-around this by cleaning up what ever piled
2448 	 * up in accepted and pending lists.
2449 	 */
2450 	mutex_lock(&isert_np->mutex);
2451 	if (!list_empty(&isert_np->pending)) {
2452 		isert_info("Still have isert pending connections\n");
2453 		list_for_each_entry_safe(isert_conn, n,
2454 					 &isert_np->pending,
2455 					 node) {
2456 			isert_info("cleaning isert_conn %p state (%d)\n",
2457 				   isert_conn, isert_conn->state);
2458 			list_move_tail(&isert_conn->node, &drop_conn_list);
2459 		}
2460 	}
2461 
2462 	if (!list_empty(&isert_np->accepted)) {
2463 		isert_info("Still have isert accepted connections\n");
2464 		list_for_each_entry_safe(isert_conn, n,
2465 					 &isert_np->accepted,
2466 					 node) {
2467 			isert_info("cleaning isert_conn %p state (%d)\n",
2468 				   isert_conn, isert_conn->state);
2469 			list_move_tail(&isert_conn->node, &drop_conn_list);
2470 		}
2471 	}
2472 	mutex_unlock(&isert_np->mutex);
2473 
2474 	list_for_each_entry_safe(isert_conn, n, &drop_conn_list, node) {
2475 		list_del_init(&isert_conn->node);
2476 		isert_connect_release(isert_conn);
2477 	}
2478 
2479 	np->np_context = NULL;
2480 	kfree(isert_np);
2481 }
2482 
2483 static void isert_release_work(struct work_struct *work)
2484 {
2485 	struct isert_conn *isert_conn = container_of(work,
2486 						     struct isert_conn,
2487 						     release_work);
2488 
2489 	isert_info("Starting release conn %p\n", isert_conn);
2490 
2491 	mutex_lock(&isert_conn->mutex);
2492 	isert_conn->state = ISER_CONN_DOWN;
2493 	mutex_unlock(&isert_conn->mutex);
2494 
2495 	isert_info("Destroying conn %p\n", isert_conn);
2496 	isert_put_conn(isert_conn);
2497 }
2498 
2499 static void
2500 isert_wait4logout(struct isert_conn *isert_conn)
2501 {
2502 	struct iscsit_conn *conn = isert_conn->conn;
2503 
2504 	isert_info("conn %p\n", isert_conn);
2505 
2506 	if (isert_conn->logout_posted) {
2507 		isert_info("conn %p wait for conn_logout_comp\n", isert_conn);
2508 		wait_for_completion_timeout(&conn->conn_logout_comp,
2509 					    SECONDS_FOR_LOGOUT_COMP * HZ);
2510 	}
2511 }
2512 
2513 static void
2514 isert_wait4cmds(struct iscsit_conn *conn)
2515 {
2516 	isert_info("iscsit_conn %p\n", conn);
2517 
2518 	if (conn->sess) {
2519 		target_stop_cmd_counter(conn->cmd_cnt);
2520 		target_wait_for_cmds(conn->cmd_cnt);
2521 	}
2522 }
2523 
2524 /**
2525  * isert_put_unsol_pending_cmds() - Drop commands waiting for
2526  *     unsolicitate dataout
2527  * @conn:    iscsi connection
2528  *
2529  * We might still have commands that are waiting for unsolicited
2530  * dataouts messages. We must put the extra reference on those
2531  * before blocking on the target_wait_for_session_cmds
2532  */
2533 static void
2534 isert_put_unsol_pending_cmds(struct iscsit_conn *conn)
2535 {
2536 	struct iscsit_cmd *cmd, *tmp;
2537 	static LIST_HEAD(drop_cmd_list);
2538 
2539 	spin_lock_bh(&conn->cmd_lock);
2540 	list_for_each_entry_safe(cmd, tmp, &conn->conn_cmd_list, i_conn_node) {
2541 		if ((cmd->cmd_flags & ICF_NON_IMMEDIATE_UNSOLICITED_DATA) &&
2542 		    (cmd->write_data_done < conn->sess->sess_ops->FirstBurstLength) &&
2543 		    (cmd->write_data_done < cmd->se_cmd.data_length))
2544 			list_move_tail(&cmd->i_conn_node, &drop_cmd_list);
2545 	}
2546 	spin_unlock_bh(&conn->cmd_lock);
2547 
2548 	list_for_each_entry_safe(cmd, tmp, &drop_cmd_list, i_conn_node) {
2549 		list_del_init(&cmd->i_conn_node);
2550 		if (cmd->i_state != ISTATE_REMOVE) {
2551 			struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
2552 
2553 			isert_info("conn %p dropping cmd %p\n", conn, cmd);
2554 			isert_put_cmd(isert_cmd, true);
2555 		}
2556 	}
2557 }
2558 
2559 static void isert_wait_conn(struct iscsit_conn *conn)
2560 {
2561 	struct isert_conn *isert_conn = conn->context;
2562 
2563 	isert_info("Starting conn %p\n", isert_conn);
2564 
2565 	mutex_lock(&isert_conn->mutex);
2566 	isert_conn_terminate(isert_conn);
2567 	mutex_unlock(&isert_conn->mutex);
2568 
2569 	ib_drain_qp(isert_conn->qp);
2570 	isert_put_unsol_pending_cmds(conn);
2571 	isert_wait4cmds(conn);
2572 	isert_wait4logout(isert_conn);
2573 
2574 	queue_work(isert_release_wq, &isert_conn->release_work);
2575 }
2576 
2577 static void isert_free_conn(struct iscsit_conn *conn)
2578 {
2579 	struct isert_conn *isert_conn = conn->context;
2580 
2581 	ib_drain_qp(isert_conn->qp);
2582 	isert_put_conn(isert_conn);
2583 }
2584 
2585 static void isert_get_rx_pdu(struct iscsit_conn *conn)
2586 {
2587 	struct completion comp;
2588 
2589 	init_completion(&comp);
2590 
2591 	wait_for_completion_interruptible(&comp);
2592 }
2593 
2594 static struct iscsit_transport iser_target_transport = {
2595 	.name			= "IB/iSER",
2596 	.transport_type		= ISCSI_INFINIBAND,
2597 	.rdma_shutdown		= true,
2598 	.priv_size		= sizeof(struct isert_cmd),
2599 	.owner			= THIS_MODULE,
2600 	.iscsit_setup_np	= isert_setup_np,
2601 	.iscsit_accept_np	= isert_accept_np,
2602 	.iscsit_free_np		= isert_free_np,
2603 	.iscsit_wait_conn	= isert_wait_conn,
2604 	.iscsit_free_conn	= isert_free_conn,
2605 	.iscsit_get_login_rx	= isert_get_login_rx,
2606 	.iscsit_put_login_tx	= isert_put_login_tx,
2607 	.iscsit_immediate_queue	= isert_immediate_queue,
2608 	.iscsit_response_queue	= isert_response_queue,
2609 	.iscsit_get_dataout	= isert_get_dataout,
2610 	.iscsit_queue_data_in	= isert_put_datain,
2611 	.iscsit_queue_status	= isert_put_response,
2612 	.iscsit_aborted_task	= isert_aborted_task,
2613 	.iscsit_get_rx_pdu	= isert_get_rx_pdu,
2614 	.iscsit_get_sup_prot_ops = isert_get_sup_prot_ops,
2615 };
2616 
2617 static int __init isert_init(void)
2618 {
2619 	isert_login_wq = alloc_workqueue("isert_login_wq", 0, 0);
2620 	if (!isert_login_wq) {
2621 		isert_err("Unable to allocate isert_login_wq\n");
2622 		return -ENOMEM;
2623 	}
2624 
2625 	isert_comp_wq = alloc_workqueue("isert_comp_wq",
2626 					WQ_UNBOUND | WQ_HIGHPRI, 0);
2627 	if (!isert_comp_wq) {
2628 		isert_err("Unable to allocate isert_comp_wq\n");
2629 		goto destroy_login_wq;
2630 	}
2631 
2632 	isert_release_wq = alloc_workqueue("isert_release_wq", WQ_UNBOUND,
2633 					WQ_UNBOUND_MAX_ACTIVE);
2634 	if (!isert_release_wq) {
2635 		isert_err("Unable to allocate isert_release_wq\n");
2636 		goto destroy_comp_wq;
2637 	}
2638 
2639 	iscsit_register_transport(&iser_target_transport);
2640 	isert_info("iSER_TARGET[0] - Loaded iser_target_transport\n");
2641 
2642 	return 0;
2643 
2644 destroy_comp_wq:
2645 	destroy_workqueue(isert_comp_wq);
2646 destroy_login_wq:
2647 	destroy_workqueue(isert_login_wq);
2648 
2649 	return -ENOMEM;
2650 }
2651 
2652 static void __exit isert_exit(void)
2653 {
2654 	flush_workqueue(isert_login_wq);
2655 	destroy_workqueue(isert_release_wq);
2656 	destroy_workqueue(isert_comp_wq);
2657 	iscsit_unregister_transport(&iser_target_transport);
2658 	isert_info("iSER_TARGET[0] - Released iser_target_transport\n");
2659 	destroy_workqueue(isert_login_wq);
2660 }
2661 
2662 MODULE_DESCRIPTION("iSER-Target for mainline target infrastructure");
2663 MODULE_AUTHOR("nab@Linux-iSCSI.org");
2664 MODULE_LICENSE("GPL");
2665 
2666 module_init(isert_init);
2667 module_exit(isert_exit);
2668