1 /*
2  * Copyright (c) 2006 - 2009 Mellanox Technology Inc.  All rights reserved.
3  * Copyright (C) 2008 - 2011 Bart Van Assche <bvanassche@acm.org>.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  *
33  */
34 
35 #include <linux/module.h>
36 #include <linux/init.h>
37 #include <linux/slab.h>
38 #include <linux/err.h>
39 #include <linux/ctype.h>
40 #include <linux/kthread.h>
41 #include <linux/string.h>
42 #include <linux/delay.h>
43 #include <linux/atomic.h>
44 #include <scsi/scsi_proto.h>
45 #include <scsi/scsi_tcq.h>
46 #include <target/target_core_base.h>
47 #include <target/target_core_fabric.h>
48 #include "ib_srpt.h"
49 
50 /* Name of this kernel module. */
51 #define DRV_NAME		"ib_srpt"
52 #define DRV_VERSION		"2.0.0"
53 #define DRV_RELDATE		"2011-02-14"
54 
55 #define SRPT_ID_STRING	"Linux SRP target"
56 
57 #undef pr_fmt
58 #define pr_fmt(fmt) DRV_NAME " " fmt
59 
60 MODULE_AUTHOR("Vu Pham and Bart Van Assche");
61 MODULE_DESCRIPTION("InfiniBand SCSI RDMA Protocol target "
62 		   "v" DRV_VERSION " (" DRV_RELDATE ")");
63 MODULE_LICENSE("Dual BSD/GPL");
64 
65 /*
66  * Global Variables
67  */
68 
69 static u64 srpt_service_guid;
70 static DEFINE_SPINLOCK(srpt_dev_lock);	/* Protects srpt_dev_list. */
71 static LIST_HEAD(srpt_dev_list);	/* List of srpt_device structures. */
72 
73 static unsigned srp_max_req_size = DEFAULT_MAX_REQ_SIZE;
74 module_param(srp_max_req_size, int, 0444);
75 MODULE_PARM_DESC(srp_max_req_size,
76 		 "Maximum size of SRP request messages in bytes.");
77 
78 static int srpt_srq_size = DEFAULT_SRPT_SRQ_SIZE;
79 module_param(srpt_srq_size, int, 0444);
80 MODULE_PARM_DESC(srpt_srq_size,
81 		 "Shared receive queue (SRQ) size.");
82 
83 static int srpt_get_u64_x(char *buffer, struct kernel_param *kp)
84 {
85 	return sprintf(buffer, "0x%016llx", *(u64 *)kp->arg);
86 }
87 module_param_call(srpt_service_guid, NULL, srpt_get_u64_x, &srpt_service_guid,
88 		  0444);
89 MODULE_PARM_DESC(srpt_service_guid,
90 		 "Using this value for ioc_guid, id_ext, and cm_listen_id"
91 		 " instead of using the node_guid of the first HCA.");
92 
93 static struct ib_client srpt_client;
94 static void srpt_release_channel(struct srpt_rdma_ch *ch);
95 static int srpt_queue_status(struct se_cmd *cmd);
96 
97 /**
98  * opposite_dma_dir() - Swap DMA_TO_DEVICE and DMA_FROM_DEVICE.
99  */
100 static inline
101 enum dma_data_direction opposite_dma_dir(enum dma_data_direction dir)
102 {
103 	switch (dir) {
104 	case DMA_TO_DEVICE:	return DMA_FROM_DEVICE;
105 	case DMA_FROM_DEVICE:	return DMA_TO_DEVICE;
106 	default:		return dir;
107 	}
108 }
109 
110 /**
111  * srpt_sdev_name() - Return the name associated with the HCA.
112  *
113  * Examples are ib0, ib1, ...
114  */
115 static inline const char *srpt_sdev_name(struct srpt_device *sdev)
116 {
117 	return sdev->device->name;
118 }
119 
120 static enum rdma_ch_state srpt_get_ch_state(struct srpt_rdma_ch *ch)
121 {
122 	unsigned long flags;
123 	enum rdma_ch_state state;
124 
125 	spin_lock_irqsave(&ch->spinlock, flags);
126 	state = ch->state;
127 	spin_unlock_irqrestore(&ch->spinlock, flags);
128 	return state;
129 }
130 
131 static enum rdma_ch_state
132 srpt_set_ch_state(struct srpt_rdma_ch *ch, enum rdma_ch_state new_state)
133 {
134 	unsigned long flags;
135 	enum rdma_ch_state prev;
136 
137 	spin_lock_irqsave(&ch->spinlock, flags);
138 	prev = ch->state;
139 	ch->state = new_state;
140 	spin_unlock_irqrestore(&ch->spinlock, flags);
141 	return prev;
142 }
143 
144 /**
145  * srpt_test_and_set_ch_state() - Test and set the channel state.
146  *
147  * Returns true if and only if the channel state has been set to the new state.
148  */
149 static bool
150 srpt_test_and_set_ch_state(struct srpt_rdma_ch *ch, enum rdma_ch_state old,
151 			   enum rdma_ch_state new)
152 {
153 	unsigned long flags;
154 	enum rdma_ch_state prev;
155 
156 	spin_lock_irqsave(&ch->spinlock, flags);
157 	prev = ch->state;
158 	if (prev == old)
159 		ch->state = new;
160 	spin_unlock_irqrestore(&ch->spinlock, flags);
161 	return prev == old;
162 }
163 
164 /**
165  * srpt_event_handler() - Asynchronous IB event callback function.
166  *
167  * Callback function called by the InfiniBand core when an asynchronous IB
168  * event occurs. This callback may occur in interrupt context. See also
169  * section 11.5.2, Set Asynchronous Event Handler in the InfiniBand
170  * Architecture Specification.
171  */
172 static void srpt_event_handler(struct ib_event_handler *handler,
173 			       struct ib_event *event)
174 {
175 	struct srpt_device *sdev;
176 	struct srpt_port *sport;
177 
178 	sdev = ib_get_client_data(event->device, &srpt_client);
179 	if (!sdev || sdev->device != event->device)
180 		return;
181 
182 	pr_debug("ASYNC event= %d on device= %s\n", event->event,
183 		 srpt_sdev_name(sdev));
184 
185 	switch (event->event) {
186 	case IB_EVENT_PORT_ERR:
187 		if (event->element.port_num <= sdev->device->phys_port_cnt) {
188 			sport = &sdev->port[event->element.port_num - 1];
189 			sport->lid = 0;
190 			sport->sm_lid = 0;
191 		}
192 		break;
193 	case IB_EVENT_PORT_ACTIVE:
194 	case IB_EVENT_LID_CHANGE:
195 	case IB_EVENT_PKEY_CHANGE:
196 	case IB_EVENT_SM_CHANGE:
197 	case IB_EVENT_CLIENT_REREGISTER:
198 	case IB_EVENT_GID_CHANGE:
199 		/* Refresh port data asynchronously. */
200 		if (event->element.port_num <= sdev->device->phys_port_cnt) {
201 			sport = &sdev->port[event->element.port_num - 1];
202 			if (!sport->lid && !sport->sm_lid)
203 				schedule_work(&sport->work);
204 		}
205 		break;
206 	default:
207 		pr_err("received unrecognized IB event %d\n",
208 		       event->event);
209 		break;
210 	}
211 }
212 
213 /**
214  * srpt_srq_event() - SRQ event callback function.
215  */
216 static void srpt_srq_event(struct ib_event *event, void *ctx)
217 {
218 	pr_info("SRQ event %d\n", event->event);
219 }
220 
221 /**
222  * srpt_qp_event() - QP event callback function.
223  */
224 static void srpt_qp_event(struct ib_event *event, struct srpt_rdma_ch *ch)
225 {
226 	pr_debug("QP event %d on cm_id=%p sess_name=%s state=%d\n",
227 		 event->event, ch->cm_id, ch->sess_name, srpt_get_ch_state(ch));
228 
229 	switch (event->event) {
230 	case IB_EVENT_COMM_EST:
231 		ib_cm_notify(ch->cm_id, event->event);
232 		break;
233 	case IB_EVENT_QP_LAST_WQE_REACHED:
234 		if (srpt_test_and_set_ch_state(ch, CH_DRAINING,
235 					       CH_RELEASING))
236 			srpt_release_channel(ch);
237 		else
238 			pr_debug("%s: state %d - ignored LAST_WQE.\n",
239 				 ch->sess_name, srpt_get_ch_state(ch));
240 		break;
241 	default:
242 		pr_err("received unrecognized IB QP event %d\n", event->event);
243 		break;
244 	}
245 }
246 
247 /**
248  * srpt_set_ioc() - Helper function for initializing an IOUnitInfo structure.
249  *
250  * @slot: one-based slot number.
251  * @value: four-bit value.
252  *
253  * Copies the lowest four bits of value in element slot of the array of four
254  * bit elements called c_list (controller list). The index slot is one-based.
255  */
256 static void srpt_set_ioc(u8 *c_list, u32 slot, u8 value)
257 {
258 	u16 id;
259 	u8 tmp;
260 
261 	id = (slot - 1) / 2;
262 	if (slot & 0x1) {
263 		tmp = c_list[id] & 0xf;
264 		c_list[id] = (value << 4) | tmp;
265 	} else {
266 		tmp = c_list[id] & 0xf0;
267 		c_list[id] = (value & 0xf) | tmp;
268 	}
269 }
270 
271 /**
272  * srpt_get_class_port_info() - Copy ClassPortInfo to a management datagram.
273  *
274  * See also section 16.3.3.1 ClassPortInfo in the InfiniBand Architecture
275  * Specification.
276  */
277 static void srpt_get_class_port_info(struct ib_dm_mad *mad)
278 {
279 	struct ib_class_port_info *cif;
280 
281 	cif = (struct ib_class_port_info *)mad->data;
282 	memset(cif, 0, sizeof *cif);
283 	cif->base_version = 1;
284 	cif->class_version = 1;
285 	cif->resp_time_value = 20;
286 
287 	mad->mad_hdr.status = 0;
288 }
289 
290 /**
291  * srpt_get_iou() - Write IOUnitInfo to a management datagram.
292  *
293  * See also section 16.3.3.3 IOUnitInfo in the InfiniBand Architecture
294  * Specification. See also section B.7, table B.6 in the SRP r16a document.
295  */
296 static void srpt_get_iou(struct ib_dm_mad *mad)
297 {
298 	struct ib_dm_iou_info *ioui;
299 	u8 slot;
300 	int i;
301 
302 	ioui = (struct ib_dm_iou_info *)mad->data;
303 	ioui->change_id = cpu_to_be16(1);
304 	ioui->max_controllers = 16;
305 
306 	/* set present for slot 1 and empty for the rest */
307 	srpt_set_ioc(ioui->controller_list, 1, 1);
308 	for (i = 1, slot = 2; i < 16; i++, slot++)
309 		srpt_set_ioc(ioui->controller_list, slot, 0);
310 
311 	mad->mad_hdr.status = 0;
312 }
313 
314 /**
315  * srpt_get_ioc() - Write IOControllerprofile to a management datagram.
316  *
317  * See also section 16.3.3.4 IOControllerProfile in the InfiniBand
318  * Architecture Specification. See also section B.7, table B.7 in the SRP
319  * r16a document.
320  */
321 static void srpt_get_ioc(struct srpt_port *sport, u32 slot,
322 			 struct ib_dm_mad *mad)
323 {
324 	struct srpt_device *sdev = sport->sdev;
325 	struct ib_dm_ioc_profile *iocp;
326 
327 	iocp = (struct ib_dm_ioc_profile *)mad->data;
328 
329 	if (!slot || slot > 16) {
330 		mad->mad_hdr.status
331 			= cpu_to_be16(DM_MAD_STATUS_INVALID_FIELD);
332 		return;
333 	}
334 
335 	if (slot > 2) {
336 		mad->mad_hdr.status
337 			= cpu_to_be16(DM_MAD_STATUS_NO_IOC);
338 		return;
339 	}
340 
341 	memset(iocp, 0, sizeof *iocp);
342 	strcpy(iocp->id_string, SRPT_ID_STRING);
343 	iocp->guid = cpu_to_be64(srpt_service_guid);
344 	iocp->vendor_id = cpu_to_be32(sdev->dev_attr.vendor_id);
345 	iocp->device_id = cpu_to_be32(sdev->dev_attr.vendor_part_id);
346 	iocp->device_version = cpu_to_be16(sdev->dev_attr.hw_ver);
347 	iocp->subsys_vendor_id = cpu_to_be32(sdev->dev_attr.vendor_id);
348 	iocp->subsys_device_id = 0x0;
349 	iocp->io_class = cpu_to_be16(SRP_REV16A_IB_IO_CLASS);
350 	iocp->io_subclass = cpu_to_be16(SRP_IO_SUBCLASS);
351 	iocp->protocol = cpu_to_be16(SRP_PROTOCOL);
352 	iocp->protocol_version = cpu_to_be16(SRP_PROTOCOL_VERSION);
353 	iocp->send_queue_depth = cpu_to_be16(sdev->srq_size);
354 	iocp->rdma_read_depth = 4;
355 	iocp->send_size = cpu_to_be32(srp_max_req_size);
356 	iocp->rdma_size = cpu_to_be32(min(sport->port_attrib.srp_max_rdma_size,
357 					  1U << 24));
358 	iocp->num_svc_entries = 1;
359 	iocp->op_cap_mask = SRP_SEND_TO_IOC | SRP_SEND_FROM_IOC |
360 		SRP_RDMA_READ_FROM_IOC | SRP_RDMA_WRITE_FROM_IOC;
361 
362 	mad->mad_hdr.status = 0;
363 }
364 
365 /**
366  * srpt_get_svc_entries() - Write ServiceEntries to a management datagram.
367  *
368  * See also section 16.3.3.5 ServiceEntries in the InfiniBand Architecture
369  * Specification. See also section B.7, table B.8 in the SRP r16a document.
370  */
371 static void srpt_get_svc_entries(u64 ioc_guid,
372 				 u16 slot, u8 hi, u8 lo, struct ib_dm_mad *mad)
373 {
374 	struct ib_dm_svc_entries *svc_entries;
375 
376 	WARN_ON(!ioc_guid);
377 
378 	if (!slot || slot > 16) {
379 		mad->mad_hdr.status
380 			= cpu_to_be16(DM_MAD_STATUS_INVALID_FIELD);
381 		return;
382 	}
383 
384 	if (slot > 2 || lo > hi || hi > 1) {
385 		mad->mad_hdr.status
386 			= cpu_to_be16(DM_MAD_STATUS_NO_IOC);
387 		return;
388 	}
389 
390 	svc_entries = (struct ib_dm_svc_entries *)mad->data;
391 	memset(svc_entries, 0, sizeof *svc_entries);
392 	svc_entries->service_entries[0].id = cpu_to_be64(ioc_guid);
393 	snprintf(svc_entries->service_entries[0].name,
394 		 sizeof(svc_entries->service_entries[0].name),
395 		 "%s%016llx",
396 		 SRP_SERVICE_NAME_PREFIX,
397 		 ioc_guid);
398 
399 	mad->mad_hdr.status = 0;
400 }
401 
402 /**
403  * srpt_mgmt_method_get() - Process a received management datagram.
404  * @sp:      source port through which the MAD has been received.
405  * @rq_mad:  received MAD.
406  * @rsp_mad: response MAD.
407  */
408 static void srpt_mgmt_method_get(struct srpt_port *sp, struct ib_mad *rq_mad,
409 				 struct ib_dm_mad *rsp_mad)
410 {
411 	u16 attr_id;
412 	u32 slot;
413 	u8 hi, lo;
414 
415 	attr_id = be16_to_cpu(rq_mad->mad_hdr.attr_id);
416 	switch (attr_id) {
417 	case DM_ATTR_CLASS_PORT_INFO:
418 		srpt_get_class_port_info(rsp_mad);
419 		break;
420 	case DM_ATTR_IOU_INFO:
421 		srpt_get_iou(rsp_mad);
422 		break;
423 	case DM_ATTR_IOC_PROFILE:
424 		slot = be32_to_cpu(rq_mad->mad_hdr.attr_mod);
425 		srpt_get_ioc(sp, slot, rsp_mad);
426 		break;
427 	case DM_ATTR_SVC_ENTRIES:
428 		slot = be32_to_cpu(rq_mad->mad_hdr.attr_mod);
429 		hi = (u8) ((slot >> 8) & 0xff);
430 		lo = (u8) (slot & 0xff);
431 		slot = (u16) ((slot >> 16) & 0xffff);
432 		srpt_get_svc_entries(srpt_service_guid,
433 				     slot, hi, lo, rsp_mad);
434 		break;
435 	default:
436 		rsp_mad->mad_hdr.status =
437 		    cpu_to_be16(DM_MAD_STATUS_UNSUP_METHOD_ATTR);
438 		break;
439 	}
440 }
441 
442 /**
443  * srpt_mad_send_handler() - Post MAD-send callback function.
444  */
445 static void srpt_mad_send_handler(struct ib_mad_agent *mad_agent,
446 				  struct ib_mad_send_wc *mad_wc)
447 {
448 	ib_destroy_ah(mad_wc->send_buf->ah);
449 	ib_free_send_mad(mad_wc->send_buf);
450 }
451 
452 /**
453  * srpt_mad_recv_handler() - MAD reception callback function.
454  */
455 static void srpt_mad_recv_handler(struct ib_mad_agent *mad_agent,
456 				  struct ib_mad_recv_wc *mad_wc)
457 {
458 	struct srpt_port *sport = (struct srpt_port *)mad_agent->context;
459 	struct ib_ah *ah;
460 	struct ib_mad_send_buf *rsp;
461 	struct ib_dm_mad *dm_mad;
462 
463 	if (!mad_wc || !mad_wc->recv_buf.mad)
464 		return;
465 
466 	ah = ib_create_ah_from_wc(mad_agent->qp->pd, mad_wc->wc,
467 				  mad_wc->recv_buf.grh, mad_agent->port_num);
468 	if (IS_ERR(ah))
469 		goto err;
470 
471 	BUILD_BUG_ON(offsetof(struct ib_dm_mad, data) != IB_MGMT_DEVICE_HDR);
472 
473 	rsp = ib_create_send_mad(mad_agent, mad_wc->wc->src_qp,
474 				 mad_wc->wc->pkey_index, 0,
475 				 IB_MGMT_DEVICE_HDR, IB_MGMT_DEVICE_DATA,
476 				 GFP_KERNEL,
477 				 IB_MGMT_BASE_VERSION);
478 	if (IS_ERR(rsp))
479 		goto err_rsp;
480 
481 	rsp->ah = ah;
482 
483 	dm_mad = rsp->mad;
484 	memcpy(dm_mad, mad_wc->recv_buf.mad, sizeof *dm_mad);
485 	dm_mad->mad_hdr.method = IB_MGMT_METHOD_GET_RESP;
486 	dm_mad->mad_hdr.status = 0;
487 
488 	switch (mad_wc->recv_buf.mad->mad_hdr.method) {
489 	case IB_MGMT_METHOD_GET:
490 		srpt_mgmt_method_get(sport, mad_wc->recv_buf.mad, dm_mad);
491 		break;
492 	case IB_MGMT_METHOD_SET:
493 		dm_mad->mad_hdr.status =
494 		    cpu_to_be16(DM_MAD_STATUS_UNSUP_METHOD_ATTR);
495 		break;
496 	default:
497 		dm_mad->mad_hdr.status =
498 		    cpu_to_be16(DM_MAD_STATUS_UNSUP_METHOD);
499 		break;
500 	}
501 
502 	if (!ib_post_send_mad(rsp, NULL)) {
503 		ib_free_recv_mad(mad_wc);
504 		/* will destroy_ah & free_send_mad in send completion */
505 		return;
506 	}
507 
508 	ib_free_send_mad(rsp);
509 
510 err_rsp:
511 	ib_destroy_ah(ah);
512 err:
513 	ib_free_recv_mad(mad_wc);
514 }
515 
516 /**
517  * srpt_refresh_port() - Configure a HCA port.
518  *
519  * Enable InfiniBand management datagram processing, update the cached sm_lid,
520  * lid and gid values, and register a callback function for processing MADs
521  * on the specified port.
522  *
523  * Note: It is safe to call this function more than once for the same port.
524  */
525 static int srpt_refresh_port(struct srpt_port *sport)
526 {
527 	struct ib_mad_reg_req reg_req;
528 	struct ib_port_modify port_modify;
529 	struct ib_port_attr port_attr;
530 	int ret;
531 
532 	memset(&port_modify, 0, sizeof port_modify);
533 	port_modify.set_port_cap_mask = IB_PORT_DEVICE_MGMT_SUP;
534 	port_modify.clr_port_cap_mask = 0;
535 
536 	ret = ib_modify_port(sport->sdev->device, sport->port, 0, &port_modify);
537 	if (ret)
538 		goto err_mod_port;
539 
540 	ret = ib_query_port(sport->sdev->device, sport->port, &port_attr);
541 	if (ret)
542 		goto err_query_port;
543 
544 	sport->sm_lid = port_attr.sm_lid;
545 	sport->lid = port_attr.lid;
546 
547 	ret = ib_query_gid(sport->sdev->device, sport->port, 0, &sport->gid,
548 			   NULL);
549 	if (ret)
550 		goto err_query_port;
551 
552 	if (!sport->mad_agent) {
553 		memset(&reg_req, 0, sizeof reg_req);
554 		reg_req.mgmt_class = IB_MGMT_CLASS_DEVICE_MGMT;
555 		reg_req.mgmt_class_version = IB_MGMT_BASE_VERSION;
556 		set_bit(IB_MGMT_METHOD_GET, reg_req.method_mask);
557 		set_bit(IB_MGMT_METHOD_SET, reg_req.method_mask);
558 
559 		sport->mad_agent = ib_register_mad_agent(sport->sdev->device,
560 							 sport->port,
561 							 IB_QPT_GSI,
562 							 &reg_req, 0,
563 							 srpt_mad_send_handler,
564 							 srpt_mad_recv_handler,
565 							 sport, 0);
566 		if (IS_ERR(sport->mad_agent)) {
567 			ret = PTR_ERR(sport->mad_agent);
568 			sport->mad_agent = NULL;
569 			goto err_query_port;
570 		}
571 	}
572 
573 	return 0;
574 
575 err_query_port:
576 
577 	port_modify.set_port_cap_mask = 0;
578 	port_modify.clr_port_cap_mask = IB_PORT_DEVICE_MGMT_SUP;
579 	ib_modify_port(sport->sdev->device, sport->port, 0, &port_modify);
580 
581 err_mod_port:
582 
583 	return ret;
584 }
585 
586 /**
587  * srpt_unregister_mad_agent() - Unregister MAD callback functions.
588  *
589  * Note: It is safe to call this function more than once for the same device.
590  */
591 static void srpt_unregister_mad_agent(struct srpt_device *sdev)
592 {
593 	struct ib_port_modify port_modify = {
594 		.clr_port_cap_mask = IB_PORT_DEVICE_MGMT_SUP,
595 	};
596 	struct srpt_port *sport;
597 	int i;
598 
599 	for (i = 1; i <= sdev->device->phys_port_cnt; i++) {
600 		sport = &sdev->port[i - 1];
601 		WARN_ON(sport->port != i);
602 		if (ib_modify_port(sdev->device, i, 0, &port_modify) < 0)
603 			pr_err("disabling MAD processing failed.\n");
604 		if (sport->mad_agent) {
605 			ib_unregister_mad_agent(sport->mad_agent);
606 			sport->mad_agent = NULL;
607 		}
608 	}
609 }
610 
611 /**
612  * srpt_alloc_ioctx() - Allocate an SRPT I/O context structure.
613  */
614 static struct srpt_ioctx *srpt_alloc_ioctx(struct srpt_device *sdev,
615 					   int ioctx_size, int dma_size,
616 					   enum dma_data_direction dir)
617 {
618 	struct srpt_ioctx *ioctx;
619 
620 	ioctx = kmalloc(ioctx_size, GFP_KERNEL);
621 	if (!ioctx)
622 		goto err;
623 
624 	ioctx->buf = kmalloc(dma_size, GFP_KERNEL);
625 	if (!ioctx->buf)
626 		goto err_free_ioctx;
627 
628 	ioctx->dma = ib_dma_map_single(sdev->device, ioctx->buf, dma_size, dir);
629 	if (ib_dma_mapping_error(sdev->device, ioctx->dma))
630 		goto err_free_buf;
631 
632 	return ioctx;
633 
634 err_free_buf:
635 	kfree(ioctx->buf);
636 err_free_ioctx:
637 	kfree(ioctx);
638 err:
639 	return NULL;
640 }
641 
642 /**
643  * srpt_free_ioctx() - Free an SRPT I/O context structure.
644  */
645 static void srpt_free_ioctx(struct srpt_device *sdev, struct srpt_ioctx *ioctx,
646 			    int dma_size, enum dma_data_direction dir)
647 {
648 	if (!ioctx)
649 		return;
650 
651 	ib_dma_unmap_single(sdev->device, ioctx->dma, dma_size, dir);
652 	kfree(ioctx->buf);
653 	kfree(ioctx);
654 }
655 
656 /**
657  * srpt_alloc_ioctx_ring() - Allocate a ring of SRPT I/O context structures.
658  * @sdev:       Device to allocate the I/O context ring for.
659  * @ring_size:  Number of elements in the I/O context ring.
660  * @ioctx_size: I/O context size.
661  * @dma_size:   DMA buffer size.
662  * @dir:        DMA data direction.
663  */
664 static struct srpt_ioctx **srpt_alloc_ioctx_ring(struct srpt_device *sdev,
665 				int ring_size, int ioctx_size,
666 				int dma_size, enum dma_data_direction dir)
667 {
668 	struct srpt_ioctx **ring;
669 	int i;
670 
671 	WARN_ON(ioctx_size != sizeof(struct srpt_recv_ioctx)
672 		&& ioctx_size != sizeof(struct srpt_send_ioctx));
673 
674 	ring = kmalloc(ring_size * sizeof(ring[0]), GFP_KERNEL);
675 	if (!ring)
676 		goto out;
677 	for (i = 0; i < ring_size; ++i) {
678 		ring[i] = srpt_alloc_ioctx(sdev, ioctx_size, dma_size, dir);
679 		if (!ring[i])
680 			goto err;
681 		ring[i]->index = i;
682 	}
683 	goto out;
684 
685 err:
686 	while (--i >= 0)
687 		srpt_free_ioctx(sdev, ring[i], dma_size, dir);
688 	kfree(ring);
689 	ring = NULL;
690 out:
691 	return ring;
692 }
693 
694 /**
695  * srpt_free_ioctx_ring() - Free the ring of SRPT I/O context structures.
696  */
697 static void srpt_free_ioctx_ring(struct srpt_ioctx **ioctx_ring,
698 				 struct srpt_device *sdev, int ring_size,
699 				 int dma_size, enum dma_data_direction dir)
700 {
701 	int i;
702 
703 	for (i = 0; i < ring_size; ++i)
704 		srpt_free_ioctx(sdev, ioctx_ring[i], dma_size, dir);
705 	kfree(ioctx_ring);
706 }
707 
708 /**
709  * srpt_get_cmd_state() - Get the state of a SCSI command.
710  */
711 static enum srpt_command_state srpt_get_cmd_state(struct srpt_send_ioctx *ioctx)
712 {
713 	enum srpt_command_state state;
714 	unsigned long flags;
715 
716 	BUG_ON(!ioctx);
717 
718 	spin_lock_irqsave(&ioctx->spinlock, flags);
719 	state = ioctx->state;
720 	spin_unlock_irqrestore(&ioctx->spinlock, flags);
721 	return state;
722 }
723 
724 /**
725  * srpt_set_cmd_state() - Set the state of a SCSI command.
726  *
727  * Does not modify the state of aborted commands. Returns the previous command
728  * state.
729  */
730 static enum srpt_command_state srpt_set_cmd_state(struct srpt_send_ioctx *ioctx,
731 						  enum srpt_command_state new)
732 {
733 	enum srpt_command_state previous;
734 	unsigned long flags;
735 
736 	BUG_ON(!ioctx);
737 
738 	spin_lock_irqsave(&ioctx->spinlock, flags);
739 	previous = ioctx->state;
740 	if (previous != SRPT_STATE_DONE)
741 		ioctx->state = new;
742 	spin_unlock_irqrestore(&ioctx->spinlock, flags);
743 
744 	return previous;
745 }
746 
747 /**
748  * srpt_test_and_set_cmd_state() - Test and set the state of a command.
749  *
750  * Returns true if and only if the previous command state was equal to 'old'.
751  */
752 static bool srpt_test_and_set_cmd_state(struct srpt_send_ioctx *ioctx,
753 					enum srpt_command_state old,
754 					enum srpt_command_state new)
755 {
756 	enum srpt_command_state previous;
757 	unsigned long flags;
758 
759 	WARN_ON(!ioctx);
760 	WARN_ON(old == SRPT_STATE_DONE);
761 	WARN_ON(new == SRPT_STATE_NEW);
762 
763 	spin_lock_irqsave(&ioctx->spinlock, flags);
764 	previous = ioctx->state;
765 	if (previous == old)
766 		ioctx->state = new;
767 	spin_unlock_irqrestore(&ioctx->spinlock, flags);
768 	return previous == old;
769 }
770 
771 /**
772  * srpt_post_recv() - Post an IB receive request.
773  */
774 static int srpt_post_recv(struct srpt_device *sdev,
775 			  struct srpt_recv_ioctx *ioctx)
776 {
777 	struct ib_sge list;
778 	struct ib_recv_wr wr, *bad_wr;
779 
780 	BUG_ON(!sdev);
781 	wr.wr_id = encode_wr_id(SRPT_RECV, ioctx->ioctx.index);
782 
783 	list.addr = ioctx->ioctx.dma;
784 	list.length = srp_max_req_size;
785 	list.lkey = sdev->pd->local_dma_lkey;
786 
787 	wr.next = NULL;
788 	wr.sg_list = &list;
789 	wr.num_sge = 1;
790 
791 	return ib_post_srq_recv(sdev->srq, &wr, &bad_wr);
792 }
793 
794 /**
795  * srpt_post_send() - Post an IB send request.
796  *
797  * Returns zero upon success and a non-zero value upon failure.
798  */
799 static int srpt_post_send(struct srpt_rdma_ch *ch,
800 			  struct srpt_send_ioctx *ioctx, int len)
801 {
802 	struct ib_sge list;
803 	struct ib_send_wr wr, *bad_wr;
804 	struct srpt_device *sdev = ch->sport->sdev;
805 	int ret;
806 
807 	atomic_inc(&ch->req_lim);
808 
809 	ret = -ENOMEM;
810 	if (unlikely(atomic_dec_return(&ch->sq_wr_avail) < 0)) {
811 		pr_warn("IB send queue full (needed 1)\n");
812 		goto out;
813 	}
814 
815 	ib_dma_sync_single_for_device(sdev->device, ioctx->ioctx.dma, len,
816 				      DMA_TO_DEVICE);
817 
818 	list.addr = ioctx->ioctx.dma;
819 	list.length = len;
820 	list.lkey = sdev->pd->local_dma_lkey;
821 
822 	wr.next = NULL;
823 	wr.wr_id = encode_wr_id(SRPT_SEND, ioctx->ioctx.index);
824 	wr.sg_list = &list;
825 	wr.num_sge = 1;
826 	wr.opcode = IB_WR_SEND;
827 	wr.send_flags = IB_SEND_SIGNALED;
828 
829 	ret = ib_post_send(ch->qp, &wr, &bad_wr);
830 
831 out:
832 	if (ret < 0) {
833 		atomic_inc(&ch->sq_wr_avail);
834 		atomic_dec(&ch->req_lim);
835 	}
836 	return ret;
837 }
838 
839 /**
840  * srpt_get_desc_tbl() - Parse the data descriptors of an SRP_CMD request.
841  * @ioctx: Pointer to the I/O context associated with the request.
842  * @srp_cmd: Pointer to the SRP_CMD request data.
843  * @dir: Pointer to the variable to which the transfer direction will be
844  *   written.
845  * @data_len: Pointer to the variable to which the total data length of all
846  *   descriptors in the SRP_CMD request will be written.
847  *
848  * This function initializes ioctx->nrbuf and ioctx->r_bufs.
849  *
850  * Returns -EINVAL when the SRP_CMD request contains inconsistent descriptors;
851  * -ENOMEM when memory allocation fails and zero upon success.
852  */
853 static int srpt_get_desc_tbl(struct srpt_send_ioctx *ioctx,
854 			     struct srp_cmd *srp_cmd,
855 			     enum dma_data_direction *dir, u64 *data_len)
856 {
857 	struct srp_indirect_buf *idb;
858 	struct srp_direct_buf *db;
859 	unsigned add_cdb_offset;
860 	int ret;
861 
862 	/*
863 	 * The pointer computations below will only be compiled correctly
864 	 * if srp_cmd::add_data is declared as s8*, u8*, s8[] or u8[], so check
865 	 * whether srp_cmd::add_data has been declared as a byte pointer.
866 	 */
867 	BUILD_BUG_ON(!__same_type(srp_cmd->add_data[0], (s8)0)
868 		     && !__same_type(srp_cmd->add_data[0], (u8)0));
869 
870 	BUG_ON(!dir);
871 	BUG_ON(!data_len);
872 
873 	ret = 0;
874 	*data_len = 0;
875 
876 	/*
877 	 * The lower four bits of the buffer format field contain the DATA-IN
878 	 * buffer descriptor format, and the highest four bits contain the
879 	 * DATA-OUT buffer descriptor format.
880 	 */
881 	*dir = DMA_NONE;
882 	if (srp_cmd->buf_fmt & 0xf)
883 		/* DATA-IN: transfer data from target to initiator (read). */
884 		*dir = DMA_FROM_DEVICE;
885 	else if (srp_cmd->buf_fmt >> 4)
886 		/* DATA-OUT: transfer data from initiator to target (write). */
887 		*dir = DMA_TO_DEVICE;
888 
889 	/*
890 	 * According to the SRP spec, the lower two bits of the 'ADDITIONAL
891 	 * CDB LENGTH' field are reserved and the size in bytes of this field
892 	 * is four times the value specified in bits 3..7. Hence the "& ~3".
893 	 */
894 	add_cdb_offset = srp_cmd->add_cdb_len & ~3;
895 	if (((srp_cmd->buf_fmt & 0xf) == SRP_DATA_DESC_DIRECT) ||
896 	    ((srp_cmd->buf_fmt >> 4) == SRP_DATA_DESC_DIRECT)) {
897 		ioctx->n_rbuf = 1;
898 		ioctx->rbufs = &ioctx->single_rbuf;
899 
900 		db = (struct srp_direct_buf *)(srp_cmd->add_data
901 					       + add_cdb_offset);
902 		memcpy(ioctx->rbufs, db, sizeof *db);
903 		*data_len = be32_to_cpu(db->len);
904 	} else if (((srp_cmd->buf_fmt & 0xf) == SRP_DATA_DESC_INDIRECT) ||
905 		   ((srp_cmd->buf_fmt >> 4) == SRP_DATA_DESC_INDIRECT)) {
906 		idb = (struct srp_indirect_buf *)(srp_cmd->add_data
907 						  + add_cdb_offset);
908 
909 		ioctx->n_rbuf = be32_to_cpu(idb->table_desc.len) / sizeof *db;
910 
911 		if (ioctx->n_rbuf >
912 		    (srp_cmd->data_out_desc_cnt + srp_cmd->data_in_desc_cnt)) {
913 			pr_err("received unsupported SRP_CMD request"
914 			       " type (%u out + %u in != %u / %zu)\n",
915 			       srp_cmd->data_out_desc_cnt,
916 			       srp_cmd->data_in_desc_cnt,
917 			       be32_to_cpu(idb->table_desc.len),
918 			       sizeof(*db));
919 			ioctx->n_rbuf = 0;
920 			ret = -EINVAL;
921 			goto out;
922 		}
923 
924 		if (ioctx->n_rbuf == 1)
925 			ioctx->rbufs = &ioctx->single_rbuf;
926 		else {
927 			ioctx->rbufs =
928 				kmalloc(ioctx->n_rbuf * sizeof *db, GFP_ATOMIC);
929 			if (!ioctx->rbufs) {
930 				ioctx->n_rbuf = 0;
931 				ret = -ENOMEM;
932 				goto out;
933 			}
934 		}
935 
936 		db = idb->desc_list;
937 		memcpy(ioctx->rbufs, db, ioctx->n_rbuf * sizeof *db);
938 		*data_len = be32_to_cpu(idb->len);
939 	}
940 out:
941 	return ret;
942 }
943 
944 /**
945  * srpt_init_ch_qp() - Initialize queue pair attributes.
946  *
947  * Initialized the attributes of queue pair 'qp' by allowing local write,
948  * remote read and remote write. Also transitions 'qp' to state IB_QPS_INIT.
949  */
950 static int srpt_init_ch_qp(struct srpt_rdma_ch *ch, struct ib_qp *qp)
951 {
952 	struct ib_qp_attr *attr;
953 	int ret;
954 
955 	attr = kzalloc(sizeof *attr, GFP_KERNEL);
956 	if (!attr)
957 		return -ENOMEM;
958 
959 	attr->qp_state = IB_QPS_INIT;
960 	attr->qp_access_flags = IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_READ |
961 	    IB_ACCESS_REMOTE_WRITE;
962 	attr->port_num = ch->sport->port;
963 	attr->pkey_index = 0;
964 
965 	ret = ib_modify_qp(qp, attr,
966 			   IB_QP_STATE | IB_QP_ACCESS_FLAGS | IB_QP_PORT |
967 			   IB_QP_PKEY_INDEX);
968 
969 	kfree(attr);
970 	return ret;
971 }
972 
973 /**
974  * srpt_ch_qp_rtr() - Change the state of a channel to 'ready to receive' (RTR).
975  * @ch: channel of the queue pair.
976  * @qp: queue pair to change the state of.
977  *
978  * Returns zero upon success and a negative value upon failure.
979  *
980  * Note: currently a struct ib_qp_attr takes 136 bytes on a 64-bit system.
981  * If this structure ever becomes larger, it might be necessary to allocate
982  * it dynamically instead of on the stack.
983  */
984 static int srpt_ch_qp_rtr(struct srpt_rdma_ch *ch, struct ib_qp *qp)
985 {
986 	struct ib_qp_attr qp_attr;
987 	int attr_mask;
988 	int ret;
989 
990 	qp_attr.qp_state = IB_QPS_RTR;
991 	ret = ib_cm_init_qp_attr(ch->cm_id, &qp_attr, &attr_mask);
992 	if (ret)
993 		goto out;
994 
995 	qp_attr.max_dest_rd_atomic = 4;
996 
997 	ret = ib_modify_qp(qp, &qp_attr, attr_mask);
998 
999 out:
1000 	return ret;
1001 }
1002 
1003 /**
1004  * srpt_ch_qp_rts() - Change the state of a channel to 'ready to send' (RTS).
1005  * @ch: channel of the queue pair.
1006  * @qp: queue pair to change the state of.
1007  *
1008  * Returns zero upon success and a negative value upon failure.
1009  *
1010  * Note: currently a struct ib_qp_attr takes 136 bytes on a 64-bit system.
1011  * If this structure ever becomes larger, it might be necessary to allocate
1012  * it dynamically instead of on the stack.
1013  */
1014 static int srpt_ch_qp_rts(struct srpt_rdma_ch *ch, struct ib_qp *qp)
1015 {
1016 	struct ib_qp_attr qp_attr;
1017 	int attr_mask;
1018 	int ret;
1019 
1020 	qp_attr.qp_state = IB_QPS_RTS;
1021 	ret = ib_cm_init_qp_attr(ch->cm_id, &qp_attr, &attr_mask);
1022 	if (ret)
1023 		goto out;
1024 
1025 	qp_attr.max_rd_atomic = 4;
1026 
1027 	ret = ib_modify_qp(qp, &qp_attr, attr_mask);
1028 
1029 out:
1030 	return ret;
1031 }
1032 
1033 /**
1034  * srpt_ch_qp_err() - Set the channel queue pair state to 'error'.
1035  */
1036 static int srpt_ch_qp_err(struct srpt_rdma_ch *ch)
1037 {
1038 	struct ib_qp_attr qp_attr;
1039 
1040 	qp_attr.qp_state = IB_QPS_ERR;
1041 	return ib_modify_qp(ch->qp, &qp_attr, IB_QP_STATE);
1042 }
1043 
1044 /**
1045  * srpt_unmap_sg_to_ib_sge() - Unmap an IB SGE list.
1046  */
1047 static void srpt_unmap_sg_to_ib_sge(struct srpt_rdma_ch *ch,
1048 				    struct srpt_send_ioctx *ioctx)
1049 {
1050 	struct scatterlist *sg;
1051 	enum dma_data_direction dir;
1052 
1053 	BUG_ON(!ch);
1054 	BUG_ON(!ioctx);
1055 	BUG_ON(ioctx->n_rdma && !ioctx->rdma_ius);
1056 
1057 	while (ioctx->n_rdma)
1058 		kfree(ioctx->rdma_ius[--ioctx->n_rdma].sge);
1059 
1060 	kfree(ioctx->rdma_ius);
1061 	ioctx->rdma_ius = NULL;
1062 
1063 	if (ioctx->mapped_sg_count) {
1064 		sg = ioctx->sg;
1065 		WARN_ON(!sg);
1066 		dir = ioctx->cmd.data_direction;
1067 		BUG_ON(dir == DMA_NONE);
1068 		ib_dma_unmap_sg(ch->sport->sdev->device, sg, ioctx->sg_cnt,
1069 				opposite_dma_dir(dir));
1070 		ioctx->mapped_sg_count = 0;
1071 	}
1072 }
1073 
1074 /**
1075  * srpt_map_sg_to_ib_sge() - Map an SG list to an IB SGE list.
1076  */
1077 static int srpt_map_sg_to_ib_sge(struct srpt_rdma_ch *ch,
1078 				 struct srpt_send_ioctx *ioctx)
1079 {
1080 	struct ib_device *dev = ch->sport->sdev->device;
1081 	struct se_cmd *cmd;
1082 	struct scatterlist *sg, *sg_orig;
1083 	int sg_cnt;
1084 	enum dma_data_direction dir;
1085 	struct rdma_iu *riu;
1086 	struct srp_direct_buf *db;
1087 	dma_addr_t dma_addr;
1088 	struct ib_sge *sge;
1089 	u64 raddr;
1090 	u32 rsize;
1091 	u32 tsize;
1092 	u32 dma_len;
1093 	int count, nrdma;
1094 	int i, j, k;
1095 
1096 	BUG_ON(!ch);
1097 	BUG_ON(!ioctx);
1098 	cmd = &ioctx->cmd;
1099 	dir = cmd->data_direction;
1100 	BUG_ON(dir == DMA_NONE);
1101 
1102 	ioctx->sg = sg = sg_orig = cmd->t_data_sg;
1103 	ioctx->sg_cnt = sg_cnt = cmd->t_data_nents;
1104 
1105 	count = ib_dma_map_sg(ch->sport->sdev->device, sg, sg_cnt,
1106 			      opposite_dma_dir(dir));
1107 	if (unlikely(!count))
1108 		return -EAGAIN;
1109 
1110 	ioctx->mapped_sg_count = count;
1111 
1112 	if (ioctx->rdma_ius && ioctx->n_rdma_ius)
1113 		nrdma = ioctx->n_rdma_ius;
1114 	else {
1115 		nrdma = (count + SRPT_DEF_SG_PER_WQE - 1) / SRPT_DEF_SG_PER_WQE
1116 			+ ioctx->n_rbuf;
1117 
1118 		ioctx->rdma_ius = kzalloc(nrdma * sizeof *riu, GFP_KERNEL);
1119 		if (!ioctx->rdma_ius)
1120 			goto free_mem;
1121 
1122 		ioctx->n_rdma_ius = nrdma;
1123 	}
1124 
1125 	db = ioctx->rbufs;
1126 	tsize = cmd->data_length;
1127 	dma_len = ib_sg_dma_len(dev, &sg[0]);
1128 	riu = ioctx->rdma_ius;
1129 
1130 	/*
1131 	 * For each remote desc - calculate the #ib_sge.
1132 	 * If #ib_sge < SRPT_DEF_SG_PER_WQE per rdma operation then
1133 	 *      each remote desc rdma_iu is required a rdma wr;
1134 	 * else
1135 	 *      we need to allocate extra rdma_iu to carry extra #ib_sge in
1136 	 *      another rdma wr
1137 	 */
1138 	for (i = 0, j = 0;
1139 	     j < count && i < ioctx->n_rbuf && tsize > 0; ++i, ++riu, ++db) {
1140 		rsize = be32_to_cpu(db->len);
1141 		raddr = be64_to_cpu(db->va);
1142 		riu->raddr = raddr;
1143 		riu->rkey = be32_to_cpu(db->key);
1144 		riu->sge_cnt = 0;
1145 
1146 		/* calculate how many sge required for this remote_buf */
1147 		while (rsize > 0 && tsize > 0) {
1148 
1149 			if (rsize >= dma_len) {
1150 				tsize -= dma_len;
1151 				rsize -= dma_len;
1152 				raddr += dma_len;
1153 
1154 				if (tsize > 0) {
1155 					++j;
1156 					if (j < count) {
1157 						sg = sg_next(sg);
1158 						dma_len = ib_sg_dma_len(
1159 								dev, sg);
1160 					}
1161 				}
1162 			} else {
1163 				tsize -= rsize;
1164 				dma_len -= rsize;
1165 				rsize = 0;
1166 			}
1167 
1168 			++riu->sge_cnt;
1169 
1170 			if (rsize > 0 && riu->sge_cnt == SRPT_DEF_SG_PER_WQE) {
1171 				++ioctx->n_rdma;
1172 				riu->sge =
1173 				    kmalloc(riu->sge_cnt * sizeof *riu->sge,
1174 					    GFP_KERNEL);
1175 				if (!riu->sge)
1176 					goto free_mem;
1177 
1178 				++riu;
1179 				riu->sge_cnt = 0;
1180 				riu->raddr = raddr;
1181 				riu->rkey = be32_to_cpu(db->key);
1182 			}
1183 		}
1184 
1185 		++ioctx->n_rdma;
1186 		riu->sge = kmalloc(riu->sge_cnt * sizeof *riu->sge,
1187 				   GFP_KERNEL);
1188 		if (!riu->sge)
1189 			goto free_mem;
1190 	}
1191 
1192 	db = ioctx->rbufs;
1193 	tsize = cmd->data_length;
1194 	riu = ioctx->rdma_ius;
1195 	sg = sg_orig;
1196 	dma_len = ib_sg_dma_len(dev, &sg[0]);
1197 	dma_addr = ib_sg_dma_address(dev, &sg[0]);
1198 
1199 	/* this second loop is really mapped sg_addres to rdma_iu->ib_sge */
1200 	for (i = 0, j = 0;
1201 	     j < count && i < ioctx->n_rbuf && tsize > 0; ++i, ++riu, ++db) {
1202 		rsize = be32_to_cpu(db->len);
1203 		sge = riu->sge;
1204 		k = 0;
1205 
1206 		while (rsize > 0 && tsize > 0) {
1207 			sge->addr = dma_addr;
1208 			sge->lkey = ch->sport->sdev->pd->local_dma_lkey;
1209 
1210 			if (rsize >= dma_len) {
1211 				sge->length =
1212 					(tsize < dma_len) ? tsize : dma_len;
1213 				tsize -= dma_len;
1214 				rsize -= dma_len;
1215 
1216 				if (tsize > 0) {
1217 					++j;
1218 					if (j < count) {
1219 						sg = sg_next(sg);
1220 						dma_len = ib_sg_dma_len(
1221 								dev, sg);
1222 						dma_addr = ib_sg_dma_address(
1223 								dev, sg);
1224 					}
1225 				}
1226 			} else {
1227 				sge->length = (tsize < rsize) ? tsize : rsize;
1228 				tsize -= rsize;
1229 				dma_len -= rsize;
1230 				dma_addr += rsize;
1231 				rsize = 0;
1232 			}
1233 
1234 			++k;
1235 			if (k == riu->sge_cnt && rsize > 0 && tsize > 0) {
1236 				++riu;
1237 				sge = riu->sge;
1238 				k = 0;
1239 			} else if (rsize > 0 && tsize > 0)
1240 				++sge;
1241 		}
1242 	}
1243 
1244 	return 0;
1245 
1246 free_mem:
1247 	srpt_unmap_sg_to_ib_sge(ch, ioctx);
1248 
1249 	return -ENOMEM;
1250 }
1251 
1252 /**
1253  * srpt_get_send_ioctx() - Obtain an I/O context for sending to the initiator.
1254  */
1255 static struct srpt_send_ioctx *srpt_get_send_ioctx(struct srpt_rdma_ch *ch)
1256 {
1257 	struct srpt_send_ioctx *ioctx;
1258 	unsigned long flags;
1259 
1260 	BUG_ON(!ch);
1261 
1262 	ioctx = NULL;
1263 	spin_lock_irqsave(&ch->spinlock, flags);
1264 	if (!list_empty(&ch->free_list)) {
1265 		ioctx = list_first_entry(&ch->free_list,
1266 					 struct srpt_send_ioctx, free_list);
1267 		list_del(&ioctx->free_list);
1268 	}
1269 	spin_unlock_irqrestore(&ch->spinlock, flags);
1270 
1271 	if (!ioctx)
1272 		return ioctx;
1273 
1274 	BUG_ON(ioctx->ch != ch);
1275 	spin_lock_init(&ioctx->spinlock);
1276 	ioctx->state = SRPT_STATE_NEW;
1277 	ioctx->n_rbuf = 0;
1278 	ioctx->rbufs = NULL;
1279 	ioctx->n_rdma = 0;
1280 	ioctx->n_rdma_ius = 0;
1281 	ioctx->rdma_ius = NULL;
1282 	ioctx->mapped_sg_count = 0;
1283 	init_completion(&ioctx->tx_done);
1284 	ioctx->queue_status_only = false;
1285 	/*
1286 	 * transport_init_se_cmd() does not initialize all fields, so do it
1287 	 * here.
1288 	 */
1289 	memset(&ioctx->cmd, 0, sizeof(ioctx->cmd));
1290 	memset(&ioctx->sense_data, 0, sizeof(ioctx->sense_data));
1291 
1292 	return ioctx;
1293 }
1294 
1295 /**
1296  * srpt_abort_cmd() - Abort a SCSI command.
1297  * @ioctx:   I/O context associated with the SCSI command.
1298  * @context: Preferred execution context.
1299  */
1300 static int srpt_abort_cmd(struct srpt_send_ioctx *ioctx)
1301 {
1302 	enum srpt_command_state state;
1303 	unsigned long flags;
1304 
1305 	BUG_ON(!ioctx);
1306 
1307 	/*
1308 	 * If the command is in a state where the target core is waiting for
1309 	 * the ib_srpt driver, change the state to the next state. Changing
1310 	 * the state of the command from SRPT_STATE_NEED_DATA to
1311 	 * SRPT_STATE_DATA_IN ensures that srpt_xmit_response() will call this
1312 	 * function a second time.
1313 	 */
1314 
1315 	spin_lock_irqsave(&ioctx->spinlock, flags);
1316 	state = ioctx->state;
1317 	switch (state) {
1318 	case SRPT_STATE_NEED_DATA:
1319 		ioctx->state = SRPT_STATE_DATA_IN;
1320 		break;
1321 	case SRPT_STATE_DATA_IN:
1322 	case SRPT_STATE_CMD_RSP_SENT:
1323 	case SRPT_STATE_MGMT_RSP_SENT:
1324 		ioctx->state = SRPT_STATE_DONE;
1325 		break;
1326 	default:
1327 		break;
1328 	}
1329 	spin_unlock_irqrestore(&ioctx->spinlock, flags);
1330 
1331 	if (state == SRPT_STATE_DONE) {
1332 		struct srpt_rdma_ch *ch = ioctx->ch;
1333 
1334 		BUG_ON(ch->sess == NULL);
1335 
1336 		target_put_sess_cmd(&ioctx->cmd);
1337 		goto out;
1338 	}
1339 
1340 	pr_debug("Aborting cmd with state %d and tag %lld\n", state,
1341 		 ioctx->cmd.tag);
1342 
1343 	switch (state) {
1344 	case SRPT_STATE_NEW:
1345 	case SRPT_STATE_DATA_IN:
1346 	case SRPT_STATE_MGMT:
1347 		/*
1348 		 * Do nothing - defer abort processing until
1349 		 * srpt_queue_response() is invoked.
1350 		 */
1351 		WARN_ON(!transport_check_aborted_status(&ioctx->cmd, false));
1352 		break;
1353 	case SRPT_STATE_NEED_DATA:
1354 		/* DMA_TO_DEVICE (write) - RDMA read error. */
1355 
1356 		/* XXX(hch): this is a horrible layering violation.. */
1357 		spin_lock_irqsave(&ioctx->cmd.t_state_lock, flags);
1358 		ioctx->cmd.transport_state &= ~CMD_T_ACTIVE;
1359 		spin_unlock_irqrestore(&ioctx->cmd.t_state_lock, flags);
1360 		break;
1361 	case SRPT_STATE_CMD_RSP_SENT:
1362 		/*
1363 		 * SRP_RSP sending failed or the SRP_RSP send completion has
1364 		 * not been received in time.
1365 		 */
1366 		srpt_unmap_sg_to_ib_sge(ioctx->ch, ioctx);
1367 		target_put_sess_cmd(&ioctx->cmd);
1368 		break;
1369 	case SRPT_STATE_MGMT_RSP_SENT:
1370 		srpt_set_cmd_state(ioctx, SRPT_STATE_DONE);
1371 		target_put_sess_cmd(&ioctx->cmd);
1372 		break;
1373 	default:
1374 		WARN(1, "Unexpected command state (%d)", state);
1375 		break;
1376 	}
1377 
1378 out:
1379 	return state;
1380 }
1381 
1382 /**
1383  * srpt_handle_send_err_comp() - Process an IB_WC_SEND error completion.
1384  */
1385 static void srpt_handle_send_err_comp(struct srpt_rdma_ch *ch, u64 wr_id)
1386 {
1387 	struct srpt_send_ioctx *ioctx;
1388 	enum srpt_command_state state;
1389 	u32 index;
1390 
1391 	atomic_inc(&ch->sq_wr_avail);
1392 
1393 	index = idx_from_wr_id(wr_id);
1394 	ioctx = ch->ioctx_ring[index];
1395 	state = srpt_get_cmd_state(ioctx);
1396 
1397 	WARN_ON(state != SRPT_STATE_CMD_RSP_SENT
1398 		&& state != SRPT_STATE_MGMT_RSP_SENT
1399 		&& state != SRPT_STATE_NEED_DATA
1400 		&& state != SRPT_STATE_DONE);
1401 
1402 	/* If SRP_RSP sending failed, undo the ch->req_lim change. */
1403 	if (state == SRPT_STATE_CMD_RSP_SENT
1404 	    || state == SRPT_STATE_MGMT_RSP_SENT)
1405 		atomic_dec(&ch->req_lim);
1406 
1407 	srpt_abort_cmd(ioctx);
1408 }
1409 
1410 /**
1411  * srpt_handle_send_comp() - Process an IB send completion notification.
1412  */
1413 static void srpt_handle_send_comp(struct srpt_rdma_ch *ch,
1414 				  struct srpt_send_ioctx *ioctx)
1415 {
1416 	enum srpt_command_state state;
1417 
1418 	atomic_inc(&ch->sq_wr_avail);
1419 
1420 	state = srpt_set_cmd_state(ioctx, SRPT_STATE_DONE);
1421 
1422 	if (WARN_ON(state != SRPT_STATE_CMD_RSP_SENT
1423 		    && state != SRPT_STATE_MGMT_RSP_SENT
1424 		    && state != SRPT_STATE_DONE))
1425 		pr_debug("state = %d\n", state);
1426 
1427 	if (state != SRPT_STATE_DONE) {
1428 		srpt_unmap_sg_to_ib_sge(ch, ioctx);
1429 		transport_generic_free_cmd(&ioctx->cmd, 0);
1430 	} else {
1431 		pr_err("IB completion has been received too late for"
1432 		       " wr_id = %u.\n", ioctx->ioctx.index);
1433 	}
1434 }
1435 
1436 /**
1437  * srpt_handle_rdma_comp() - Process an IB RDMA completion notification.
1438  *
1439  * XXX: what is now target_execute_cmd used to be asynchronous, and unmapping
1440  * the data that has been transferred via IB RDMA had to be postponed until the
1441  * check_stop_free() callback.  None of this is necessary anymore and needs to
1442  * be cleaned up.
1443  */
1444 static void srpt_handle_rdma_comp(struct srpt_rdma_ch *ch,
1445 				  struct srpt_send_ioctx *ioctx,
1446 				  enum srpt_opcode opcode)
1447 {
1448 	WARN_ON(ioctx->n_rdma <= 0);
1449 	atomic_add(ioctx->n_rdma, &ch->sq_wr_avail);
1450 
1451 	if (opcode == SRPT_RDMA_READ_LAST) {
1452 		if (srpt_test_and_set_cmd_state(ioctx, SRPT_STATE_NEED_DATA,
1453 						SRPT_STATE_DATA_IN))
1454 			target_execute_cmd(&ioctx->cmd);
1455 		else
1456 			pr_err("%s[%d]: wrong state = %d\n", __func__,
1457 			       __LINE__, srpt_get_cmd_state(ioctx));
1458 	} else if (opcode == SRPT_RDMA_ABORT) {
1459 		ioctx->rdma_aborted = true;
1460 	} else {
1461 		WARN(true, "unexpected opcode %d\n", opcode);
1462 	}
1463 }
1464 
1465 /**
1466  * srpt_handle_rdma_err_comp() - Process an IB RDMA error completion.
1467  */
1468 static void srpt_handle_rdma_err_comp(struct srpt_rdma_ch *ch,
1469 				      struct srpt_send_ioctx *ioctx,
1470 				      enum srpt_opcode opcode)
1471 {
1472 	enum srpt_command_state state;
1473 
1474 	state = srpt_get_cmd_state(ioctx);
1475 	switch (opcode) {
1476 	case SRPT_RDMA_READ_LAST:
1477 		if (ioctx->n_rdma <= 0) {
1478 			pr_err("Received invalid RDMA read"
1479 			       " error completion with idx %d\n",
1480 			       ioctx->ioctx.index);
1481 			break;
1482 		}
1483 		atomic_add(ioctx->n_rdma, &ch->sq_wr_avail);
1484 		if (state == SRPT_STATE_NEED_DATA)
1485 			srpt_abort_cmd(ioctx);
1486 		else
1487 			pr_err("%s[%d]: wrong state = %d\n",
1488 			       __func__, __LINE__, state);
1489 		break;
1490 	case SRPT_RDMA_WRITE_LAST:
1491 		break;
1492 	default:
1493 		pr_err("%s[%d]: opcode = %u\n", __func__, __LINE__, opcode);
1494 		break;
1495 	}
1496 }
1497 
1498 /**
1499  * srpt_build_cmd_rsp() - Build an SRP_RSP response.
1500  * @ch: RDMA channel through which the request has been received.
1501  * @ioctx: I/O context associated with the SRP_CMD request. The response will
1502  *   be built in the buffer ioctx->buf points at and hence this function will
1503  *   overwrite the request data.
1504  * @tag: tag of the request for which this response is being generated.
1505  * @status: value for the STATUS field of the SRP_RSP information unit.
1506  *
1507  * Returns the size in bytes of the SRP_RSP response.
1508  *
1509  * An SRP_RSP response contains a SCSI status or service response. See also
1510  * section 6.9 in the SRP r16a document for the format of an SRP_RSP
1511  * response. See also SPC-2 for more information about sense data.
1512  */
1513 static int srpt_build_cmd_rsp(struct srpt_rdma_ch *ch,
1514 			      struct srpt_send_ioctx *ioctx, u64 tag,
1515 			      int status)
1516 {
1517 	struct srp_rsp *srp_rsp;
1518 	const u8 *sense_data;
1519 	int sense_data_len, max_sense_len;
1520 
1521 	/*
1522 	 * The lowest bit of all SAM-3 status codes is zero (see also
1523 	 * paragraph 5.3 in SAM-3).
1524 	 */
1525 	WARN_ON(status & 1);
1526 
1527 	srp_rsp = ioctx->ioctx.buf;
1528 	BUG_ON(!srp_rsp);
1529 
1530 	sense_data = ioctx->sense_data;
1531 	sense_data_len = ioctx->cmd.scsi_sense_length;
1532 	WARN_ON(sense_data_len > sizeof(ioctx->sense_data));
1533 
1534 	memset(srp_rsp, 0, sizeof *srp_rsp);
1535 	srp_rsp->opcode = SRP_RSP;
1536 	srp_rsp->req_lim_delta =
1537 		cpu_to_be32(1 + atomic_xchg(&ch->req_lim_delta, 0));
1538 	srp_rsp->tag = tag;
1539 	srp_rsp->status = status;
1540 
1541 	if (sense_data_len) {
1542 		BUILD_BUG_ON(MIN_MAX_RSP_SIZE <= sizeof(*srp_rsp));
1543 		max_sense_len = ch->max_ti_iu_len - sizeof(*srp_rsp);
1544 		if (sense_data_len > max_sense_len) {
1545 			pr_warn("truncated sense data from %d to %d"
1546 				" bytes\n", sense_data_len, max_sense_len);
1547 			sense_data_len = max_sense_len;
1548 		}
1549 
1550 		srp_rsp->flags |= SRP_RSP_FLAG_SNSVALID;
1551 		srp_rsp->sense_data_len = cpu_to_be32(sense_data_len);
1552 		memcpy(srp_rsp + 1, sense_data, sense_data_len);
1553 	}
1554 
1555 	return sizeof(*srp_rsp) + sense_data_len;
1556 }
1557 
1558 /**
1559  * srpt_build_tskmgmt_rsp() - Build a task management response.
1560  * @ch:       RDMA channel through which the request has been received.
1561  * @ioctx:    I/O context in which the SRP_RSP response will be built.
1562  * @rsp_code: RSP_CODE that will be stored in the response.
1563  * @tag:      Tag of the request for which this response is being generated.
1564  *
1565  * Returns the size in bytes of the SRP_RSP response.
1566  *
1567  * An SRP_RSP response contains a SCSI status or service response. See also
1568  * section 6.9 in the SRP r16a document for the format of an SRP_RSP
1569  * response.
1570  */
1571 static int srpt_build_tskmgmt_rsp(struct srpt_rdma_ch *ch,
1572 				  struct srpt_send_ioctx *ioctx,
1573 				  u8 rsp_code, u64 tag)
1574 {
1575 	struct srp_rsp *srp_rsp;
1576 	int resp_data_len;
1577 	int resp_len;
1578 
1579 	resp_data_len = 4;
1580 	resp_len = sizeof(*srp_rsp) + resp_data_len;
1581 
1582 	srp_rsp = ioctx->ioctx.buf;
1583 	BUG_ON(!srp_rsp);
1584 	memset(srp_rsp, 0, sizeof *srp_rsp);
1585 
1586 	srp_rsp->opcode = SRP_RSP;
1587 	srp_rsp->req_lim_delta =
1588 		cpu_to_be32(1 + atomic_xchg(&ch->req_lim_delta, 0));
1589 	srp_rsp->tag = tag;
1590 
1591 	srp_rsp->flags |= SRP_RSP_FLAG_RSPVALID;
1592 	srp_rsp->resp_data_len = cpu_to_be32(resp_data_len);
1593 	srp_rsp->data[3] = rsp_code;
1594 
1595 	return resp_len;
1596 }
1597 
1598 #define NO_SUCH_LUN ((uint64_t)-1LL)
1599 
1600 /*
1601  * SCSI LUN addressing method. See also SAM-2 and the section about
1602  * eight byte LUNs.
1603  */
1604 enum scsi_lun_addr_method {
1605 	SCSI_LUN_ADDR_METHOD_PERIPHERAL   = 0,
1606 	SCSI_LUN_ADDR_METHOD_FLAT         = 1,
1607 	SCSI_LUN_ADDR_METHOD_LUN          = 2,
1608 	SCSI_LUN_ADDR_METHOD_EXTENDED_LUN = 3,
1609 };
1610 
1611 /*
1612  * srpt_unpack_lun() - Convert from network LUN to linear LUN.
1613  *
1614  * Convert an 2-byte, 4-byte, 6-byte or 8-byte LUN structure in network byte
1615  * order (big endian) to a linear LUN. Supports three LUN addressing methods:
1616  * peripheral, flat and logical unit. See also SAM-2, section 4.9.4 (page 40).
1617  */
1618 static uint64_t srpt_unpack_lun(const uint8_t *lun, int len)
1619 {
1620 	uint64_t res = NO_SUCH_LUN;
1621 	int addressing_method;
1622 
1623 	if (unlikely(len < 2)) {
1624 		pr_err("Illegal LUN length %d, expected 2 bytes or more\n",
1625 		       len);
1626 		goto out;
1627 	}
1628 
1629 	switch (len) {
1630 	case 8:
1631 		if ((*((__be64 *)lun) &
1632 		     cpu_to_be64(0x0000FFFFFFFFFFFFLL)) != 0)
1633 			goto out_err;
1634 		break;
1635 	case 4:
1636 		if (*((__be16 *)&lun[2]) != 0)
1637 			goto out_err;
1638 		break;
1639 	case 6:
1640 		if (*((__be32 *)&lun[2]) != 0)
1641 			goto out_err;
1642 		break;
1643 	case 2:
1644 		break;
1645 	default:
1646 		goto out_err;
1647 	}
1648 
1649 	addressing_method = (*lun) >> 6; /* highest two bits of byte 0 */
1650 	switch (addressing_method) {
1651 	case SCSI_LUN_ADDR_METHOD_PERIPHERAL:
1652 	case SCSI_LUN_ADDR_METHOD_FLAT:
1653 	case SCSI_LUN_ADDR_METHOD_LUN:
1654 		res = *(lun + 1) | (((*lun) & 0x3f) << 8);
1655 		break;
1656 
1657 	case SCSI_LUN_ADDR_METHOD_EXTENDED_LUN:
1658 	default:
1659 		pr_err("Unimplemented LUN addressing method %u\n",
1660 		       addressing_method);
1661 		break;
1662 	}
1663 
1664 out:
1665 	return res;
1666 
1667 out_err:
1668 	pr_err("Support for multi-level LUNs has not yet been implemented\n");
1669 	goto out;
1670 }
1671 
1672 static int srpt_check_stop_free(struct se_cmd *cmd)
1673 {
1674 	struct srpt_send_ioctx *ioctx = container_of(cmd,
1675 				struct srpt_send_ioctx, cmd);
1676 
1677 	return target_put_sess_cmd(&ioctx->cmd);
1678 }
1679 
1680 /**
1681  * srpt_handle_cmd() - Process SRP_CMD.
1682  */
1683 static int srpt_handle_cmd(struct srpt_rdma_ch *ch,
1684 			   struct srpt_recv_ioctx *recv_ioctx,
1685 			   struct srpt_send_ioctx *send_ioctx)
1686 {
1687 	struct se_cmd *cmd;
1688 	struct srp_cmd *srp_cmd;
1689 	uint64_t unpacked_lun;
1690 	u64 data_len;
1691 	enum dma_data_direction dir;
1692 	sense_reason_t ret;
1693 	int rc;
1694 
1695 	BUG_ON(!send_ioctx);
1696 
1697 	srp_cmd = recv_ioctx->ioctx.buf;
1698 	cmd = &send_ioctx->cmd;
1699 	cmd->tag = srp_cmd->tag;
1700 
1701 	switch (srp_cmd->task_attr) {
1702 	case SRP_CMD_SIMPLE_Q:
1703 		cmd->sam_task_attr = TCM_SIMPLE_TAG;
1704 		break;
1705 	case SRP_CMD_ORDERED_Q:
1706 	default:
1707 		cmd->sam_task_attr = TCM_ORDERED_TAG;
1708 		break;
1709 	case SRP_CMD_HEAD_OF_Q:
1710 		cmd->sam_task_attr = TCM_HEAD_TAG;
1711 		break;
1712 	case SRP_CMD_ACA:
1713 		cmd->sam_task_attr = TCM_ACA_TAG;
1714 		break;
1715 	}
1716 
1717 	if (srpt_get_desc_tbl(send_ioctx, srp_cmd, &dir, &data_len)) {
1718 		pr_err("0x%llx: parsing SRP descriptor table failed.\n",
1719 		       srp_cmd->tag);
1720 		ret = TCM_INVALID_CDB_FIELD;
1721 		goto send_sense;
1722 	}
1723 
1724 	unpacked_lun = srpt_unpack_lun((uint8_t *)&srp_cmd->lun,
1725 				       sizeof(srp_cmd->lun));
1726 	rc = target_submit_cmd(cmd, ch->sess, srp_cmd->cdb,
1727 			&send_ioctx->sense_data[0], unpacked_lun, data_len,
1728 			TCM_SIMPLE_TAG, dir, TARGET_SCF_ACK_KREF);
1729 	if (rc != 0) {
1730 		ret = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1731 		goto send_sense;
1732 	}
1733 	return 0;
1734 
1735 send_sense:
1736 	transport_send_check_condition_and_sense(cmd, ret, 0);
1737 	return -1;
1738 }
1739 
1740 /**
1741  * srpt_rx_mgmt_fn_tag() - Process a task management function by tag.
1742  * @ch: RDMA channel of the task management request.
1743  * @fn: Task management function to perform.
1744  * @req_tag: Tag of the SRP task management request.
1745  * @mgmt_ioctx: I/O context of the task management request.
1746  *
1747  * Returns zero if the target core will process the task management
1748  * request asynchronously.
1749  *
1750  * Note: It is assumed that the initiator serializes tag-based task management
1751  * requests.
1752  */
1753 static int srpt_rx_mgmt_fn_tag(struct srpt_send_ioctx *ioctx, u64 tag)
1754 {
1755 	struct srpt_device *sdev;
1756 	struct srpt_rdma_ch *ch;
1757 	struct srpt_send_ioctx *target;
1758 	int ret, i;
1759 
1760 	ret = -EINVAL;
1761 	ch = ioctx->ch;
1762 	BUG_ON(!ch);
1763 	BUG_ON(!ch->sport);
1764 	sdev = ch->sport->sdev;
1765 	BUG_ON(!sdev);
1766 	spin_lock_irq(&sdev->spinlock);
1767 	for (i = 0; i < ch->rq_size; ++i) {
1768 		target = ch->ioctx_ring[i];
1769 		if (target->cmd.se_lun == ioctx->cmd.se_lun &&
1770 		    target->cmd.tag == tag &&
1771 		    srpt_get_cmd_state(target) != SRPT_STATE_DONE) {
1772 			ret = 0;
1773 			/* now let the target core abort &target->cmd; */
1774 			break;
1775 		}
1776 	}
1777 	spin_unlock_irq(&sdev->spinlock);
1778 	return ret;
1779 }
1780 
1781 static int srp_tmr_to_tcm(int fn)
1782 {
1783 	switch (fn) {
1784 	case SRP_TSK_ABORT_TASK:
1785 		return TMR_ABORT_TASK;
1786 	case SRP_TSK_ABORT_TASK_SET:
1787 		return TMR_ABORT_TASK_SET;
1788 	case SRP_TSK_CLEAR_TASK_SET:
1789 		return TMR_CLEAR_TASK_SET;
1790 	case SRP_TSK_LUN_RESET:
1791 		return TMR_LUN_RESET;
1792 	case SRP_TSK_CLEAR_ACA:
1793 		return TMR_CLEAR_ACA;
1794 	default:
1795 		return -1;
1796 	}
1797 }
1798 
1799 /**
1800  * srpt_handle_tsk_mgmt() - Process an SRP_TSK_MGMT information unit.
1801  *
1802  * Returns 0 if and only if the request will be processed by the target core.
1803  *
1804  * For more information about SRP_TSK_MGMT information units, see also section
1805  * 6.7 in the SRP r16a document.
1806  */
1807 static void srpt_handle_tsk_mgmt(struct srpt_rdma_ch *ch,
1808 				 struct srpt_recv_ioctx *recv_ioctx,
1809 				 struct srpt_send_ioctx *send_ioctx)
1810 {
1811 	struct srp_tsk_mgmt *srp_tsk;
1812 	struct se_cmd *cmd;
1813 	struct se_session *sess = ch->sess;
1814 	uint64_t unpacked_lun;
1815 	uint32_t tag = 0;
1816 	int tcm_tmr;
1817 	int rc;
1818 
1819 	BUG_ON(!send_ioctx);
1820 
1821 	srp_tsk = recv_ioctx->ioctx.buf;
1822 	cmd = &send_ioctx->cmd;
1823 
1824 	pr_debug("recv tsk_mgmt fn %d for task_tag %lld and cmd tag %lld"
1825 		 " cm_id %p sess %p\n", srp_tsk->tsk_mgmt_func,
1826 		 srp_tsk->task_tag, srp_tsk->tag, ch->cm_id, ch->sess);
1827 
1828 	srpt_set_cmd_state(send_ioctx, SRPT_STATE_MGMT);
1829 	send_ioctx->cmd.tag = srp_tsk->tag;
1830 	tcm_tmr = srp_tmr_to_tcm(srp_tsk->tsk_mgmt_func);
1831 	if (tcm_tmr < 0) {
1832 		send_ioctx->cmd.se_tmr_req->response =
1833 			TMR_TASK_MGMT_FUNCTION_NOT_SUPPORTED;
1834 		goto fail;
1835 	}
1836 	unpacked_lun = srpt_unpack_lun((uint8_t *)&srp_tsk->lun,
1837 				       sizeof(srp_tsk->lun));
1838 
1839 	if (srp_tsk->tsk_mgmt_func == SRP_TSK_ABORT_TASK) {
1840 		rc = srpt_rx_mgmt_fn_tag(send_ioctx, srp_tsk->task_tag);
1841 		if (rc < 0) {
1842 			send_ioctx->cmd.se_tmr_req->response =
1843 					TMR_TASK_DOES_NOT_EXIST;
1844 			goto fail;
1845 		}
1846 		tag = srp_tsk->task_tag;
1847 	}
1848 	rc = target_submit_tmr(&send_ioctx->cmd, sess, NULL, unpacked_lun,
1849 				srp_tsk, tcm_tmr, GFP_KERNEL, tag,
1850 				TARGET_SCF_ACK_KREF);
1851 	if (rc != 0) {
1852 		send_ioctx->cmd.se_tmr_req->response = TMR_FUNCTION_REJECTED;
1853 		goto fail;
1854 	}
1855 	return;
1856 fail:
1857 	transport_send_check_condition_and_sense(cmd, 0, 0); // XXX:
1858 }
1859 
1860 /**
1861  * srpt_handle_new_iu() - Process a newly received information unit.
1862  * @ch:    RDMA channel through which the information unit has been received.
1863  * @ioctx: SRPT I/O context associated with the information unit.
1864  */
1865 static void srpt_handle_new_iu(struct srpt_rdma_ch *ch,
1866 			       struct srpt_recv_ioctx *recv_ioctx,
1867 			       struct srpt_send_ioctx *send_ioctx)
1868 {
1869 	struct srp_cmd *srp_cmd;
1870 	enum rdma_ch_state ch_state;
1871 
1872 	BUG_ON(!ch);
1873 	BUG_ON(!recv_ioctx);
1874 
1875 	ib_dma_sync_single_for_cpu(ch->sport->sdev->device,
1876 				   recv_ioctx->ioctx.dma, srp_max_req_size,
1877 				   DMA_FROM_DEVICE);
1878 
1879 	ch_state = srpt_get_ch_state(ch);
1880 	if (unlikely(ch_state == CH_CONNECTING)) {
1881 		list_add_tail(&recv_ioctx->wait_list, &ch->cmd_wait_list);
1882 		goto out;
1883 	}
1884 
1885 	if (unlikely(ch_state != CH_LIVE))
1886 		goto out;
1887 
1888 	srp_cmd = recv_ioctx->ioctx.buf;
1889 	if (srp_cmd->opcode == SRP_CMD || srp_cmd->opcode == SRP_TSK_MGMT) {
1890 		if (!send_ioctx)
1891 			send_ioctx = srpt_get_send_ioctx(ch);
1892 		if (unlikely(!send_ioctx)) {
1893 			list_add_tail(&recv_ioctx->wait_list,
1894 				      &ch->cmd_wait_list);
1895 			goto out;
1896 		}
1897 	}
1898 
1899 	switch (srp_cmd->opcode) {
1900 	case SRP_CMD:
1901 		srpt_handle_cmd(ch, recv_ioctx, send_ioctx);
1902 		break;
1903 	case SRP_TSK_MGMT:
1904 		srpt_handle_tsk_mgmt(ch, recv_ioctx, send_ioctx);
1905 		break;
1906 	case SRP_I_LOGOUT:
1907 		pr_err("Not yet implemented: SRP_I_LOGOUT\n");
1908 		break;
1909 	case SRP_CRED_RSP:
1910 		pr_debug("received SRP_CRED_RSP\n");
1911 		break;
1912 	case SRP_AER_RSP:
1913 		pr_debug("received SRP_AER_RSP\n");
1914 		break;
1915 	case SRP_RSP:
1916 		pr_err("Received SRP_RSP\n");
1917 		break;
1918 	default:
1919 		pr_err("received IU with unknown opcode 0x%x\n",
1920 		       srp_cmd->opcode);
1921 		break;
1922 	}
1923 
1924 	srpt_post_recv(ch->sport->sdev, recv_ioctx);
1925 out:
1926 	return;
1927 }
1928 
1929 static void srpt_process_rcv_completion(struct ib_cq *cq,
1930 					struct srpt_rdma_ch *ch,
1931 					struct ib_wc *wc)
1932 {
1933 	struct srpt_device *sdev = ch->sport->sdev;
1934 	struct srpt_recv_ioctx *ioctx;
1935 	u32 index;
1936 
1937 	index = idx_from_wr_id(wc->wr_id);
1938 	if (wc->status == IB_WC_SUCCESS) {
1939 		int req_lim;
1940 
1941 		req_lim = atomic_dec_return(&ch->req_lim);
1942 		if (unlikely(req_lim < 0))
1943 			pr_err("req_lim = %d < 0\n", req_lim);
1944 		ioctx = sdev->ioctx_ring[index];
1945 		srpt_handle_new_iu(ch, ioctx, NULL);
1946 	} else {
1947 		pr_info("receiving failed for idx %u with status %d\n",
1948 			index, wc->status);
1949 	}
1950 }
1951 
1952 /**
1953  * srpt_process_send_completion() - Process an IB send completion.
1954  *
1955  * Note: Although this has not yet been observed during tests, at least in
1956  * theory it is possible that the srpt_get_send_ioctx() call invoked by
1957  * srpt_handle_new_iu() fails. This is possible because the req_lim_delta
1958  * value in each response is set to one, and it is possible that this response
1959  * makes the initiator send a new request before the send completion for that
1960  * response has been processed. This could e.g. happen if the call to
1961  * srpt_put_send_iotcx() is delayed because of a higher priority interrupt or
1962  * if IB retransmission causes generation of the send completion to be
1963  * delayed. Incoming information units for which srpt_get_send_ioctx() fails
1964  * are queued on cmd_wait_list. The code below processes these delayed
1965  * requests one at a time.
1966  */
1967 static void srpt_process_send_completion(struct ib_cq *cq,
1968 					 struct srpt_rdma_ch *ch,
1969 					 struct ib_wc *wc)
1970 {
1971 	struct srpt_send_ioctx *send_ioctx;
1972 	uint32_t index;
1973 	enum srpt_opcode opcode;
1974 
1975 	index = idx_from_wr_id(wc->wr_id);
1976 	opcode = opcode_from_wr_id(wc->wr_id);
1977 	send_ioctx = ch->ioctx_ring[index];
1978 	if (wc->status == IB_WC_SUCCESS) {
1979 		if (opcode == SRPT_SEND)
1980 			srpt_handle_send_comp(ch, send_ioctx);
1981 		else {
1982 			WARN_ON(opcode != SRPT_RDMA_ABORT &&
1983 				wc->opcode != IB_WC_RDMA_READ);
1984 			srpt_handle_rdma_comp(ch, send_ioctx, opcode);
1985 		}
1986 	} else {
1987 		if (opcode == SRPT_SEND) {
1988 			pr_info("sending response for idx %u failed"
1989 				" with status %d\n", index, wc->status);
1990 			srpt_handle_send_err_comp(ch, wc->wr_id);
1991 		} else if (opcode != SRPT_RDMA_MID) {
1992 			pr_info("RDMA t %d for idx %u failed with"
1993 				" status %d\n", opcode, index, wc->status);
1994 			srpt_handle_rdma_err_comp(ch, send_ioctx, opcode);
1995 		}
1996 	}
1997 
1998 	while (unlikely(opcode == SRPT_SEND
1999 			&& !list_empty(&ch->cmd_wait_list)
2000 			&& srpt_get_ch_state(ch) == CH_LIVE
2001 			&& (send_ioctx = srpt_get_send_ioctx(ch)) != NULL)) {
2002 		struct srpt_recv_ioctx *recv_ioctx;
2003 
2004 		recv_ioctx = list_first_entry(&ch->cmd_wait_list,
2005 					      struct srpt_recv_ioctx,
2006 					      wait_list);
2007 		list_del(&recv_ioctx->wait_list);
2008 		srpt_handle_new_iu(ch, recv_ioctx, send_ioctx);
2009 	}
2010 }
2011 
2012 static void srpt_process_completion(struct ib_cq *cq, struct srpt_rdma_ch *ch)
2013 {
2014 	struct ib_wc *const wc = ch->wc;
2015 	int i, n;
2016 
2017 	WARN_ON(cq != ch->cq);
2018 
2019 	ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
2020 	while ((n = ib_poll_cq(cq, ARRAY_SIZE(ch->wc), wc)) > 0) {
2021 		for (i = 0; i < n; i++) {
2022 			if (opcode_from_wr_id(wc[i].wr_id) == SRPT_RECV)
2023 				srpt_process_rcv_completion(cq, ch, &wc[i]);
2024 			else
2025 				srpt_process_send_completion(cq, ch, &wc[i]);
2026 		}
2027 	}
2028 }
2029 
2030 /**
2031  * srpt_completion() - IB completion queue callback function.
2032  *
2033  * Notes:
2034  * - It is guaranteed that a completion handler will never be invoked
2035  *   concurrently on two different CPUs for the same completion queue. See also
2036  *   Documentation/infiniband/core_locking.txt and the implementation of
2037  *   handle_edge_irq() in kernel/irq/chip.c.
2038  * - When threaded IRQs are enabled, completion handlers are invoked in thread
2039  *   context instead of interrupt context.
2040  */
2041 static void srpt_completion(struct ib_cq *cq, void *ctx)
2042 {
2043 	struct srpt_rdma_ch *ch = ctx;
2044 
2045 	wake_up_interruptible(&ch->wait_queue);
2046 }
2047 
2048 static int srpt_compl_thread(void *arg)
2049 {
2050 	struct srpt_rdma_ch *ch;
2051 
2052 	/* Hibernation / freezing of the SRPT kernel thread is not supported. */
2053 	current->flags |= PF_NOFREEZE;
2054 
2055 	ch = arg;
2056 	BUG_ON(!ch);
2057 	pr_info("Session %s: kernel thread %s (PID %d) started\n",
2058 		ch->sess_name, ch->thread->comm, current->pid);
2059 	while (!kthread_should_stop()) {
2060 		wait_event_interruptible(ch->wait_queue,
2061 			(srpt_process_completion(ch->cq, ch),
2062 			 kthread_should_stop()));
2063 	}
2064 	pr_info("Session %s: kernel thread %s (PID %d) stopped\n",
2065 		ch->sess_name, ch->thread->comm, current->pid);
2066 	return 0;
2067 }
2068 
2069 /**
2070  * srpt_create_ch_ib() - Create receive and send completion queues.
2071  */
2072 static int srpt_create_ch_ib(struct srpt_rdma_ch *ch)
2073 {
2074 	struct ib_qp_init_attr *qp_init;
2075 	struct srpt_port *sport = ch->sport;
2076 	struct srpt_device *sdev = sport->sdev;
2077 	u32 srp_sq_size = sport->port_attrib.srp_sq_size;
2078 	struct ib_cq_init_attr cq_attr = {};
2079 	int ret;
2080 
2081 	WARN_ON(ch->rq_size < 1);
2082 
2083 	ret = -ENOMEM;
2084 	qp_init = kzalloc(sizeof *qp_init, GFP_KERNEL);
2085 	if (!qp_init)
2086 		goto out;
2087 
2088 retry:
2089 	cq_attr.cqe = ch->rq_size + srp_sq_size;
2090 	ch->cq = ib_create_cq(sdev->device, srpt_completion, NULL, ch,
2091 			      &cq_attr);
2092 	if (IS_ERR(ch->cq)) {
2093 		ret = PTR_ERR(ch->cq);
2094 		pr_err("failed to create CQ cqe= %d ret= %d\n",
2095 		       ch->rq_size + srp_sq_size, ret);
2096 		goto out;
2097 	}
2098 
2099 	qp_init->qp_context = (void *)ch;
2100 	qp_init->event_handler
2101 		= (void(*)(struct ib_event *, void*))srpt_qp_event;
2102 	qp_init->send_cq = ch->cq;
2103 	qp_init->recv_cq = ch->cq;
2104 	qp_init->srq = sdev->srq;
2105 	qp_init->sq_sig_type = IB_SIGNAL_REQ_WR;
2106 	qp_init->qp_type = IB_QPT_RC;
2107 	qp_init->cap.max_send_wr = srp_sq_size;
2108 	qp_init->cap.max_send_sge = SRPT_DEF_SG_PER_WQE;
2109 
2110 	ch->qp = ib_create_qp(sdev->pd, qp_init);
2111 	if (IS_ERR(ch->qp)) {
2112 		ret = PTR_ERR(ch->qp);
2113 		if (ret == -ENOMEM) {
2114 			srp_sq_size /= 2;
2115 			if (srp_sq_size >= MIN_SRPT_SQ_SIZE) {
2116 				ib_destroy_cq(ch->cq);
2117 				goto retry;
2118 			}
2119 		}
2120 		pr_err("failed to create_qp ret= %d\n", ret);
2121 		goto err_destroy_cq;
2122 	}
2123 
2124 	atomic_set(&ch->sq_wr_avail, qp_init->cap.max_send_wr);
2125 
2126 	pr_debug("%s: max_cqe= %d max_sge= %d sq_size = %d cm_id= %p\n",
2127 		 __func__, ch->cq->cqe, qp_init->cap.max_send_sge,
2128 		 qp_init->cap.max_send_wr, ch->cm_id);
2129 
2130 	ret = srpt_init_ch_qp(ch, ch->qp);
2131 	if (ret)
2132 		goto err_destroy_qp;
2133 
2134 	init_waitqueue_head(&ch->wait_queue);
2135 
2136 	pr_debug("creating thread for session %s\n", ch->sess_name);
2137 
2138 	ch->thread = kthread_run(srpt_compl_thread, ch, "ib_srpt_compl");
2139 	if (IS_ERR(ch->thread)) {
2140 		pr_err("failed to create kernel thread %ld\n",
2141 		       PTR_ERR(ch->thread));
2142 		ch->thread = NULL;
2143 		goto err_destroy_qp;
2144 	}
2145 
2146 out:
2147 	kfree(qp_init);
2148 	return ret;
2149 
2150 err_destroy_qp:
2151 	ib_destroy_qp(ch->qp);
2152 err_destroy_cq:
2153 	ib_destroy_cq(ch->cq);
2154 	goto out;
2155 }
2156 
2157 static void srpt_destroy_ch_ib(struct srpt_rdma_ch *ch)
2158 {
2159 	if (ch->thread)
2160 		kthread_stop(ch->thread);
2161 
2162 	ib_destroy_qp(ch->qp);
2163 	ib_destroy_cq(ch->cq);
2164 }
2165 
2166 /**
2167  * __srpt_close_ch() - Close an RDMA channel by setting the QP error state.
2168  *
2169  * Reset the QP and make sure all resources associated with the channel will
2170  * be deallocated at an appropriate time.
2171  *
2172  * Note: The caller must hold ch->sport->sdev->spinlock.
2173  */
2174 static void __srpt_close_ch(struct srpt_rdma_ch *ch)
2175 {
2176 	enum rdma_ch_state prev_state;
2177 	unsigned long flags;
2178 
2179 	spin_lock_irqsave(&ch->spinlock, flags);
2180 	prev_state = ch->state;
2181 	switch (prev_state) {
2182 	case CH_CONNECTING:
2183 	case CH_LIVE:
2184 		ch->state = CH_DISCONNECTING;
2185 		break;
2186 	default:
2187 		break;
2188 	}
2189 	spin_unlock_irqrestore(&ch->spinlock, flags);
2190 
2191 	switch (prev_state) {
2192 	case CH_CONNECTING:
2193 		ib_send_cm_rej(ch->cm_id, IB_CM_REJ_NO_RESOURCES, NULL, 0,
2194 			       NULL, 0);
2195 		/* fall through */
2196 	case CH_LIVE:
2197 		if (ib_send_cm_dreq(ch->cm_id, NULL, 0) < 0)
2198 			pr_err("sending CM DREQ failed.\n");
2199 		break;
2200 	case CH_DISCONNECTING:
2201 		break;
2202 	case CH_DRAINING:
2203 	case CH_RELEASING:
2204 		break;
2205 	}
2206 }
2207 
2208 /**
2209  * srpt_close_ch() - Close an RDMA channel.
2210  */
2211 static void srpt_close_ch(struct srpt_rdma_ch *ch)
2212 {
2213 	struct srpt_device *sdev;
2214 
2215 	sdev = ch->sport->sdev;
2216 	spin_lock_irq(&sdev->spinlock);
2217 	__srpt_close_ch(ch);
2218 	spin_unlock_irq(&sdev->spinlock);
2219 }
2220 
2221 /**
2222  * srpt_shutdown_session() - Whether or not a session may be shut down.
2223  */
2224 static int srpt_shutdown_session(struct se_session *se_sess)
2225 {
2226 	struct srpt_rdma_ch *ch = se_sess->fabric_sess_ptr;
2227 	unsigned long flags;
2228 
2229 	spin_lock_irqsave(&ch->spinlock, flags);
2230 	if (ch->in_shutdown) {
2231 		spin_unlock_irqrestore(&ch->spinlock, flags);
2232 		return true;
2233 	}
2234 
2235 	ch->in_shutdown = true;
2236 	target_sess_cmd_list_set_waiting(se_sess);
2237 	spin_unlock_irqrestore(&ch->spinlock, flags);
2238 
2239 	return true;
2240 }
2241 
2242 /**
2243  * srpt_drain_channel() - Drain a channel by resetting the IB queue pair.
2244  * @cm_id: Pointer to the CM ID of the channel to be drained.
2245  *
2246  * Note: Must be called from inside srpt_cm_handler to avoid a race between
2247  * accessing sdev->spinlock and the call to kfree(sdev) in srpt_remove_one()
2248  * (the caller of srpt_cm_handler holds the cm_id spinlock; srpt_remove_one()
2249  * waits until all target sessions for the associated IB device have been
2250  * unregistered and target session registration involves a call to
2251  * ib_destroy_cm_id(), which locks the cm_id spinlock and hence waits until
2252  * this function has finished).
2253  */
2254 static void srpt_drain_channel(struct ib_cm_id *cm_id)
2255 {
2256 	struct srpt_device *sdev;
2257 	struct srpt_rdma_ch *ch;
2258 	int ret;
2259 	bool do_reset = false;
2260 
2261 	WARN_ON_ONCE(irqs_disabled());
2262 
2263 	sdev = cm_id->context;
2264 	BUG_ON(!sdev);
2265 	spin_lock_irq(&sdev->spinlock);
2266 	list_for_each_entry(ch, &sdev->rch_list, list) {
2267 		if (ch->cm_id == cm_id) {
2268 			do_reset = srpt_test_and_set_ch_state(ch,
2269 					CH_CONNECTING, CH_DRAINING) ||
2270 				   srpt_test_and_set_ch_state(ch,
2271 					CH_LIVE, CH_DRAINING) ||
2272 				   srpt_test_and_set_ch_state(ch,
2273 					CH_DISCONNECTING, CH_DRAINING);
2274 			break;
2275 		}
2276 	}
2277 	spin_unlock_irq(&sdev->spinlock);
2278 
2279 	if (do_reset) {
2280 		if (ch->sess)
2281 			srpt_shutdown_session(ch->sess);
2282 
2283 		ret = srpt_ch_qp_err(ch);
2284 		if (ret < 0)
2285 			pr_err("Setting queue pair in error state"
2286 			       " failed: %d\n", ret);
2287 	}
2288 }
2289 
2290 /**
2291  * srpt_find_channel() - Look up an RDMA channel.
2292  * @cm_id: Pointer to the CM ID of the channel to be looked up.
2293  *
2294  * Return NULL if no matching RDMA channel has been found.
2295  */
2296 static struct srpt_rdma_ch *srpt_find_channel(struct srpt_device *sdev,
2297 					      struct ib_cm_id *cm_id)
2298 {
2299 	struct srpt_rdma_ch *ch;
2300 	bool found;
2301 
2302 	WARN_ON_ONCE(irqs_disabled());
2303 	BUG_ON(!sdev);
2304 
2305 	found = false;
2306 	spin_lock_irq(&sdev->spinlock);
2307 	list_for_each_entry(ch, &sdev->rch_list, list) {
2308 		if (ch->cm_id == cm_id) {
2309 			found = true;
2310 			break;
2311 		}
2312 	}
2313 	spin_unlock_irq(&sdev->spinlock);
2314 
2315 	return found ? ch : NULL;
2316 }
2317 
2318 /**
2319  * srpt_release_channel() - Release channel resources.
2320  *
2321  * Schedules the actual release because:
2322  * - Calling the ib_destroy_cm_id() call from inside an IB CM callback would
2323  *   trigger a deadlock.
2324  * - It is not safe to call TCM transport_* functions from interrupt context.
2325  */
2326 static void srpt_release_channel(struct srpt_rdma_ch *ch)
2327 {
2328 	schedule_work(&ch->release_work);
2329 }
2330 
2331 static void srpt_release_channel_work(struct work_struct *w)
2332 {
2333 	struct srpt_rdma_ch *ch;
2334 	struct srpt_device *sdev;
2335 	struct se_session *se_sess;
2336 
2337 	ch = container_of(w, struct srpt_rdma_ch, release_work);
2338 	pr_debug("ch = %p; ch->sess = %p; release_done = %p\n", ch, ch->sess,
2339 		 ch->release_done);
2340 
2341 	sdev = ch->sport->sdev;
2342 	BUG_ON(!sdev);
2343 
2344 	se_sess = ch->sess;
2345 	BUG_ON(!se_sess);
2346 
2347 	target_wait_for_sess_cmds(se_sess);
2348 
2349 	transport_deregister_session_configfs(se_sess);
2350 	transport_deregister_session(se_sess);
2351 	ch->sess = NULL;
2352 
2353 	ib_destroy_cm_id(ch->cm_id);
2354 
2355 	srpt_destroy_ch_ib(ch);
2356 
2357 	srpt_free_ioctx_ring((struct srpt_ioctx **)ch->ioctx_ring,
2358 			     ch->sport->sdev, ch->rq_size,
2359 			     ch->rsp_size, DMA_TO_DEVICE);
2360 
2361 	spin_lock_irq(&sdev->spinlock);
2362 	list_del(&ch->list);
2363 	spin_unlock_irq(&sdev->spinlock);
2364 
2365 	if (ch->release_done)
2366 		complete(ch->release_done);
2367 
2368 	wake_up(&sdev->ch_releaseQ);
2369 
2370 	kfree(ch);
2371 }
2372 
2373 static struct srpt_node_acl *__srpt_lookup_acl(struct srpt_port *sport,
2374 					       u8 i_port_id[16])
2375 {
2376 	struct srpt_node_acl *nacl;
2377 
2378 	list_for_each_entry(nacl, &sport->port_acl_list, list)
2379 		if (memcmp(nacl->i_port_id, i_port_id,
2380 			   sizeof(nacl->i_port_id)) == 0)
2381 			return nacl;
2382 
2383 	return NULL;
2384 }
2385 
2386 static struct srpt_node_acl *srpt_lookup_acl(struct srpt_port *sport,
2387 					     u8 i_port_id[16])
2388 {
2389 	struct srpt_node_acl *nacl;
2390 
2391 	spin_lock_irq(&sport->port_acl_lock);
2392 	nacl = __srpt_lookup_acl(sport, i_port_id);
2393 	spin_unlock_irq(&sport->port_acl_lock);
2394 
2395 	return nacl;
2396 }
2397 
2398 /**
2399  * srpt_cm_req_recv() - Process the event IB_CM_REQ_RECEIVED.
2400  *
2401  * Ownership of the cm_id is transferred to the target session if this
2402  * functions returns zero. Otherwise the caller remains the owner of cm_id.
2403  */
2404 static int srpt_cm_req_recv(struct ib_cm_id *cm_id,
2405 			    struct ib_cm_req_event_param *param,
2406 			    void *private_data)
2407 {
2408 	struct srpt_device *sdev = cm_id->context;
2409 	struct srpt_port *sport = &sdev->port[param->port - 1];
2410 	struct srp_login_req *req;
2411 	struct srp_login_rsp *rsp;
2412 	struct srp_login_rej *rej;
2413 	struct ib_cm_rep_param *rep_param;
2414 	struct srpt_rdma_ch *ch, *tmp_ch;
2415 	struct srpt_node_acl *nacl;
2416 	u32 it_iu_len;
2417 	int i;
2418 	int ret = 0;
2419 
2420 	WARN_ON_ONCE(irqs_disabled());
2421 
2422 	if (WARN_ON(!sdev || !private_data))
2423 		return -EINVAL;
2424 
2425 	req = (struct srp_login_req *)private_data;
2426 
2427 	it_iu_len = be32_to_cpu(req->req_it_iu_len);
2428 
2429 	pr_info("Received SRP_LOGIN_REQ with i_port_id 0x%llx:0x%llx,"
2430 		" t_port_id 0x%llx:0x%llx and it_iu_len %d on port %d"
2431 		" (guid=0x%llx:0x%llx)\n",
2432 		be64_to_cpu(*(__be64 *)&req->initiator_port_id[0]),
2433 		be64_to_cpu(*(__be64 *)&req->initiator_port_id[8]),
2434 		be64_to_cpu(*(__be64 *)&req->target_port_id[0]),
2435 		be64_to_cpu(*(__be64 *)&req->target_port_id[8]),
2436 		it_iu_len,
2437 		param->port,
2438 		be64_to_cpu(*(__be64 *)&sdev->port[param->port - 1].gid.raw[0]),
2439 		be64_to_cpu(*(__be64 *)&sdev->port[param->port - 1].gid.raw[8]));
2440 
2441 	rsp = kzalloc(sizeof *rsp, GFP_KERNEL);
2442 	rej = kzalloc(sizeof *rej, GFP_KERNEL);
2443 	rep_param = kzalloc(sizeof *rep_param, GFP_KERNEL);
2444 
2445 	if (!rsp || !rej || !rep_param) {
2446 		ret = -ENOMEM;
2447 		goto out;
2448 	}
2449 
2450 	if (it_iu_len > srp_max_req_size || it_iu_len < 64) {
2451 		rej->reason = cpu_to_be32(
2452 			      SRP_LOGIN_REJ_REQ_IT_IU_LENGTH_TOO_LARGE);
2453 		ret = -EINVAL;
2454 		pr_err("rejected SRP_LOGIN_REQ because its"
2455 		       " length (%d bytes) is out of range (%d .. %d)\n",
2456 		       it_iu_len, 64, srp_max_req_size);
2457 		goto reject;
2458 	}
2459 
2460 	if (!sport->enabled) {
2461 		rej->reason = cpu_to_be32(
2462 			      SRP_LOGIN_REJ_INSUFFICIENT_RESOURCES);
2463 		ret = -EINVAL;
2464 		pr_err("rejected SRP_LOGIN_REQ because the target port"
2465 		       " has not yet been enabled\n");
2466 		goto reject;
2467 	}
2468 
2469 	if ((req->req_flags & SRP_MTCH_ACTION) == SRP_MULTICHAN_SINGLE) {
2470 		rsp->rsp_flags = SRP_LOGIN_RSP_MULTICHAN_NO_CHAN;
2471 
2472 		spin_lock_irq(&sdev->spinlock);
2473 
2474 		list_for_each_entry_safe(ch, tmp_ch, &sdev->rch_list, list) {
2475 			if (!memcmp(ch->i_port_id, req->initiator_port_id, 16)
2476 			    && !memcmp(ch->t_port_id, req->target_port_id, 16)
2477 			    && param->port == ch->sport->port
2478 			    && param->listen_id == ch->sport->sdev->cm_id
2479 			    && ch->cm_id) {
2480 				enum rdma_ch_state ch_state;
2481 
2482 				ch_state = srpt_get_ch_state(ch);
2483 				if (ch_state != CH_CONNECTING
2484 				    && ch_state != CH_LIVE)
2485 					continue;
2486 
2487 				/* found an existing channel */
2488 				pr_debug("Found existing channel %s"
2489 					 " cm_id= %p state= %d\n",
2490 					 ch->sess_name, ch->cm_id, ch_state);
2491 
2492 				__srpt_close_ch(ch);
2493 
2494 				rsp->rsp_flags =
2495 					SRP_LOGIN_RSP_MULTICHAN_TERMINATED;
2496 			}
2497 		}
2498 
2499 		spin_unlock_irq(&sdev->spinlock);
2500 
2501 	} else
2502 		rsp->rsp_flags = SRP_LOGIN_RSP_MULTICHAN_MAINTAINED;
2503 
2504 	if (*(__be64 *)req->target_port_id != cpu_to_be64(srpt_service_guid)
2505 	    || *(__be64 *)(req->target_port_id + 8) !=
2506 	       cpu_to_be64(srpt_service_guid)) {
2507 		rej->reason = cpu_to_be32(
2508 			      SRP_LOGIN_REJ_UNABLE_ASSOCIATE_CHANNEL);
2509 		ret = -ENOMEM;
2510 		pr_err("rejected SRP_LOGIN_REQ because it"
2511 		       " has an invalid target port identifier.\n");
2512 		goto reject;
2513 	}
2514 
2515 	ch = kzalloc(sizeof *ch, GFP_KERNEL);
2516 	if (!ch) {
2517 		rej->reason = cpu_to_be32(
2518 			      SRP_LOGIN_REJ_INSUFFICIENT_RESOURCES);
2519 		pr_err("rejected SRP_LOGIN_REQ because no memory.\n");
2520 		ret = -ENOMEM;
2521 		goto reject;
2522 	}
2523 
2524 	INIT_WORK(&ch->release_work, srpt_release_channel_work);
2525 	memcpy(ch->i_port_id, req->initiator_port_id, 16);
2526 	memcpy(ch->t_port_id, req->target_port_id, 16);
2527 	ch->sport = &sdev->port[param->port - 1];
2528 	ch->cm_id = cm_id;
2529 	/*
2530 	 * Avoid QUEUE_FULL conditions by limiting the number of buffers used
2531 	 * for the SRP protocol to the command queue size.
2532 	 */
2533 	ch->rq_size = SRPT_RQ_SIZE;
2534 	spin_lock_init(&ch->spinlock);
2535 	ch->state = CH_CONNECTING;
2536 	INIT_LIST_HEAD(&ch->cmd_wait_list);
2537 	ch->rsp_size = ch->sport->port_attrib.srp_max_rsp_size;
2538 
2539 	ch->ioctx_ring = (struct srpt_send_ioctx **)
2540 		srpt_alloc_ioctx_ring(ch->sport->sdev, ch->rq_size,
2541 				      sizeof(*ch->ioctx_ring[0]),
2542 				      ch->rsp_size, DMA_TO_DEVICE);
2543 	if (!ch->ioctx_ring)
2544 		goto free_ch;
2545 
2546 	INIT_LIST_HEAD(&ch->free_list);
2547 	for (i = 0; i < ch->rq_size; i++) {
2548 		ch->ioctx_ring[i]->ch = ch;
2549 		list_add_tail(&ch->ioctx_ring[i]->free_list, &ch->free_list);
2550 	}
2551 
2552 	ret = srpt_create_ch_ib(ch);
2553 	if (ret) {
2554 		rej->reason = cpu_to_be32(
2555 			      SRP_LOGIN_REJ_INSUFFICIENT_RESOURCES);
2556 		pr_err("rejected SRP_LOGIN_REQ because creating"
2557 		       " a new RDMA channel failed.\n");
2558 		goto free_ring;
2559 	}
2560 
2561 	ret = srpt_ch_qp_rtr(ch, ch->qp);
2562 	if (ret) {
2563 		rej->reason = cpu_to_be32(SRP_LOGIN_REJ_INSUFFICIENT_RESOURCES);
2564 		pr_err("rejected SRP_LOGIN_REQ because enabling"
2565 		       " RTR failed (error code = %d)\n", ret);
2566 		goto destroy_ib;
2567 	}
2568 	/*
2569 	 * Use the initator port identifier as the session name.
2570 	 */
2571 	snprintf(ch->sess_name, sizeof(ch->sess_name), "0x%016llx%016llx",
2572 			be64_to_cpu(*(__be64 *)ch->i_port_id),
2573 			be64_to_cpu(*(__be64 *)(ch->i_port_id + 8)));
2574 
2575 	pr_debug("registering session %s\n", ch->sess_name);
2576 
2577 	nacl = srpt_lookup_acl(sport, ch->i_port_id);
2578 	if (!nacl) {
2579 		pr_info("Rejected login because no ACL has been"
2580 			" configured yet for initiator %s.\n", ch->sess_name);
2581 		rej->reason = cpu_to_be32(
2582 			      SRP_LOGIN_REJ_CHANNEL_LIMIT_REACHED);
2583 		goto destroy_ib;
2584 	}
2585 
2586 	ch->sess = transport_init_session(TARGET_PROT_NORMAL);
2587 	if (IS_ERR(ch->sess)) {
2588 		rej->reason = cpu_to_be32(
2589 			      SRP_LOGIN_REJ_INSUFFICIENT_RESOURCES);
2590 		pr_debug("Failed to create session\n");
2591 		goto deregister_session;
2592 	}
2593 	ch->sess->se_node_acl = &nacl->nacl;
2594 	transport_register_session(&sport->port_tpg_1, &nacl->nacl, ch->sess, ch);
2595 
2596 	pr_debug("Establish connection sess=%p name=%s cm_id=%p\n", ch->sess,
2597 		 ch->sess_name, ch->cm_id);
2598 
2599 	/* create srp_login_response */
2600 	rsp->opcode = SRP_LOGIN_RSP;
2601 	rsp->tag = req->tag;
2602 	rsp->max_it_iu_len = req->req_it_iu_len;
2603 	rsp->max_ti_iu_len = req->req_it_iu_len;
2604 	ch->max_ti_iu_len = it_iu_len;
2605 	rsp->buf_fmt = cpu_to_be16(SRP_BUF_FORMAT_DIRECT
2606 				   | SRP_BUF_FORMAT_INDIRECT);
2607 	rsp->req_lim_delta = cpu_to_be32(ch->rq_size);
2608 	atomic_set(&ch->req_lim, ch->rq_size);
2609 	atomic_set(&ch->req_lim_delta, 0);
2610 
2611 	/* create cm reply */
2612 	rep_param->qp_num = ch->qp->qp_num;
2613 	rep_param->private_data = (void *)rsp;
2614 	rep_param->private_data_len = sizeof *rsp;
2615 	rep_param->rnr_retry_count = 7;
2616 	rep_param->flow_control = 1;
2617 	rep_param->failover_accepted = 0;
2618 	rep_param->srq = 1;
2619 	rep_param->responder_resources = 4;
2620 	rep_param->initiator_depth = 4;
2621 
2622 	ret = ib_send_cm_rep(cm_id, rep_param);
2623 	if (ret) {
2624 		pr_err("sending SRP_LOGIN_REQ response failed"
2625 		       " (error code = %d)\n", ret);
2626 		goto release_channel;
2627 	}
2628 
2629 	spin_lock_irq(&sdev->spinlock);
2630 	list_add_tail(&ch->list, &sdev->rch_list);
2631 	spin_unlock_irq(&sdev->spinlock);
2632 
2633 	goto out;
2634 
2635 release_channel:
2636 	srpt_set_ch_state(ch, CH_RELEASING);
2637 	transport_deregister_session_configfs(ch->sess);
2638 
2639 deregister_session:
2640 	transport_deregister_session(ch->sess);
2641 	ch->sess = NULL;
2642 
2643 destroy_ib:
2644 	srpt_destroy_ch_ib(ch);
2645 
2646 free_ring:
2647 	srpt_free_ioctx_ring((struct srpt_ioctx **)ch->ioctx_ring,
2648 			     ch->sport->sdev, ch->rq_size,
2649 			     ch->rsp_size, DMA_TO_DEVICE);
2650 free_ch:
2651 	kfree(ch);
2652 
2653 reject:
2654 	rej->opcode = SRP_LOGIN_REJ;
2655 	rej->tag = req->tag;
2656 	rej->buf_fmt = cpu_to_be16(SRP_BUF_FORMAT_DIRECT
2657 				   | SRP_BUF_FORMAT_INDIRECT);
2658 
2659 	ib_send_cm_rej(cm_id, IB_CM_REJ_CONSUMER_DEFINED, NULL, 0,
2660 			     (void *)rej, sizeof *rej);
2661 
2662 out:
2663 	kfree(rep_param);
2664 	kfree(rsp);
2665 	kfree(rej);
2666 
2667 	return ret;
2668 }
2669 
2670 static void srpt_cm_rej_recv(struct ib_cm_id *cm_id)
2671 {
2672 	pr_info("Received IB REJ for cm_id %p.\n", cm_id);
2673 	srpt_drain_channel(cm_id);
2674 }
2675 
2676 /**
2677  * srpt_cm_rtu_recv() - Process an IB_CM_RTU_RECEIVED or USER_ESTABLISHED event.
2678  *
2679  * An IB_CM_RTU_RECEIVED message indicates that the connection is established
2680  * and that the recipient may begin transmitting (RTU = ready to use).
2681  */
2682 static void srpt_cm_rtu_recv(struct ib_cm_id *cm_id)
2683 {
2684 	struct srpt_rdma_ch *ch;
2685 	int ret;
2686 
2687 	ch = srpt_find_channel(cm_id->context, cm_id);
2688 	BUG_ON(!ch);
2689 
2690 	if (srpt_test_and_set_ch_state(ch, CH_CONNECTING, CH_LIVE)) {
2691 		struct srpt_recv_ioctx *ioctx, *ioctx_tmp;
2692 
2693 		ret = srpt_ch_qp_rts(ch, ch->qp);
2694 
2695 		list_for_each_entry_safe(ioctx, ioctx_tmp, &ch->cmd_wait_list,
2696 					 wait_list) {
2697 			list_del(&ioctx->wait_list);
2698 			srpt_handle_new_iu(ch, ioctx, NULL);
2699 		}
2700 		if (ret)
2701 			srpt_close_ch(ch);
2702 	}
2703 }
2704 
2705 static void srpt_cm_timewait_exit(struct ib_cm_id *cm_id)
2706 {
2707 	pr_info("Received IB TimeWait exit for cm_id %p.\n", cm_id);
2708 	srpt_drain_channel(cm_id);
2709 }
2710 
2711 static void srpt_cm_rep_error(struct ib_cm_id *cm_id)
2712 {
2713 	pr_info("Received IB REP error for cm_id %p.\n", cm_id);
2714 	srpt_drain_channel(cm_id);
2715 }
2716 
2717 /**
2718  * srpt_cm_dreq_recv() - Process reception of a DREQ message.
2719  */
2720 static void srpt_cm_dreq_recv(struct ib_cm_id *cm_id)
2721 {
2722 	struct srpt_rdma_ch *ch;
2723 	unsigned long flags;
2724 	bool send_drep = false;
2725 
2726 	ch = srpt_find_channel(cm_id->context, cm_id);
2727 	BUG_ON(!ch);
2728 
2729 	pr_debug("cm_id= %p ch->state= %d\n", cm_id, srpt_get_ch_state(ch));
2730 
2731 	spin_lock_irqsave(&ch->spinlock, flags);
2732 	switch (ch->state) {
2733 	case CH_CONNECTING:
2734 	case CH_LIVE:
2735 		send_drep = true;
2736 		ch->state = CH_DISCONNECTING;
2737 		break;
2738 	case CH_DISCONNECTING:
2739 	case CH_DRAINING:
2740 	case CH_RELEASING:
2741 		WARN(true, "unexpected channel state %d\n", ch->state);
2742 		break;
2743 	}
2744 	spin_unlock_irqrestore(&ch->spinlock, flags);
2745 
2746 	if (send_drep) {
2747 		if (ib_send_cm_drep(ch->cm_id, NULL, 0) < 0)
2748 			pr_err("Sending IB DREP failed.\n");
2749 		pr_info("Received DREQ and sent DREP for session %s.\n",
2750 			ch->sess_name);
2751 	}
2752 }
2753 
2754 /**
2755  * srpt_cm_drep_recv() - Process reception of a DREP message.
2756  */
2757 static void srpt_cm_drep_recv(struct ib_cm_id *cm_id)
2758 {
2759 	pr_info("Received InfiniBand DREP message for cm_id %p.\n", cm_id);
2760 	srpt_drain_channel(cm_id);
2761 }
2762 
2763 /**
2764  * srpt_cm_handler() - IB connection manager callback function.
2765  *
2766  * A non-zero return value will cause the caller destroy the CM ID.
2767  *
2768  * Note: srpt_cm_handler() must only return a non-zero value when transferring
2769  * ownership of the cm_id to a channel by srpt_cm_req_recv() failed. Returning
2770  * a non-zero value in any other case will trigger a race with the
2771  * ib_destroy_cm_id() call in srpt_release_channel().
2772  */
2773 static int srpt_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event)
2774 {
2775 	int ret;
2776 
2777 	ret = 0;
2778 	switch (event->event) {
2779 	case IB_CM_REQ_RECEIVED:
2780 		ret = srpt_cm_req_recv(cm_id, &event->param.req_rcvd,
2781 				       event->private_data);
2782 		break;
2783 	case IB_CM_REJ_RECEIVED:
2784 		srpt_cm_rej_recv(cm_id);
2785 		break;
2786 	case IB_CM_RTU_RECEIVED:
2787 	case IB_CM_USER_ESTABLISHED:
2788 		srpt_cm_rtu_recv(cm_id);
2789 		break;
2790 	case IB_CM_DREQ_RECEIVED:
2791 		srpt_cm_dreq_recv(cm_id);
2792 		break;
2793 	case IB_CM_DREP_RECEIVED:
2794 		srpt_cm_drep_recv(cm_id);
2795 		break;
2796 	case IB_CM_TIMEWAIT_EXIT:
2797 		srpt_cm_timewait_exit(cm_id);
2798 		break;
2799 	case IB_CM_REP_ERROR:
2800 		srpt_cm_rep_error(cm_id);
2801 		break;
2802 	case IB_CM_DREQ_ERROR:
2803 		pr_info("Received IB DREQ ERROR event.\n");
2804 		break;
2805 	case IB_CM_MRA_RECEIVED:
2806 		pr_info("Received IB MRA event\n");
2807 		break;
2808 	default:
2809 		pr_err("received unrecognized IB CM event %d\n", event->event);
2810 		break;
2811 	}
2812 
2813 	return ret;
2814 }
2815 
2816 /**
2817  * srpt_perform_rdmas() - Perform IB RDMA.
2818  *
2819  * Returns zero upon success or a negative number upon failure.
2820  */
2821 static int srpt_perform_rdmas(struct srpt_rdma_ch *ch,
2822 			      struct srpt_send_ioctx *ioctx)
2823 {
2824 	struct ib_rdma_wr wr;
2825 	struct ib_send_wr *bad_wr;
2826 	struct rdma_iu *riu;
2827 	int i;
2828 	int ret;
2829 	int sq_wr_avail;
2830 	enum dma_data_direction dir;
2831 	const int n_rdma = ioctx->n_rdma;
2832 
2833 	dir = ioctx->cmd.data_direction;
2834 	if (dir == DMA_TO_DEVICE) {
2835 		/* write */
2836 		ret = -ENOMEM;
2837 		sq_wr_avail = atomic_sub_return(n_rdma, &ch->sq_wr_avail);
2838 		if (sq_wr_avail < 0) {
2839 			pr_warn("IB send queue full (needed %d)\n",
2840 				n_rdma);
2841 			goto out;
2842 		}
2843 	}
2844 
2845 	ioctx->rdma_aborted = false;
2846 	ret = 0;
2847 	riu = ioctx->rdma_ius;
2848 	memset(&wr, 0, sizeof wr);
2849 
2850 	for (i = 0; i < n_rdma; ++i, ++riu) {
2851 		if (dir == DMA_FROM_DEVICE) {
2852 			wr.wr.opcode = IB_WR_RDMA_WRITE;
2853 			wr.wr.wr_id = encode_wr_id(i == n_rdma - 1 ?
2854 						SRPT_RDMA_WRITE_LAST :
2855 						SRPT_RDMA_MID,
2856 						ioctx->ioctx.index);
2857 		} else {
2858 			wr.wr.opcode = IB_WR_RDMA_READ;
2859 			wr.wr.wr_id = encode_wr_id(i == n_rdma - 1 ?
2860 						SRPT_RDMA_READ_LAST :
2861 						SRPT_RDMA_MID,
2862 						ioctx->ioctx.index);
2863 		}
2864 		wr.wr.next = NULL;
2865 		wr.remote_addr = riu->raddr;
2866 		wr.rkey = riu->rkey;
2867 		wr.wr.num_sge = riu->sge_cnt;
2868 		wr.wr.sg_list = riu->sge;
2869 
2870 		/* only get completion event for the last rdma write */
2871 		if (i == (n_rdma - 1) && dir == DMA_TO_DEVICE)
2872 			wr.wr.send_flags = IB_SEND_SIGNALED;
2873 
2874 		ret = ib_post_send(ch->qp, &wr.wr, &bad_wr);
2875 		if (ret)
2876 			break;
2877 	}
2878 
2879 	if (ret)
2880 		pr_err("%s[%d]: ib_post_send() returned %d for %d/%d\n",
2881 				 __func__, __LINE__, ret, i, n_rdma);
2882 	if (ret && i > 0) {
2883 		wr.wr.num_sge = 0;
2884 		wr.wr.wr_id = encode_wr_id(SRPT_RDMA_ABORT, ioctx->ioctx.index);
2885 		wr.wr.send_flags = IB_SEND_SIGNALED;
2886 		while (ch->state == CH_LIVE &&
2887 			ib_post_send(ch->qp, &wr.wr, &bad_wr) != 0) {
2888 			pr_info("Trying to abort failed RDMA transfer [%d]\n",
2889 				ioctx->ioctx.index);
2890 			msleep(1000);
2891 		}
2892 		while (ch->state != CH_RELEASING && !ioctx->rdma_aborted) {
2893 			pr_info("Waiting until RDMA abort finished [%d]\n",
2894 				ioctx->ioctx.index);
2895 			msleep(1000);
2896 		}
2897 	}
2898 out:
2899 	if (unlikely(dir == DMA_TO_DEVICE && ret < 0))
2900 		atomic_add(n_rdma, &ch->sq_wr_avail);
2901 	return ret;
2902 }
2903 
2904 /**
2905  * srpt_xfer_data() - Start data transfer from initiator to target.
2906  */
2907 static int srpt_xfer_data(struct srpt_rdma_ch *ch,
2908 			  struct srpt_send_ioctx *ioctx)
2909 {
2910 	int ret;
2911 
2912 	ret = srpt_map_sg_to_ib_sge(ch, ioctx);
2913 	if (ret) {
2914 		pr_err("%s[%d] ret=%d\n", __func__, __LINE__, ret);
2915 		goto out;
2916 	}
2917 
2918 	ret = srpt_perform_rdmas(ch, ioctx);
2919 	if (ret) {
2920 		if (ret == -EAGAIN || ret == -ENOMEM)
2921 			pr_info("%s[%d] queue full -- ret=%d\n",
2922 				__func__, __LINE__, ret);
2923 		else
2924 			pr_err("%s[%d] fatal error -- ret=%d\n",
2925 			       __func__, __LINE__, ret);
2926 		goto out_unmap;
2927 	}
2928 
2929 out:
2930 	return ret;
2931 out_unmap:
2932 	srpt_unmap_sg_to_ib_sge(ch, ioctx);
2933 	goto out;
2934 }
2935 
2936 static int srpt_write_pending_status(struct se_cmd *se_cmd)
2937 {
2938 	struct srpt_send_ioctx *ioctx;
2939 
2940 	ioctx = container_of(se_cmd, struct srpt_send_ioctx, cmd);
2941 	return srpt_get_cmd_state(ioctx) == SRPT_STATE_NEED_DATA;
2942 }
2943 
2944 /*
2945  * srpt_write_pending() - Start data transfer from initiator to target (write).
2946  */
2947 static int srpt_write_pending(struct se_cmd *se_cmd)
2948 {
2949 	struct srpt_rdma_ch *ch;
2950 	struct srpt_send_ioctx *ioctx;
2951 	enum srpt_command_state new_state;
2952 	enum rdma_ch_state ch_state;
2953 	int ret;
2954 
2955 	ioctx = container_of(se_cmd, struct srpt_send_ioctx, cmd);
2956 
2957 	new_state = srpt_set_cmd_state(ioctx, SRPT_STATE_NEED_DATA);
2958 	WARN_ON(new_state == SRPT_STATE_DONE);
2959 
2960 	ch = ioctx->ch;
2961 	BUG_ON(!ch);
2962 
2963 	ch_state = srpt_get_ch_state(ch);
2964 	switch (ch_state) {
2965 	case CH_CONNECTING:
2966 		WARN(true, "unexpected channel state %d\n", ch_state);
2967 		ret = -EINVAL;
2968 		goto out;
2969 	case CH_LIVE:
2970 		break;
2971 	case CH_DISCONNECTING:
2972 	case CH_DRAINING:
2973 	case CH_RELEASING:
2974 		pr_debug("cmd with tag %lld: channel disconnecting\n",
2975 			 ioctx->cmd.tag);
2976 		srpt_set_cmd_state(ioctx, SRPT_STATE_DATA_IN);
2977 		ret = -EINVAL;
2978 		goto out;
2979 	}
2980 	ret = srpt_xfer_data(ch, ioctx);
2981 
2982 out:
2983 	return ret;
2984 }
2985 
2986 static u8 tcm_to_srp_tsk_mgmt_status(const int tcm_mgmt_status)
2987 {
2988 	switch (tcm_mgmt_status) {
2989 	case TMR_FUNCTION_COMPLETE:
2990 		return SRP_TSK_MGMT_SUCCESS;
2991 	case TMR_FUNCTION_REJECTED:
2992 		return SRP_TSK_MGMT_FUNC_NOT_SUPP;
2993 	}
2994 	return SRP_TSK_MGMT_FAILED;
2995 }
2996 
2997 /**
2998  * srpt_queue_response() - Transmits the response to a SCSI command.
2999  *
3000  * Callback function called by the TCM core. Must not block since it can be
3001  * invoked on the context of the IB completion handler.
3002  */
3003 static void srpt_queue_response(struct se_cmd *cmd)
3004 {
3005 	struct srpt_rdma_ch *ch;
3006 	struct srpt_send_ioctx *ioctx;
3007 	enum srpt_command_state state;
3008 	unsigned long flags;
3009 	int ret;
3010 	enum dma_data_direction dir;
3011 	int resp_len;
3012 	u8 srp_tm_status;
3013 
3014 	ioctx = container_of(cmd, struct srpt_send_ioctx, cmd);
3015 	ch = ioctx->ch;
3016 	BUG_ON(!ch);
3017 
3018 	spin_lock_irqsave(&ioctx->spinlock, flags);
3019 	state = ioctx->state;
3020 	switch (state) {
3021 	case SRPT_STATE_NEW:
3022 	case SRPT_STATE_DATA_IN:
3023 		ioctx->state = SRPT_STATE_CMD_RSP_SENT;
3024 		break;
3025 	case SRPT_STATE_MGMT:
3026 		ioctx->state = SRPT_STATE_MGMT_RSP_SENT;
3027 		break;
3028 	default:
3029 		WARN(true, "ch %p; cmd %d: unexpected command state %d\n",
3030 			ch, ioctx->ioctx.index, ioctx->state);
3031 		break;
3032 	}
3033 	spin_unlock_irqrestore(&ioctx->spinlock, flags);
3034 
3035 	if (unlikely(transport_check_aborted_status(&ioctx->cmd, false)
3036 		     || WARN_ON_ONCE(state == SRPT_STATE_CMD_RSP_SENT))) {
3037 		atomic_inc(&ch->req_lim_delta);
3038 		srpt_abort_cmd(ioctx);
3039 		return;
3040 	}
3041 
3042 	dir = ioctx->cmd.data_direction;
3043 
3044 	/* For read commands, transfer the data to the initiator. */
3045 	if (dir == DMA_FROM_DEVICE && ioctx->cmd.data_length &&
3046 	    !ioctx->queue_status_only) {
3047 		ret = srpt_xfer_data(ch, ioctx);
3048 		if (ret) {
3049 			pr_err("xfer_data failed for tag %llu\n",
3050 			       ioctx->cmd.tag);
3051 			return;
3052 		}
3053 	}
3054 
3055 	if (state != SRPT_STATE_MGMT)
3056 		resp_len = srpt_build_cmd_rsp(ch, ioctx, ioctx->cmd.tag,
3057 					      cmd->scsi_status);
3058 	else {
3059 		srp_tm_status
3060 			= tcm_to_srp_tsk_mgmt_status(cmd->se_tmr_req->response);
3061 		resp_len = srpt_build_tskmgmt_rsp(ch, ioctx, srp_tm_status,
3062 						 ioctx->cmd.tag);
3063 	}
3064 	ret = srpt_post_send(ch, ioctx, resp_len);
3065 	if (ret) {
3066 		pr_err("sending cmd response failed for tag %llu\n",
3067 		       ioctx->cmd.tag);
3068 		srpt_unmap_sg_to_ib_sge(ch, ioctx);
3069 		srpt_set_cmd_state(ioctx, SRPT_STATE_DONE);
3070 		target_put_sess_cmd(&ioctx->cmd);
3071 	}
3072 }
3073 
3074 static int srpt_queue_data_in(struct se_cmd *cmd)
3075 {
3076 	srpt_queue_response(cmd);
3077 	return 0;
3078 }
3079 
3080 static void srpt_queue_tm_rsp(struct se_cmd *cmd)
3081 {
3082 	srpt_queue_response(cmd);
3083 }
3084 
3085 static void srpt_aborted_task(struct se_cmd *cmd)
3086 {
3087 	struct srpt_send_ioctx *ioctx = container_of(cmd,
3088 				struct srpt_send_ioctx, cmd);
3089 
3090 	srpt_unmap_sg_to_ib_sge(ioctx->ch, ioctx);
3091 }
3092 
3093 static int srpt_queue_status(struct se_cmd *cmd)
3094 {
3095 	struct srpt_send_ioctx *ioctx;
3096 
3097 	ioctx = container_of(cmd, struct srpt_send_ioctx, cmd);
3098 	BUG_ON(ioctx->sense_data != cmd->sense_buffer);
3099 	if (cmd->se_cmd_flags &
3100 	    (SCF_TRANSPORT_TASK_SENSE | SCF_EMULATED_TASK_SENSE))
3101 		WARN_ON(cmd->scsi_status != SAM_STAT_CHECK_CONDITION);
3102 	ioctx->queue_status_only = true;
3103 	srpt_queue_response(cmd);
3104 	return 0;
3105 }
3106 
3107 static void srpt_refresh_port_work(struct work_struct *work)
3108 {
3109 	struct srpt_port *sport = container_of(work, struct srpt_port, work);
3110 
3111 	srpt_refresh_port(sport);
3112 }
3113 
3114 static int srpt_ch_list_empty(struct srpt_device *sdev)
3115 {
3116 	int res;
3117 
3118 	spin_lock_irq(&sdev->spinlock);
3119 	res = list_empty(&sdev->rch_list);
3120 	spin_unlock_irq(&sdev->spinlock);
3121 
3122 	return res;
3123 }
3124 
3125 /**
3126  * srpt_release_sdev() - Free the channel resources associated with a target.
3127  */
3128 static int srpt_release_sdev(struct srpt_device *sdev)
3129 {
3130 	struct srpt_rdma_ch *ch, *tmp_ch;
3131 	int res;
3132 
3133 	WARN_ON_ONCE(irqs_disabled());
3134 
3135 	BUG_ON(!sdev);
3136 
3137 	spin_lock_irq(&sdev->spinlock);
3138 	list_for_each_entry_safe(ch, tmp_ch, &sdev->rch_list, list)
3139 		__srpt_close_ch(ch);
3140 	spin_unlock_irq(&sdev->spinlock);
3141 
3142 	res = wait_event_interruptible(sdev->ch_releaseQ,
3143 				       srpt_ch_list_empty(sdev));
3144 	if (res)
3145 		pr_err("%s: interrupted.\n", __func__);
3146 
3147 	return 0;
3148 }
3149 
3150 static struct srpt_port *__srpt_lookup_port(const char *name)
3151 {
3152 	struct ib_device *dev;
3153 	struct srpt_device *sdev;
3154 	struct srpt_port *sport;
3155 	int i;
3156 
3157 	list_for_each_entry(sdev, &srpt_dev_list, list) {
3158 		dev = sdev->device;
3159 		if (!dev)
3160 			continue;
3161 
3162 		for (i = 0; i < dev->phys_port_cnt; i++) {
3163 			sport = &sdev->port[i];
3164 
3165 			if (!strcmp(sport->port_guid, name))
3166 				return sport;
3167 		}
3168 	}
3169 
3170 	return NULL;
3171 }
3172 
3173 static struct srpt_port *srpt_lookup_port(const char *name)
3174 {
3175 	struct srpt_port *sport;
3176 
3177 	spin_lock(&srpt_dev_lock);
3178 	sport = __srpt_lookup_port(name);
3179 	spin_unlock(&srpt_dev_lock);
3180 
3181 	return sport;
3182 }
3183 
3184 /**
3185  * srpt_add_one() - Infiniband device addition callback function.
3186  */
3187 static void srpt_add_one(struct ib_device *device)
3188 {
3189 	struct srpt_device *sdev;
3190 	struct srpt_port *sport;
3191 	struct ib_srq_init_attr srq_attr;
3192 	int i;
3193 
3194 	pr_debug("device = %p, device->dma_ops = %p\n", device,
3195 		 device->dma_ops);
3196 
3197 	sdev = kzalloc(sizeof *sdev, GFP_KERNEL);
3198 	if (!sdev)
3199 		goto err;
3200 
3201 	sdev->device = device;
3202 	INIT_LIST_HEAD(&sdev->rch_list);
3203 	init_waitqueue_head(&sdev->ch_releaseQ);
3204 	spin_lock_init(&sdev->spinlock);
3205 
3206 	if (ib_query_device(device, &sdev->dev_attr))
3207 		goto free_dev;
3208 
3209 	sdev->pd = ib_alloc_pd(device);
3210 	if (IS_ERR(sdev->pd))
3211 		goto free_dev;
3212 
3213 	sdev->srq_size = min(srpt_srq_size, sdev->dev_attr.max_srq_wr);
3214 
3215 	srq_attr.event_handler = srpt_srq_event;
3216 	srq_attr.srq_context = (void *)sdev;
3217 	srq_attr.attr.max_wr = sdev->srq_size;
3218 	srq_attr.attr.max_sge = 1;
3219 	srq_attr.attr.srq_limit = 0;
3220 	srq_attr.srq_type = IB_SRQT_BASIC;
3221 
3222 	sdev->srq = ib_create_srq(sdev->pd, &srq_attr);
3223 	if (IS_ERR(sdev->srq))
3224 		goto err_pd;
3225 
3226 	pr_debug("%s: create SRQ #wr= %d max_allow=%d dev= %s\n",
3227 		 __func__, sdev->srq_size, sdev->dev_attr.max_srq_wr,
3228 		 device->name);
3229 
3230 	if (!srpt_service_guid)
3231 		srpt_service_guid = be64_to_cpu(device->node_guid);
3232 
3233 	sdev->cm_id = ib_create_cm_id(device, srpt_cm_handler, sdev);
3234 	if (IS_ERR(sdev->cm_id))
3235 		goto err_srq;
3236 
3237 	/* print out target login information */
3238 	pr_debug("Target login info: id_ext=%016llx,ioc_guid=%016llx,"
3239 		 "pkey=ffff,service_id=%016llx\n", srpt_service_guid,
3240 		 srpt_service_guid, srpt_service_guid);
3241 
3242 	/*
3243 	 * We do not have a consistent service_id (ie. also id_ext of target_id)
3244 	 * to identify this target. We currently use the guid of the first HCA
3245 	 * in the system as service_id; therefore, the target_id will change
3246 	 * if this HCA is gone bad and replaced by different HCA
3247 	 */
3248 	if (ib_cm_listen(sdev->cm_id, cpu_to_be64(srpt_service_guid), 0))
3249 		goto err_cm;
3250 
3251 	INIT_IB_EVENT_HANDLER(&sdev->event_handler, sdev->device,
3252 			      srpt_event_handler);
3253 	if (ib_register_event_handler(&sdev->event_handler))
3254 		goto err_cm;
3255 
3256 	sdev->ioctx_ring = (struct srpt_recv_ioctx **)
3257 		srpt_alloc_ioctx_ring(sdev, sdev->srq_size,
3258 				      sizeof(*sdev->ioctx_ring[0]),
3259 				      srp_max_req_size, DMA_FROM_DEVICE);
3260 	if (!sdev->ioctx_ring)
3261 		goto err_event;
3262 
3263 	for (i = 0; i < sdev->srq_size; ++i)
3264 		srpt_post_recv(sdev, sdev->ioctx_ring[i]);
3265 
3266 	WARN_ON(sdev->device->phys_port_cnt > ARRAY_SIZE(sdev->port));
3267 
3268 	for (i = 1; i <= sdev->device->phys_port_cnt; i++) {
3269 		sport = &sdev->port[i - 1];
3270 		sport->sdev = sdev;
3271 		sport->port = i;
3272 		sport->port_attrib.srp_max_rdma_size = DEFAULT_MAX_RDMA_SIZE;
3273 		sport->port_attrib.srp_max_rsp_size = DEFAULT_MAX_RSP_SIZE;
3274 		sport->port_attrib.srp_sq_size = DEF_SRPT_SQ_SIZE;
3275 		INIT_WORK(&sport->work, srpt_refresh_port_work);
3276 		INIT_LIST_HEAD(&sport->port_acl_list);
3277 		spin_lock_init(&sport->port_acl_lock);
3278 
3279 		if (srpt_refresh_port(sport)) {
3280 			pr_err("MAD registration failed for %s-%d.\n",
3281 			       srpt_sdev_name(sdev), i);
3282 			goto err_ring;
3283 		}
3284 		snprintf(sport->port_guid, sizeof(sport->port_guid),
3285 			"0x%016llx%016llx",
3286 			be64_to_cpu(sport->gid.global.subnet_prefix),
3287 			be64_to_cpu(sport->gid.global.interface_id));
3288 	}
3289 
3290 	spin_lock(&srpt_dev_lock);
3291 	list_add_tail(&sdev->list, &srpt_dev_list);
3292 	spin_unlock(&srpt_dev_lock);
3293 
3294 out:
3295 	ib_set_client_data(device, &srpt_client, sdev);
3296 	pr_debug("added %s.\n", device->name);
3297 	return;
3298 
3299 err_ring:
3300 	srpt_free_ioctx_ring((struct srpt_ioctx **)sdev->ioctx_ring, sdev,
3301 			     sdev->srq_size, srp_max_req_size,
3302 			     DMA_FROM_DEVICE);
3303 err_event:
3304 	ib_unregister_event_handler(&sdev->event_handler);
3305 err_cm:
3306 	ib_destroy_cm_id(sdev->cm_id);
3307 err_srq:
3308 	ib_destroy_srq(sdev->srq);
3309 err_pd:
3310 	ib_dealloc_pd(sdev->pd);
3311 free_dev:
3312 	kfree(sdev);
3313 err:
3314 	sdev = NULL;
3315 	pr_info("%s(%s) failed.\n", __func__, device->name);
3316 	goto out;
3317 }
3318 
3319 /**
3320  * srpt_remove_one() - InfiniBand device removal callback function.
3321  */
3322 static void srpt_remove_one(struct ib_device *device, void *client_data)
3323 {
3324 	struct srpt_device *sdev = client_data;
3325 	int i;
3326 
3327 	if (!sdev) {
3328 		pr_info("%s(%s): nothing to do.\n", __func__, device->name);
3329 		return;
3330 	}
3331 
3332 	srpt_unregister_mad_agent(sdev);
3333 
3334 	ib_unregister_event_handler(&sdev->event_handler);
3335 
3336 	/* Cancel any work queued by the just unregistered IB event handler. */
3337 	for (i = 0; i < sdev->device->phys_port_cnt; i++)
3338 		cancel_work_sync(&sdev->port[i].work);
3339 
3340 	ib_destroy_cm_id(sdev->cm_id);
3341 
3342 	/*
3343 	 * Unregistering a target must happen after destroying sdev->cm_id
3344 	 * such that no new SRP_LOGIN_REQ information units can arrive while
3345 	 * destroying the target.
3346 	 */
3347 	spin_lock(&srpt_dev_lock);
3348 	list_del(&sdev->list);
3349 	spin_unlock(&srpt_dev_lock);
3350 	srpt_release_sdev(sdev);
3351 
3352 	ib_destroy_srq(sdev->srq);
3353 	ib_dealloc_pd(sdev->pd);
3354 
3355 	srpt_free_ioctx_ring((struct srpt_ioctx **)sdev->ioctx_ring, sdev,
3356 			     sdev->srq_size, srp_max_req_size, DMA_FROM_DEVICE);
3357 	sdev->ioctx_ring = NULL;
3358 	kfree(sdev);
3359 }
3360 
3361 static struct ib_client srpt_client = {
3362 	.name = DRV_NAME,
3363 	.add = srpt_add_one,
3364 	.remove = srpt_remove_one
3365 };
3366 
3367 static int srpt_check_true(struct se_portal_group *se_tpg)
3368 {
3369 	return 1;
3370 }
3371 
3372 static int srpt_check_false(struct se_portal_group *se_tpg)
3373 {
3374 	return 0;
3375 }
3376 
3377 static char *srpt_get_fabric_name(void)
3378 {
3379 	return "srpt";
3380 }
3381 
3382 static char *srpt_get_fabric_wwn(struct se_portal_group *tpg)
3383 {
3384 	struct srpt_port *sport = container_of(tpg, struct srpt_port, port_tpg_1);
3385 
3386 	return sport->port_guid;
3387 }
3388 
3389 static u16 srpt_get_tag(struct se_portal_group *tpg)
3390 {
3391 	return 1;
3392 }
3393 
3394 static u32 srpt_tpg_get_inst_index(struct se_portal_group *se_tpg)
3395 {
3396 	return 1;
3397 }
3398 
3399 static void srpt_release_cmd(struct se_cmd *se_cmd)
3400 {
3401 	struct srpt_send_ioctx *ioctx = container_of(se_cmd,
3402 				struct srpt_send_ioctx, cmd);
3403 	struct srpt_rdma_ch *ch = ioctx->ch;
3404 	unsigned long flags;
3405 
3406 	WARN_ON(ioctx->state != SRPT_STATE_DONE);
3407 	WARN_ON(ioctx->mapped_sg_count != 0);
3408 
3409 	if (ioctx->n_rbuf > 1) {
3410 		kfree(ioctx->rbufs);
3411 		ioctx->rbufs = NULL;
3412 		ioctx->n_rbuf = 0;
3413 	}
3414 
3415 	spin_lock_irqsave(&ch->spinlock, flags);
3416 	list_add(&ioctx->free_list, &ch->free_list);
3417 	spin_unlock_irqrestore(&ch->spinlock, flags);
3418 }
3419 
3420 /**
3421  * srpt_close_session() - Forcibly close a session.
3422  *
3423  * Callback function invoked by the TCM core to clean up sessions associated
3424  * with a node ACL when the user invokes
3425  * rmdir /sys/kernel/config/target/$driver/$port/$tpg/acls/$i_port_id
3426  */
3427 static void srpt_close_session(struct se_session *se_sess)
3428 {
3429 	DECLARE_COMPLETION_ONSTACK(release_done);
3430 	struct srpt_rdma_ch *ch;
3431 	struct srpt_device *sdev;
3432 	unsigned long res;
3433 
3434 	ch = se_sess->fabric_sess_ptr;
3435 	WARN_ON(ch->sess != se_sess);
3436 
3437 	pr_debug("ch %p state %d\n", ch, srpt_get_ch_state(ch));
3438 
3439 	sdev = ch->sport->sdev;
3440 	spin_lock_irq(&sdev->spinlock);
3441 	BUG_ON(ch->release_done);
3442 	ch->release_done = &release_done;
3443 	__srpt_close_ch(ch);
3444 	spin_unlock_irq(&sdev->spinlock);
3445 
3446 	res = wait_for_completion_timeout(&release_done, 60 * HZ);
3447 	WARN_ON(res == 0);
3448 }
3449 
3450 /**
3451  * srpt_sess_get_index() - Return the value of scsiAttIntrPortIndex (SCSI-MIB).
3452  *
3453  * A quote from RFC 4455 (SCSI-MIB) about this MIB object:
3454  * This object represents an arbitrary integer used to uniquely identify a
3455  * particular attached remote initiator port to a particular SCSI target port
3456  * within a particular SCSI target device within a particular SCSI instance.
3457  */
3458 static u32 srpt_sess_get_index(struct se_session *se_sess)
3459 {
3460 	return 0;
3461 }
3462 
3463 static void srpt_set_default_node_attrs(struct se_node_acl *nacl)
3464 {
3465 }
3466 
3467 /* Note: only used from inside debug printk's by the TCM core. */
3468 static int srpt_get_tcm_cmd_state(struct se_cmd *se_cmd)
3469 {
3470 	struct srpt_send_ioctx *ioctx;
3471 
3472 	ioctx = container_of(se_cmd, struct srpt_send_ioctx, cmd);
3473 	return srpt_get_cmd_state(ioctx);
3474 }
3475 
3476 /**
3477  * srpt_parse_i_port_id() - Parse an initiator port ID.
3478  * @name: ASCII representation of a 128-bit initiator port ID.
3479  * @i_port_id: Binary 128-bit port ID.
3480  */
3481 static int srpt_parse_i_port_id(u8 i_port_id[16], const char *name)
3482 {
3483 	const char *p;
3484 	unsigned len, count, leading_zero_bytes;
3485 	int ret, rc;
3486 
3487 	p = name;
3488 	if (strncasecmp(p, "0x", 2) == 0)
3489 		p += 2;
3490 	ret = -EINVAL;
3491 	len = strlen(p);
3492 	if (len % 2)
3493 		goto out;
3494 	count = min(len / 2, 16U);
3495 	leading_zero_bytes = 16 - count;
3496 	memset(i_port_id, 0, leading_zero_bytes);
3497 	rc = hex2bin(i_port_id + leading_zero_bytes, p, count);
3498 	if (rc < 0)
3499 		pr_debug("hex2bin failed for srpt_parse_i_port_id: %d\n", rc);
3500 	ret = 0;
3501 out:
3502 	return ret;
3503 }
3504 
3505 /*
3506  * configfs callback function invoked for
3507  * mkdir /sys/kernel/config/target/$driver/$port/$tpg/acls/$i_port_id
3508  */
3509 static int srpt_init_nodeacl(struct se_node_acl *se_nacl, const char *name)
3510 {
3511 	struct srpt_port *sport =
3512 		container_of(se_nacl->se_tpg, struct srpt_port, port_tpg_1);
3513 	struct srpt_node_acl *nacl =
3514 		container_of(se_nacl, struct srpt_node_acl, nacl);
3515 	u8 i_port_id[16];
3516 
3517 	if (srpt_parse_i_port_id(i_port_id, name) < 0) {
3518 		pr_err("invalid initiator port ID %s\n", name);
3519 		return -EINVAL;
3520 	}
3521 
3522 	memcpy(&nacl->i_port_id[0], &i_port_id[0], 16);
3523 	nacl->sport = sport;
3524 
3525 	spin_lock_irq(&sport->port_acl_lock);
3526 	list_add_tail(&nacl->list, &sport->port_acl_list);
3527 	spin_unlock_irq(&sport->port_acl_lock);
3528 
3529 	return 0;
3530 }
3531 
3532 /*
3533  * configfs callback function invoked for
3534  * rmdir /sys/kernel/config/target/$driver/$port/$tpg/acls/$i_port_id
3535  */
3536 static void srpt_cleanup_nodeacl(struct se_node_acl *se_nacl)
3537 {
3538 	struct srpt_node_acl *nacl =
3539 		container_of(se_nacl, struct srpt_node_acl, nacl);
3540 	struct srpt_port *sport = nacl->sport;
3541 
3542 	spin_lock_irq(&sport->port_acl_lock);
3543 	list_del(&nacl->list);
3544 	spin_unlock_irq(&sport->port_acl_lock);
3545 }
3546 
3547 static ssize_t srpt_tpg_attrib_srp_max_rdma_size_show(struct config_item *item,
3548 		char *page)
3549 {
3550 	struct se_portal_group *se_tpg = attrib_to_tpg(item);
3551 	struct srpt_port *sport = container_of(se_tpg, struct srpt_port, port_tpg_1);
3552 
3553 	return sprintf(page, "%u\n", sport->port_attrib.srp_max_rdma_size);
3554 }
3555 
3556 static ssize_t srpt_tpg_attrib_srp_max_rdma_size_store(struct config_item *item,
3557 		const char *page, size_t count)
3558 {
3559 	struct se_portal_group *se_tpg = attrib_to_tpg(item);
3560 	struct srpt_port *sport = container_of(se_tpg, struct srpt_port, port_tpg_1);
3561 	unsigned long val;
3562 	int ret;
3563 
3564 	ret = kstrtoul(page, 0, &val);
3565 	if (ret < 0) {
3566 		pr_err("kstrtoul() failed with ret: %d\n", ret);
3567 		return -EINVAL;
3568 	}
3569 	if (val > MAX_SRPT_RDMA_SIZE) {
3570 		pr_err("val: %lu exceeds MAX_SRPT_RDMA_SIZE: %d\n", val,
3571 			MAX_SRPT_RDMA_SIZE);
3572 		return -EINVAL;
3573 	}
3574 	if (val < DEFAULT_MAX_RDMA_SIZE) {
3575 		pr_err("val: %lu smaller than DEFAULT_MAX_RDMA_SIZE: %d\n",
3576 			val, DEFAULT_MAX_RDMA_SIZE);
3577 		return -EINVAL;
3578 	}
3579 	sport->port_attrib.srp_max_rdma_size = val;
3580 
3581 	return count;
3582 }
3583 
3584 static ssize_t srpt_tpg_attrib_srp_max_rsp_size_show(struct config_item *item,
3585 		char *page)
3586 {
3587 	struct se_portal_group *se_tpg = attrib_to_tpg(item);
3588 	struct srpt_port *sport = container_of(se_tpg, struct srpt_port, port_tpg_1);
3589 
3590 	return sprintf(page, "%u\n", sport->port_attrib.srp_max_rsp_size);
3591 }
3592 
3593 static ssize_t srpt_tpg_attrib_srp_max_rsp_size_store(struct config_item *item,
3594 		const char *page, size_t count)
3595 {
3596 	struct se_portal_group *se_tpg = attrib_to_tpg(item);
3597 	struct srpt_port *sport = container_of(se_tpg, struct srpt_port, port_tpg_1);
3598 	unsigned long val;
3599 	int ret;
3600 
3601 	ret = kstrtoul(page, 0, &val);
3602 	if (ret < 0) {
3603 		pr_err("kstrtoul() failed with ret: %d\n", ret);
3604 		return -EINVAL;
3605 	}
3606 	if (val > MAX_SRPT_RSP_SIZE) {
3607 		pr_err("val: %lu exceeds MAX_SRPT_RSP_SIZE: %d\n", val,
3608 			MAX_SRPT_RSP_SIZE);
3609 		return -EINVAL;
3610 	}
3611 	if (val < MIN_MAX_RSP_SIZE) {
3612 		pr_err("val: %lu smaller than MIN_MAX_RSP_SIZE: %d\n", val,
3613 			MIN_MAX_RSP_SIZE);
3614 		return -EINVAL;
3615 	}
3616 	sport->port_attrib.srp_max_rsp_size = val;
3617 
3618 	return count;
3619 }
3620 
3621 static ssize_t srpt_tpg_attrib_srp_sq_size_show(struct config_item *item,
3622 		char *page)
3623 {
3624 	struct se_portal_group *se_tpg = attrib_to_tpg(item);
3625 	struct srpt_port *sport = container_of(se_tpg, struct srpt_port, port_tpg_1);
3626 
3627 	return sprintf(page, "%u\n", sport->port_attrib.srp_sq_size);
3628 }
3629 
3630 static ssize_t srpt_tpg_attrib_srp_sq_size_store(struct config_item *item,
3631 		const char *page, size_t count)
3632 {
3633 	struct se_portal_group *se_tpg = attrib_to_tpg(item);
3634 	struct srpt_port *sport = container_of(se_tpg, struct srpt_port, port_tpg_1);
3635 	unsigned long val;
3636 	int ret;
3637 
3638 	ret = kstrtoul(page, 0, &val);
3639 	if (ret < 0) {
3640 		pr_err("kstrtoul() failed with ret: %d\n", ret);
3641 		return -EINVAL;
3642 	}
3643 	if (val > MAX_SRPT_SRQ_SIZE) {
3644 		pr_err("val: %lu exceeds MAX_SRPT_SRQ_SIZE: %d\n", val,
3645 			MAX_SRPT_SRQ_SIZE);
3646 		return -EINVAL;
3647 	}
3648 	if (val < MIN_SRPT_SRQ_SIZE) {
3649 		pr_err("val: %lu smaller than MIN_SRPT_SRQ_SIZE: %d\n", val,
3650 			MIN_SRPT_SRQ_SIZE);
3651 		return -EINVAL;
3652 	}
3653 	sport->port_attrib.srp_sq_size = val;
3654 
3655 	return count;
3656 }
3657 
3658 CONFIGFS_ATTR(srpt_tpg_attrib_,  srp_max_rdma_size);
3659 CONFIGFS_ATTR(srpt_tpg_attrib_,  srp_max_rsp_size);
3660 CONFIGFS_ATTR(srpt_tpg_attrib_,  srp_sq_size);
3661 
3662 static struct configfs_attribute *srpt_tpg_attrib_attrs[] = {
3663 	&srpt_tpg_attrib_attr_srp_max_rdma_size,
3664 	&srpt_tpg_attrib_attr_srp_max_rsp_size,
3665 	&srpt_tpg_attrib_attr_srp_sq_size,
3666 	NULL,
3667 };
3668 
3669 static ssize_t srpt_tpg_enable_show(struct config_item *item, char *page)
3670 {
3671 	struct se_portal_group *se_tpg = to_tpg(item);
3672 	struct srpt_port *sport = container_of(se_tpg, struct srpt_port, port_tpg_1);
3673 
3674 	return snprintf(page, PAGE_SIZE, "%d\n", (sport->enabled) ? 1: 0);
3675 }
3676 
3677 static ssize_t srpt_tpg_enable_store(struct config_item *item,
3678 		const char *page, size_t count)
3679 {
3680 	struct se_portal_group *se_tpg = to_tpg(item);
3681 	struct srpt_port *sport = container_of(se_tpg, struct srpt_port, port_tpg_1);
3682 	unsigned long tmp;
3683         int ret;
3684 
3685 	ret = kstrtoul(page, 0, &tmp);
3686 	if (ret < 0) {
3687 		pr_err("Unable to extract srpt_tpg_store_enable\n");
3688 		return -EINVAL;
3689 	}
3690 
3691 	if ((tmp != 0) && (tmp != 1)) {
3692 		pr_err("Illegal value for srpt_tpg_store_enable: %lu\n", tmp);
3693 		return -EINVAL;
3694 	}
3695 	if (tmp == 1)
3696 		sport->enabled = true;
3697 	else
3698 		sport->enabled = false;
3699 
3700 	return count;
3701 }
3702 
3703 CONFIGFS_ATTR(srpt_tpg_, enable);
3704 
3705 static struct configfs_attribute *srpt_tpg_attrs[] = {
3706 	&srpt_tpg_attr_enable,
3707 	NULL,
3708 };
3709 
3710 /**
3711  * configfs callback invoked for
3712  * mkdir /sys/kernel/config/target/$driver/$port/$tpg
3713  */
3714 static struct se_portal_group *srpt_make_tpg(struct se_wwn *wwn,
3715 					     struct config_group *group,
3716 					     const char *name)
3717 {
3718 	struct srpt_port *sport = container_of(wwn, struct srpt_port, port_wwn);
3719 	int res;
3720 
3721 	/* Initialize sport->port_wwn and sport->port_tpg_1 */
3722 	res = core_tpg_register(&sport->port_wwn, &sport->port_tpg_1, SCSI_PROTOCOL_SRP);
3723 	if (res)
3724 		return ERR_PTR(res);
3725 
3726 	return &sport->port_tpg_1;
3727 }
3728 
3729 /**
3730  * configfs callback invoked for
3731  * rmdir /sys/kernel/config/target/$driver/$port/$tpg
3732  */
3733 static void srpt_drop_tpg(struct se_portal_group *tpg)
3734 {
3735 	struct srpt_port *sport = container_of(tpg,
3736 				struct srpt_port, port_tpg_1);
3737 
3738 	sport->enabled = false;
3739 	core_tpg_deregister(&sport->port_tpg_1);
3740 }
3741 
3742 /**
3743  * configfs callback invoked for
3744  * mkdir /sys/kernel/config/target/$driver/$port
3745  */
3746 static struct se_wwn *srpt_make_tport(struct target_fabric_configfs *tf,
3747 				      struct config_group *group,
3748 				      const char *name)
3749 {
3750 	struct srpt_port *sport;
3751 	int ret;
3752 
3753 	sport = srpt_lookup_port(name);
3754 	pr_debug("make_tport(%s)\n", name);
3755 	ret = -EINVAL;
3756 	if (!sport)
3757 		goto err;
3758 
3759 	return &sport->port_wwn;
3760 
3761 err:
3762 	return ERR_PTR(ret);
3763 }
3764 
3765 /**
3766  * configfs callback invoked for
3767  * rmdir /sys/kernel/config/target/$driver/$port
3768  */
3769 static void srpt_drop_tport(struct se_wwn *wwn)
3770 {
3771 	struct srpt_port *sport = container_of(wwn, struct srpt_port, port_wwn);
3772 
3773 	pr_debug("drop_tport(%s\n", config_item_name(&sport->port_wwn.wwn_group.cg_item));
3774 }
3775 
3776 static ssize_t srpt_wwn_version_show(struct config_item *item, char *buf)
3777 {
3778 	return scnprintf(buf, PAGE_SIZE, "%s\n", DRV_VERSION);
3779 }
3780 
3781 CONFIGFS_ATTR_RO(srpt_wwn_, version);
3782 
3783 static struct configfs_attribute *srpt_wwn_attrs[] = {
3784 	&srpt_wwn_attr_version,
3785 	NULL,
3786 };
3787 
3788 static const struct target_core_fabric_ops srpt_template = {
3789 	.module				= THIS_MODULE,
3790 	.name				= "srpt",
3791 	.node_acl_size			= sizeof(struct srpt_node_acl),
3792 	.get_fabric_name		= srpt_get_fabric_name,
3793 	.tpg_get_wwn			= srpt_get_fabric_wwn,
3794 	.tpg_get_tag			= srpt_get_tag,
3795 	.tpg_check_demo_mode		= srpt_check_false,
3796 	.tpg_check_demo_mode_cache	= srpt_check_true,
3797 	.tpg_check_demo_mode_write_protect = srpt_check_true,
3798 	.tpg_check_prod_mode_write_protect = srpt_check_false,
3799 	.tpg_get_inst_index		= srpt_tpg_get_inst_index,
3800 	.release_cmd			= srpt_release_cmd,
3801 	.check_stop_free		= srpt_check_stop_free,
3802 	.shutdown_session		= srpt_shutdown_session,
3803 	.close_session			= srpt_close_session,
3804 	.sess_get_index			= srpt_sess_get_index,
3805 	.sess_get_initiator_sid		= NULL,
3806 	.write_pending			= srpt_write_pending,
3807 	.write_pending_status		= srpt_write_pending_status,
3808 	.set_default_node_attributes	= srpt_set_default_node_attrs,
3809 	.get_cmd_state			= srpt_get_tcm_cmd_state,
3810 	.queue_data_in			= srpt_queue_data_in,
3811 	.queue_status			= srpt_queue_status,
3812 	.queue_tm_rsp			= srpt_queue_tm_rsp,
3813 	.aborted_task			= srpt_aborted_task,
3814 	/*
3815 	 * Setup function pointers for generic logic in
3816 	 * target_core_fabric_configfs.c
3817 	 */
3818 	.fabric_make_wwn		= srpt_make_tport,
3819 	.fabric_drop_wwn		= srpt_drop_tport,
3820 	.fabric_make_tpg		= srpt_make_tpg,
3821 	.fabric_drop_tpg		= srpt_drop_tpg,
3822 	.fabric_init_nodeacl		= srpt_init_nodeacl,
3823 	.fabric_cleanup_nodeacl		= srpt_cleanup_nodeacl,
3824 
3825 	.tfc_wwn_attrs			= srpt_wwn_attrs,
3826 	.tfc_tpg_base_attrs		= srpt_tpg_attrs,
3827 	.tfc_tpg_attrib_attrs		= srpt_tpg_attrib_attrs,
3828 };
3829 
3830 /**
3831  * srpt_init_module() - Kernel module initialization.
3832  *
3833  * Note: Since ib_register_client() registers callback functions, and since at
3834  * least one of these callback functions (srpt_add_one()) calls target core
3835  * functions, this driver must be registered with the target core before
3836  * ib_register_client() is called.
3837  */
3838 static int __init srpt_init_module(void)
3839 {
3840 	int ret;
3841 
3842 	ret = -EINVAL;
3843 	if (srp_max_req_size < MIN_MAX_REQ_SIZE) {
3844 		pr_err("invalid value %d for kernel module parameter"
3845 		       " srp_max_req_size -- must be at least %d.\n",
3846 		       srp_max_req_size, MIN_MAX_REQ_SIZE);
3847 		goto out;
3848 	}
3849 
3850 	if (srpt_srq_size < MIN_SRPT_SRQ_SIZE
3851 	    || srpt_srq_size > MAX_SRPT_SRQ_SIZE) {
3852 		pr_err("invalid value %d for kernel module parameter"
3853 		       " srpt_srq_size -- must be in the range [%d..%d].\n",
3854 		       srpt_srq_size, MIN_SRPT_SRQ_SIZE, MAX_SRPT_SRQ_SIZE);
3855 		goto out;
3856 	}
3857 
3858 	ret = target_register_template(&srpt_template);
3859 	if (ret)
3860 		goto out;
3861 
3862 	ret = ib_register_client(&srpt_client);
3863 	if (ret) {
3864 		pr_err("couldn't register IB client\n");
3865 		goto out_unregister_target;
3866 	}
3867 
3868 	return 0;
3869 
3870 out_unregister_target:
3871 	target_unregister_template(&srpt_template);
3872 out:
3873 	return ret;
3874 }
3875 
3876 static void __exit srpt_cleanup_module(void)
3877 {
3878 	ib_unregister_client(&srpt_client);
3879 	target_unregister_template(&srpt_template);
3880 }
3881 
3882 module_init(srpt_init_module);
3883 module_exit(srpt_cleanup_module);
3884