1 /*
2  * iSCSI Initiator over iSER Data-Path
3  *
4  * Copyright (C) 2004 Dmitry Yusupov
5  * Copyright (C) 2004 Alex Aizman
6  * Copyright (C) 2005 Mike Christie
7  * Copyright (c) 2005, 2006 Voltaire, Inc. All rights reserved.
8  * Copyright (c) 2013-2014 Mellanox Technologies. All rights reserved.
9  * maintained by openib-general@openib.org
10  *
11  * This software is available to you under a choice of one of two
12  * licenses.  You may choose to be licensed under the terms of the GNU
13  * General Public License (GPL) Version 2, available from the file
14  * COPYING in the main directory of this source tree, or the
15  * OpenIB.org BSD license below:
16  *
17  *     Redistribution and use in source and binary forms, with or
18  *     without modification, are permitted provided that the following
19  *     conditions are met:
20  *
21  *	- Redistributions of source code must retain the above
22  *	  copyright notice, this list of conditions and the following
23  *	  disclaimer.
24  *
25  *	- Redistributions in binary form must reproduce the above
26  *	  copyright notice, this list of conditions and the following
27  *	  disclaimer in the documentation and/or other materials
28  *	  provided with the distribution.
29  *
30  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
31  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
32  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
33  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
34  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
35  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
36  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
37  * SOFTWARE.
38  *
39  * Credits:
40  *	Christoph Hellwig
41  *	FUJITA Tomonori
42  *	Arne Redlich
43  *	Zhenyu Wang
44  * Modified by:
45  *      Erez Zilber
46  */
47 
48 #include <linux/types.h>
49 #include <linux/list.h>
50 #include <linux/hardirq.h>
51 #include <linux/kfifo.h>
52 #include <linux/blkdev.h>
53 #include <linux/init.h>
54 #include <linux/ioctl.h>
55 #include <linux/cdev.h>
56 #include <linux/in.h>
57 #include <linux/net.h>
58 #include <linux/scatterlist.h>
59 #include <linux/delay.h>
60 #include <linux/slab.h>
61 #include <linux/module.h>
62 
63 #include <net/sock.h>
64 
65 #include <linux/uaccess.h>
66 
67 #include <scsi/scsi_cmnd.h>
68 #include <scsi/scsi_device.h>
69 #include <scsi/scsi_eh.h>
70 #include <scsi/scsi_tcq.h>
71 #include <scsi/scsi_host.h>
72 #include <scsi/scsi.h>
73 #include <scsi/scsi_transport_iscsi.h>
74 
75 #include "iscsi_iser.h"
76 
77 MODULE_DESCRIPTION("iSER (iSCSI Extensions for RDMA) Datamover");
78 MODULE_LICENSE("Dual BSD/GPL");
79 MODULE_AUTHOR("Alex Nezhinsky, Dan Bar Dov, Or Gerlitz");
80 
81 static const struct scsi_host_template iscsi_iser_sht;
82 static struct iscsi_transport iscsi_iser_transport;
83 static struct scsi_transport_template *iscsi_iser_scsi_transport;
84 static struct workqueue_struct *release_wq;
85 static DEFINE_MUTEX(unbind_iser_conn_mutex);
86 struct iser_global ig;
87 
88 int iser_debug_level = 0;
89 module_param_named(debug_level, iser_debug_level, int, S_IRUGO | S_IWUSR);
90 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0 (default:disabled)");
91 
92 static int iscsi_iser_set(const char *val, const struct kernel_param *kp);
93 static const struct kernel_param_ops iscsi_iser_size_ops = {
94 	.set = iscsi_iser_set,
95 	.get = param_get_uint,
96 };
97 
98 static unsigned int iscsi_max_lun = 512;
99 module_param_cb(max_lun, &iscsi_iser_size_ops, &iscsi_max_lun, S_IRUGO);
100 MODULE_PARM_DESC(max_lun, "Max LUNs to allow per session, should > 0 (default:512)");
101 
102 unsigned int iser_max_sectors = ISER_DEF_MAX_SECTORS;
103 module_param_cb(max_sectors, &iscsi_iser_size_ops, &iser_max_sectors,
104 		S_IRUGO | S_IWUSR);
105 MODULE_PARM_DESC(max_sectors, "Max number of sectors in a single scsi command, should > 0 (default:1024)");
106 
107 bool iser_always_reg = true;
108 module_param_named(always_register, iser_always_reg, bool, S_IRUGO);
109 MODULE_PARM_DESC(always_register,
110 		 "Always register memory, even for continuous memory regions (default:true)");
111 
112 bool iser_pi_enable = false;
113 module_param_named(pi_enable, iser_pi_enable, bool, S_IRUGO);
114 MODULE_PARM_DESC(pi_enable, "Enable T10-PI offload support (default:disabled)");
115 
116 static int iscsi_iser_set(const char *val, const struct kernel_param *kp)
117 {
118 	int ret;
119 	unsigned int n = 0;
120 
121 	ret = kstrtouint(val, 10, &n);
122 	if (ret != 0 || n == 0)
123 		return -EINVAL;
124 
125 	return param_set_uint(val, kp);
126 }
127 
128 /*
129  * iscsi_iser_recv() - Process a successful recv completion
130  * @conn:         iscsi connection
131  * @hdr:          iscsi header
132  * @rx_data:      buffer containing receive data payload
133  * @rx_data_len:  length of rx_data
134  *
135  * Notes: In case of data length errors or iscsi PDU completion failures
136  *        this routine will signal iscsi layer of connection failure.
137  */
138 void iscsi_iser_recv(struct iscsi_conn *conn, struct iscsi_hdr *hdr,
139 		     char *rx_data, int rx_data_len)
140 {
141 	int rc = 0;
142 	int datalen;
143 
144 	/* verify PDU length */
145 	datalen = ntoh24(hdr->dlength);
146 	if (datalen > rx_data_len || (datalen + 4) < rx_data_len) {
147 		iser_err("wrong datalen %d (hdr), %d (IB)\n",
148 			datalen, rx_data_len);
149 		rc = ISCSI_ERR_DATALEN;
150 		goto error;
151 	}
152 
153 	if (datalen != rx_data_len)
154 		iser_dbg("aligned datalen (%d) hdr, %d (IB)\n",
155 			datalen, rx_data_len);
156 
157 	rc = iscsi_complete_pdu(conn, hdr, rx_data, rx_data_len);
158 	if (rc && rc != ISCSI_ERR_NO_SCSI_CMD)
159 		goto error;
160 
161 	return;
162 error:
163 	iscsi_conn_failure(conn, rc);
164 }
165 
166 /**
167  * iscsi_iser_pdu_alloc() - allocate an iscsi-iser PDU
168  * @task:     iscsi task
169  * @opcode:   iscsi command opcode
170  *
171  * Netes: This routine can't fail, just assign iscsi task
172  *        hdr and max hdr size.
173  */
174 static int iscsi_iser_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
175 {
176 	struct iscsi_iser_task *iser_task = task->dd_data;
177 
178 	task->hdr = (struct iscsi_hdr *)&iser_task->desc.iscsi_header;
179 	task->hdr_max = sizeof(iser_task->desc.iscsi_header);
180 
181 	return 0;
182 }
183 
184 /**
185  * iser_initialize_task_headers() - Initialize task headers
186  * @task:       iscsi task
187  * @tx_desc:    iser tx descriptor
188  *
189  * Notes:
190  * This routine may race with iser teardown flow for scsi
191  * error handling TMFs. So for TMF we should acquire the
192  * state mutex to avoid dereferencing the IB device which
193  * may have already been terminated.
194  */
195 int iser_initialize_task_headers(struct iscsi_task *task,
196 				 struct iser_tx_desc *tx_desc)
197 {
198 	struct iser_conn *iser_conn = task->conn->dd_data;
199 	struct iser_device *device = iser_conn->ib_conn.device;
200 	struct iscsi_iser_task *iser_task = task->dd_data;
201 	u64 dma_addr;
202 
203 	if (unlikely(iser_conn->state != ISER_CONN_UP))
204 		return -ENODEV;
205 
206 	dma_addr = ib_dma_map_single(device->ib_device, (void *)tx_desc,
207 				ISER_HEADERS_LEN, DMA_TO_DEVICE);
208 	if (ib_dma_mapping_error(device->ib_device, dma_addr))
209 		return -ENOMEM;
210 
211 	tx_desc->inv_wr.next = NULL;
212 	tx_desc->reg_wr.wr.next = NULL;
213 	tx_desc->mapped = true;
214 	tx_desc->dma_addr = dma_addr;
215 	tx_desc->tx_sg[0].addr   = tx_desc->dma_addr;
216 	tx_desc->tx_sg[0].length = ISER_HEADERS_LEN;
217 	tx_desc->tx_sg[0].lkey   = device->pd->local_dma_lkey;
218 
219 	iser_task->iser_conn = iser_conn;
220 
221 	return 0;
222 }
223 
224 /**
225  * iscsi_iser_task_init() - Initialize iscsi-iser task
226  * @task: iscsi task
227  *
228  * Initialize the task for the scsi command or mgmt command.
229  *
230  * Return: Returns zero on success or -ENOMEM when failing
231  *         to init task headers (dma mapping error).
232  */
233 static int iscsi_iser_task_init(struct iscsi_task *task)
234 {
235 	struct iscsi_iser_task *iser_task = task->dd_data;
236 	int ret;
237 
238 	ret = iser_initialize_task_headers(task, &iser_task->desc);
239 	if (ret) {
240 		iser_err("Failed to init task %p, err = %d\n",
241 			 iser_task, ret);
242 		return ret;
243 	}
244 
245 	/* mgmt task */
246 	if (!task->sc)
247 		return 0;
248 
249 	iser_task->command_sent = 0;
250 	iser_task_rdma_init(iser_task);
251 	iser_task->sc = task->sc;
252 
253 	return 0;
254 }
255 
256 /**
257  * iscsi_iser_mtask_xmit() - xmit management (immediate) task
258  * @conn: iscsi connection
259  * @task: task management task
260  *
261  * Notes:
262  *	The function can return -EAGAIN in which case caller must
263  *	call it again later, or recover. '0' return code means successful
264  *	xmit.
265  *
266  **/
267 static int iscsi_iser_mtask_xmit(struct iscsi_conn *conn,
268 				 struct iscsi_task *task)
269 {
270 	int error = 0;
271 
272 	iser_dbg("mtask xmit [cid %d itt 0x%x]\n", conn->id, task->itt);
273 
274 	error = iser_send_control(conn, task);
275 
276 	/* since iser xmits control with zero copy, tasks can not be recycled
277 	 * right after sending them.
278 	 * The recycling scheme is based on whether a response is expected
279 	 * - if yes, the task is recycled at iscsi_complete_pdu
280 	 * - if no,  the task is recycled at iser_snd_completion
281 	 */
282 	return error;
283 }
284 
285 static int iscsi_iser_task_xmit_unsol_data(struct iscsi_conn *conn,
286 					   struct iscsi_task *task)
287 {
288 	struct iscsi_r2t_info *r2t = &task->unsol_r2t;
289 	struct iscsi_data hdr;
290 	int error = 0;
291 
292 	/* Send data-out PDUs while there's still unsolicited data to send */
293 	while (iscsi_task_has_unsol_data(task)) {
294 		iscsi_prep_data_out_pdu(task, r2t, &hdr);
295 		iser_dbg("Sending data-out: itt 0x%x, data count %d\n",
296 			   hdr.itt, r2t->data_count);
297 
298 		/* the buffer description has been passed with the command */
299 		/* Send the command */
300 		error = iser_send_data_out(conn, task, &hdr);
301 		if (error) {
302 			r2t->datasn--;
303 			goto iscsi_iser_task_xmit_unsol_data_exit;
304 		}
305 		r2t->sent += r2t->data_count;
306 		iser_dbg("Need to send %d more as data-out PDUs\n",
307 			   r2t->data_length - r2t->sent);
308 	}
309 
310 iscsi_iser_task_xmit_unsol_data_exit:
311 	return error;
312 }
313 
314 /**
315  * iscsi_iser_task_xmit() - xmit iscsi-iser task
316  * @task: iscsi task
317  *
318  * Return: zero on success or escalates $error on failure.
319  */
320 static int iscsi_iser_task_xmit(struct iscsi_task *task)
321 {
322 	struct iscsi_conn *conn = task->conn;
323 	struct iscsi_iser_task *iser_task = task->dd_data;
324 	int error = 0;
325 
326 	if (!task->sc)
327 		return iscsi_iser_mtask_xmit(conn, task);
328 
329 	if (task->sc->sc_data_direction == DMA_TO_DEVICE) {
330 		BUG_ON(scsi_bufflen(task->sc) == 0);
331 
332 		iser_dbg("cmd [itt %x total %d imm %d unsol_data %d\n",
333 			   task->itt, scsi_bufflen(task->sc),
334 			   task->imm_count, task->unsol_r2t.data_length);
335 	}
336 
337 	iser_dbg("ctask xmit [cid %d itt 0x%x]\n",
338 		   conn->id, task->itt);
339 
340 	/* Send the cmd PDU */
341 	if (!iser_task->command_sent) {
342 		error = iser_send_command(conn, task);
343 		if (error)
344 			goto iscsi_iser_task_xmit_exit;
345 		iser_task->command_sent = 1;
346 	}
347 
348 	/* Send unsolicited data-out PDU(s) if necessary */
349 	if (iscsi_task_has_unsol_data(task))
350 		error = iscsi_iser_task_xmit_unsol_data(conn, task);
351 
352  iscsi_iser_task_xmit_exit:
353 	return error;
354 }
355 
356 /**
357  * iscsi_iser_cleanup_task() - cleanup an iscsi-iser task
358  * @task: iscsi task
359  *
360  * Notes: In case the RDMA device is already NULL (might have
361  *        been removed in DEVICE_REMOVAL CM event it will bail-out
362  *        without doing dma unmapping.
363  */
364 static void iscsi_iser_cleanup_task(struct iscsi_task *task)
365 {
366 	struct iscsi_iser_task *iser_task = task->dd_data;
367 	struct iser_tx_desc *tx_desc = &iser_task->desc;
368 	struct iser_conn *iser_conn = task->conn->dd_data;
369 	struct iser_device *device = iser_conn->ib_conn.device;
370 
371 	/* DEVICE_REMOVAL event might have already released the device */
372 	if (!device)
373 		return;
374 
375 	if (likely(tx_desc->mapped)) {
376 		ib_dma_unmap_single(device->ib_device, tx_desc->dma_addr,
377 				    ISER_HEADERS_LEN, DMA_TO_DEVICE);
378 		tx_desc->mapped = false;
379 	}
380 
381 	/* mgmt tasks do not need special cleanup */
382 	if (!task->sc)
383 		return;
384 
385 	if (iser_task->status == ISER_TASK_STATUS_STARTED) {
386 		iser_task->status = ISER_TASK_STATUS_COMPLETED;
387 		iser_task_rdma_finalize(iser_task);
388 	}
389 }
390 
391 /**
392  * iscsi_iser_check_protection() - check protection information status of task.
393  * @task:     iscsi task
394  * @sector:   error sector if exsists (output)
395  *
396  * Return: zero if no data-integrity errors have occured
397  *         0x1: data-integrity error occured in the guard-block
398  *         0x2: data-integrity error occured in the reference tag
399  *         0x3: data-integrity error occured in the application tag
400  *
401  *         In addition the error sector is marked.
402  */
403 static u8 iscsi_iser_check_protection(struct iscsi_task *task, sector_t *sector)
404 {
405 	struct iscsi_iser_task *iser_task = task->dd_data;
406 	enum iser_data_dir dir = iser_task->dir[ISER_DIR_IN] ?
407 					ISER_DIR_IN : ISER_DIR_OUT;
408 
409 	return iser_check_task_pi_status(iser_task, dir, sector);
410 }
411 
412 /**
413  * iscsi_iser_conn_create() - create a new iscsi-iser connection
414  * @cls_session: iscsi class connection
415  * @conn_idx:    connection index within the session (for MCS)
416  *
417  * Return: iscsi_cls_conn when iscsi_conn_setup succeeds or NULL
418  *         otherwise.
419  */
420 static struct iscsi_cls_conn *
421 iscsi_iser_conn_create(struct iscsi_cls_session *cls_session,
422 		       uint32_t conn_idx)
423 {
424 	struct iscsi_conn *conn;
425 	struct iscsi_cls_conn *cls_conn;
426 
427 	cls_conn = iscsi_conn_setup(cls_session, 0, conn_idx);
428 	if (!cls_conn)
429 		return NULL;
430 	conn = cls_conn->dd_data;
431 
432 	/*
433 	 * due to issues with the login code re iser sematics
434 	 * this not set in iscsi_conn_setup - FIXME
435 	 */
436 	conn->max_recv_dlength = ISER_RECV_DATA_SEG_LEN;
437 
438 	return cls_conn;
439 }
440 
441 /**
442  * iscsi_iser_conn_bind() - bind iscsi and iser connection structures
443  * @cls_session:     iscsi class session
444  * @cls_conn:        iscsi class connection
445  * @transport_eph:   transport end-point handle
446  * @is_leading:      indicate if this is the session leading connection (MCS)
447  *
448  * Return: zero on success, $error if iscsi_conn_bind fails and
449  *         -EINVAL in case end-point doesn't exists anymore or iser connection
450  *         state is not UP (teardown already started).
451  */
452 static int iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session,
453 				struct iscsi_cls_conn *cls_conn,
454 				uint64_t transport_eph, int is_leading)
455 {
456 	struct iscsi_conn *conn = cls_conn->dd_data;
457 	struct iser_conn *iser_conn;
458 	struct iscsi_endpoint *ep;
459 	int error;
460 
461 	error = iscsi_conn_bind(cls_session, cls_conn, is_leading);
462 	if (error)
463 		return error;
464 
465 	/* the transport ep handle comes from user space so it must be
466 	 * verified against the global ib connections list */
467 	ep = iscsi_lookup_endpoint(transport_eph);
468 	if (!ep) {
469 		iser_err("can't bind eph %llx\n",
470 			 (unsigned long long)transport_eph);
471 		return -EINVAL;
472 	}
473 	iser_conn = ep->dd_data;
474 
475 	mutex_lock(&iser_conn->state_mutex);
476 	if (iser_conn->state != ISER_CONN_UP) {
477 		error = -EINVAL;
478 		iser_err("iser_conn %p state is %d, teardown started\n",
479 			 iser_conn, iser_conn->state);
480 		goto out;
481 	}
482 
483 	error = iser_alloc_rx_descriptors(iser_conn, conn->session);
484 	if (error)
485 		goto out;
486 
487 	/* binds the iSER connection retrieved from the previously
488 	 * connected ep_handle to the iSCSI layer connection. exchanges
489 	 * connection pointers */
490 	iser_info("binding iscsi conn %p to iser_conn %p\n", conn, iser_conn);
491 
492 	conn->dd_data = iser_conn;
493 	iser_conn->iscsi_conn = conn;
494 
495 out:
496 	iscsi_put_endpoint(ep);
497 	mutex_unlock(&iser_conn->state_mutex);
498 	return error;
499 }
500 
501 /**
502  * iscsi_iser_conn_start() - start iscsi-iser connection
503  * @cls_conn: iscsi class connection
504  *
505  * Notes: Here iser intialize (or re-initialize) stop_completion as
506  *        from this point iscsi must call conn_stop in session/connection
507  *        teardown so iser transport must wait for it.
508  */
509 static int iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn)
510 {
511 	struct iscsi_conn *iscsi_conn;
512 	struct iser_conn *iser_conn;
513 
514 	iscsi_conn = cls_conn->dd_data;
515 	iser_conn = iscsi_conn->dd_data;
516 	reinit_completion(&iser_conn->stop_completion);
517 
518 	return iscsi_conn_start(cls_conn);
519 }
520 
521 /**
522  * iscsi_iser_conn_stop() - stop iscsi-iser connection
523  * @cls_conn:  iscsi class connection
524  * @flag:      indicate if recover or terminate (passed as is)
525  *
526  * Notes: Calling iscsi_conn_stop might theoretically race with
527  *        DEVICE_REMOVAL event and dereference a previously freed RDMA device
528  *        handle, so we call it under iser the state lock to protect against
529  *        this kind of race.
530  */
531 static void iscsi_iser_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
532 {
533 	struct iscsi_conn *conn = cls_conn->dd_data;
534 	struct iser_conn *iser_conn = conn->dd_data;
535 
536 	iser_info("stopping iscsi_conn: %p, iser_conn: %p\n", conn, iser_conn);
537 
538 	/*
539 	 * Userspace may have goofed up and not bound the connection or
540 	 * might have only partially setup the connection.
541 	 */
542 	if (iser_conn) {
543 		mutex_lock(&iser_conn->state_mutex);
544 		mutex_lock(&unbind_iser_conn_mutex);
545 		iser_conn_terminate(iser_conn);
546 		iscsi_conn_stop(cls_conn, flag);
547 
548 		/* unbind */
549 		iser_conn->iscsi_conn = NULL;
550 		conn->dd_data = NULL;
551 		mutex_unlock(&unbind_iser_conn_mutex);
552 
553 		complete(&iser_conn->stop_completion);
554 		mutex_unlock(&iser_conn->state_mutex);
555 	} else {
556 		iscsi_conn_stop(cls_conn, flag);
557 	}
558 }
559 
560 /**
561  * iscsi_iser_session_destroy() - destroy iscsi-iser session
562  * @cls_session: iscsi class session
563  *
564  * Removes and free iscsi host.
565  */
566 static void iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session)
567 {
568 	struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
569 
570 	iscsi_session_teardown(cls_session);
571 	iscsi_host_remove(shost, false);
572 	iscsi_host_free(shost);
573 }
574 
575 static inline unsigned int iser_dif_prot_caps(int prot_caps)
576 {
577 	int ret = 0;
578 
579 	if (prot_caps & IB_PROT_T10DIF_TYPE_1)
580 		ret |= SHOST_DIF_TYPE1_PROTECTION |
581 		       SHOST_DIX_TYPE0_PROTECTION |
582 		       SHOST_DIX_TYPE1_PROTECTION;
583 	if (prot_caps & IB_PROT_T10DIF_TYPE_2)
584 		ret |= SHOST_DIF_TYPE2_PROTECTION |
585 		       SHOST_DIX_TYPE2_PROTECTION;
586 	if (prot_caps & IB_PROT_T10DIF_TYPE_3)
587 		ret |= SHOST_DIF_TYPE3_PROTECTION |
588 		       SHOST_DIX_TYPE3_PROTECTION;
589 
590 	return ret;
591 }
592 
593 /**
594  * iscsi_iser_session_create() - create an iscsi-iser session
595  * @ep:             iscsi end-point handle
596  * @cmds_max:       maximum commands in this session
597  * @qdepth:         session command queue depth
598  * @initial_cmdsn:  initiator command sequnce number
599  *
600  * Allocates and adds a scsi host, expose DIF supprot if
601  * exists, and sets up an iscsi session.
602  */
603 static struct iscsi_cls_session *
604 iscsi_iser_session_create(struct iscsi_endpoint *ep,
605 			  uint16_t cmds_max, uint16_t qdepth,
606 			  uint32_t initial_cmdsn)
607 {
608 	struct iscsi_cls_session *cls_session;
609 	struct Scsi_Host *shost;
610 	struct iser_conn *iser_conn = NULL;
611 	struct ib_conn *ib_conn;
612 	struct ib_device *ib_dev;
613 	u32 max_fr_sectors;
614 
615 	shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 0);
616 	if (!shost)
617 		return NULL;
618 	shost->transportt = iscsi_iser_scsi_transport;
619 	shost->cmd_per_lun = qdepth;
620 	shost->max_lun = iscsi_max_lun;
621 	shost->max_id = 0;
622 	shost->max_channel = 0;
623 	shost->max_cmd_len = 16;
624 
625 	/*
626 	 * older userspace tools (before 2.0-870) did not pass us
627 	 * the leading conn's ep so this will be NULL;
628 	 */
629 	if (ep) {
630 		iser_conn = ep->dd_data;
631 		shost->sg_tablesize = iser_conn->scsi_sg_tablesize;
632 		shost->can_queue = min_t(u16, cmds_max, iser_conn->max_cmds);
633 
634 		mutex_lock(&iser_conn->state_mutex);
635 		if (iser_conn->state != ISER_CONN_UP) {
636 			iser_err("iser conn %p already started teardown\n",
637 				 iser_conn);
638 			mutex_unlock(&iser_conn->state_mutex);
639 			goto free_host;
640 		}
641 
642 		ib_conn = &iser_conn->ib_conn;
643 		ib_dev = ib_conn->device->ib_device;
644 		if (ib_conn->pi_support) {
645 			u32 sig_caps = ib_dev->attrs.sig_prot_cap;
646 
647 			shost->sg_prot_tablesize = shost->sg_tablesize;
648 			scsi_host_set_prot(shost, iser_dif_prot_caps(sig_caps));
649 			scsi_host_set_guard(shost, SHOST_DIX_GUARD_IP |
650 						   SHOST_DIX_GUARD_CRC);
651 		}
652 
653 		if (!(ib_dev->attrs.kernel_cap_flags & IBK_SG_GAPS_REG))
654 			shost->virt_boundary_mask = SZ_4K - 1;
655 
656 		if (iscsi_host_add(shost, ib_dev->dev.parent)) {
657 			mutex_unlock(&iser_conn->state_mutex);
658 			goto free_host;
659 		}
660 		mutex_unlock(&iser_conn->state_mutex);
661 	} else {
662 		shost->can_queue = min_t(u16, cmds_max, ISER_DEF_XMIT_CMDS_MAX);
663 		if (iscsi_host_add(shost, NULL))
664 			goto free_host;
665 	}
666 
667 	max_fr_sectors = (shost->sg_tablesize * PAGE_SIZE) >> 9;
668 	shost->max_sectors = min(iser_max_sectors, max_fr_sectors);
669 
670 	iser_dbg("iser_conn %p, sg_tablesize %u, max_sectors %u\n",
671 		 iser_conn, shost->sg_tablesize,
672 		 shost->max_sectors);
673 
674 	if (shost->max_sectors < iser_max_sectors)
675 		iser_warn("max_sectors was reduced from %u to %u\n",
676 			  iser_max_sectors, shost->max_sectors);
677 
678 	cls_session = iscsi_session_setup(&iscsi_iser_transport, shost,
679 					  shost->can_queue, 0,
680 					  sizeof(struct iscsi_iser_task),
681 					  initial_cmdsn, 0);
682 	if (!cls_session)
683 		goto remove_host;
684 
685 	return cls_session;
686 
687 remove_host:
688 	iscsi_host_remove(shost, false);
689 free_host:
690 	iscsi_host_free(shost);
691 	return NULL;
692 }
693 
694 static int iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn,
695 				enum iscsi_param param, char *buf, int buflen)
696 {
697 	int value;
698 
699 	switch (param) {
700 	case ISCSI_PARAM_MAX_RECV_DLENGTH:
701 		/* TBD */
702 		break;
703 	case ISCSI_PARAM_HDRDGST_EN:
704 		sscanf(buf, "%d", &value);
705 		if (value) {
706 			iser_err("DataDigest wasn't negotiated to None\n");
707 			return -EPROTO;
708 		}
709 		break;
710 	case ISCSI_PARAM_DATADGST_EN:
711 		sscanf(buf, "%d", &value);
712 		if (value) {
713 			iser_err("DataDigest wasn't negotiated to None\n");
714 			return -EPROTO;
715 		}
716 		break;
717 	case ISCSI_PARAM_IFMARKER_EN:
718 		sscanf(buf, "%d", &value);
719 		if (value) {
720 			iser_err("IFMarker wasn't negotiated to No\n");
721 			return -EPROTO;
722 		}
723 		break;
724 	case ISCSI_PARAM_OFMARKER_EN:
725 		sscanf(buf, "%d", &value);
726 		if (value) {
727 			iser_err("OFMarker wasn't negotiated to No\n");
728 			return -EPROTO;
729 		}
730 		break;
731 	default:
732 		return iscsi_set_param(cls_conn, param, buf, buflen);
733 	}
734 
735 	return 0;
736 }
737 
738 /**
739  * iscsi_iser_conn_get_stats() - get iscsi connection statistics
740  * @cls_conn:    iscsi class connection
741  * @stats:       iscsi stats to output
742  *
743  * Output connection statistics.
744  */
745 static void iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn,
746 				      struct iscsi_stats *stats)
747 {
748 	struct iscsi_conn *conn = cls_conn->dd_data;
749 
750 	stats->txdata_octets = conn->txdata_octets;
751 	stats->rxdata_octets = conn->rxdata_octets;
752 	stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
753 	stats->dataout_pdus = conn->dataout_pdus_cnt;
754 	stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
755 	stats->datain_pdus = conn->datain_pdus_cnt; /* always 0 */
756 	stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */
757 	stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
758 	stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
759 	stats->custom_length = 0;
760 }
761 
762 static int iscsi_iser_get_ep_param(struct iscsi_endpoint *ep,
763 				   enum iscsi_param param, char *buf)
764 {
765 	struct iser_conn *iser_conn = ep->dd_data;
766 
767 	switch (param) {
768 	case ISCSI_PARAM_CONN_PORT:
769 	case ISCSI_PARAM_CONN_ADDRESS:
770 		if (!iser_conn || !iser_conn->ib_conn.cma_id)
771 			return -ENOTCONN;
772 
773 		return iscsi_conn_get_addr_param((struct sockaddr_storage *)
774 				&iser_conn->ib_conn.cma_id->route.addr.dst_addr,
775 				param, buf);
776 	default:
777 		break;
778 	}
779 	return -ENOSYS;
780 }
781 
782 /**
783  * iscsi_iser_ep_connect() - Initiate iSER connection establishment
784  * @shost:          scsi_host
785  * @dst_addr:       destination address
786  * @non_blocking:   indicate if routine can block
787  *
788  * Allocate an iscsi endpoint, an iser_conn structure and bind them.
789  * After that start RDMA connection establishment via rdma_cm. We
790  * don't allocate iser_conn embedded in iscsi_endpoint since in teardown
791  * the endpoint will be destroyed at ep_disconnect while iser_conn will
792  * cleanup its resources asynchronuously.
793  *
794  * Return: iscsi_endpoint created by iscsi layer or ERR_PTR(error)
795  *         if fails.
796  */
797 static struct iscsi_endpoint *iscsi_iser_ep_connect(struct Scsi_Host *shost,
798 						    struct sockaddr *dst_addr,
799 						    int non_blocking)
800 {
801 	int err;
802 	struct iser_conn *iser_conn;
803 	struct iscsi_endpoint *ep;
804 
805 	ep = iscsi_create_endpoint(0);
806 	if (!ep)
807 		return ERR_PTR(-ENOMEM);
808 
809 	iser_conn = kzalloc(sizeof(*iser_conn), GFP_KERNEL);
810 	if (!iser_conn) {
811 		err = -ENOMEM;
812 		goto failure;
813 	}
814 
815 	ep->dd_data = iser_conn;
816 	iser_conn->ep = ep;
817 	iser_conn_init(iser_conn);
818 
819 	err = iser_connect(iser_conn, NULL, dst_addr, non_blocking);
820 	if (err)
821 		goto failure;
822 
823 	return ep;
824 failure:
825 	iscsi_destroy_endpoint(ep);
826 	return ERR_PTR(err);
827 }
828 
829 /**
830  * iscsi_iser_ep_poll() - poll for iser connection establishment to complete
831  * @ep:            iscsi endpoint (created at ep_connect)
832  * @timeout_ms:    polling timeout allowed in ms.
833  *
834  * This routine boils down to waiting for up_completion signaling
835  * that cma_id got CONNECTED event.
836  *
837  * Return: 1 if succeeded in connection establishment, 0 if timeout expired
838  *         (libiscsi will retry will kick in) or -1 if interrupted by signal
839  *         or more likely iser connection state transitioned to TEMINATING or
840  *         DOWN during the wait period.
841  */
842 static int iscsi_iser_ep_poll(struct iscsi_endpoint *ep, int timeout_ms)
843 {
844 	struct iser_conn *iser_conn = ep->dd_data;
845 	int rc;
846 
847 	rc = wait_for_completion_interruptible_timeout(&iser_conn->up_completion,
848 						       msecs_to_jiffies(timeout_ms));
849 	/* if conn establishment failed, return error code to iscsi */
850 	if (rc == 0) {
851 		mutex_lock(&iser_conn->state_mutex);
852 		if (iser_conn->state == ISER_CONN_TERMINATING ||
853 		    iser_conn->state == ISER_CONN_DOWN)
854 			rc = -1;
855 		mutex_unlock(&iser_conn->state_mutex);
856 	}
857 
858 	iser_info("iser conn %p rc = %d\n", iser_conn, rc);
859 
860 	if (rc > 0)
861 		return 1; /* success, this is the equivalent of EPOLLOUT */
862 	else if (!rc)
863 		return 0; /* timeout */
864 	else
865 		return rc; /* signal */
866 }
867 
868 /**
869  * iscsi_iser_ep_disconnect() - Initiate connection teardown process
870  * @ep:    iscsi endpoint handle
871  *
872  * This routine is not blocked by iser and RDMA termination process
873  * completion as we queue a deffered work for iser/RDMA destruction
874  * and cleanup or actually call it immediately in case we didn't pass
875  * iscsi conn bind/start stage, thus it is safe.
876  */
877 static void iscsi_iser_ep_disconnect(struct iscsi_endpoint *ep)
878 {
879 	struct iser_conn *iser_conn = ep->dd_data;
880 
881 	iser_info("ep %p iser conn %p\n", ep, iser_conn);
882 
883 	mutex_lock(&iser_conn->state_mutex);
884 	iser_conn_terminate(iser_conn);
885 
886 	/*
887 	 * if iser_conn and iscsi_conn are bound, we must wait for
888 	 * iscsi_conn_stop and flush errors completion before freeing
889 	 * the iser resources. Otherwise we are safe to free resources
890 	 * immediately.
891 	 */
892 	if (iser_conn->iscsi_conn) {
893 		INIT_WORK(&iser_conn->release_work, iser_release_work);
894 		queue_work(release_wq, &iser_conn->release_work);
895 		mutex_unlock(&iser_conn->state_mutex);
896 	} else {
897 		iser_conn->state = ISER_CONN_DOWN;
898 		mutex_unlock(&iser_conn->state_mutex);
899 		iser_conn_release(iser_conn);
900 	}
901 
902 	iscsi_destroy_endpoint(ep);
903 }
904 
905 static umode_t iser_attr_is_visible(int param_type, int param)
906 {
907 	switch (param_type) {
908 	case ISCSI_HOST_PARAM:
909 		switch (param) {
910 		case ISCSI_HOST_PARAM_NETDEV_NAME:
911 		case ISCSI_HOST_PARAM_HWADDRESS:
912 		case ISCSI_HOST_PARAM_INITIATOR_NAME:
913 			return S_IRUGO;
914 		default:
915 			return 0;
916 		}
917 	case ISCSI_PARAM:
918 		switch (param) {
919 		case ISCSI_PARAM_MAX_RECV_DLENGTH:
920 		case ISCSI_PARAM_MAX_XMIT_DLENGTH:
921 		case ISCSI_PARAM_HDRDGST_EN:
922 		case ISCSI_PARAM_DATADGST_EN:
923 		case ISCSI_PARAM_CONN_ADDRESS:
924 		case ISCSI_PARAM_CONN_PORT:
925 		case ISCSI_PARAM_EXP_STATSN:
926 		case ISCSI_PARAM_PERSISTENT_ADDRESS:
927 		case ISCSI_PARAM_PERSISTENT_PORT:
928 		case ISCSI_PARAM_PING_TMO:
929 		case ISCSI_PARAM_RECV_TMO:
930 		case ISCSI_PARAM_INITIAL_R2T_EN:
931 		case ISCSI_PARAM_MAX_R2T:
932 		case ISCSI_PARAM_IMM_DATA_EN:
933 		case ISCSI_PARAM_FIRST_BURST:
934 		case ISCSI_PARAM_MAX_BURST:
935 		case ISCSI_PARAM_PDU_INORDER_EN:
936 		case ISCSI_PARAM_DATASEQ_INORDER_EN:
937 		case ISCSI_PARAM_TARGET_NAME:
938 		case ISCSI_PARAM_TPGT:
939 		case ISCSI_PARAM_USERNAME:
940 		case ISCSI_PARAM_PASSWORD:
941 		case ISCSI_PARAM_USERNAME_IN:
942 		case ISCSI_PARAM_PASSWORD_IN:
943 		case ISCSI_PARAM_FAST_ABORT:
944 		case ISCSI_PARAM_ABORT_TMO:
945 		case ISCSI_PARAM_LU_RESET_TMO:
946 		case ISCSI_PARAM_TGT_RESET_TMO:
947 		case ISCSI_PARAM_IFACE_NAME:
948 		case ISCSI_PARAM_INITIATOR_NAME:
949 		case ISCSI_PARAM_DISCOVERY_SESS:
950 			return S_IRUGO;
951 		default:
952 			return 0;
953 		}
954 	}
955 
956 	return 0;
957 }
958 
959 static const struct scsi_host_template iscsi_iser_sht = {
960 	.module                 = THIS_MODULE,
961 	.name                   = "iSCSI Initiator over iSER",
962 	.queuecommand           = iscsi_queuecommand,
963 	.change_queue_depth	= scsi_change_queue_depth,
964 	.sg_tablesize           = ISCSI_ISER_DEF_SG_TABLESIZE,
965 	.cmd_per_lun            = ISER_DEF_CMD_PER_LUN,
966 	.eh_timed_out		= iscsi_eh_cmd_timed_out,
967 	.eh_abort_handler       = iscsi_eh_abort,
968 	.eh_device_reset_handler= iscsi_eh_device_reset,
969 	.eh_target_reset_handler = iscsi_eh_recover_target,
970 	.target_alloc		= iscsi_target_alloc,
971 	.proc_name              = "iscsi_iser",
972 	.this_id                = -1,
973 	.track_queue_depth	= 1,
974 	.cmd_size		= sizeof(struct iscsi_cmd),
975 };
976 
977 static struct iscsi_transport iscsi_iser_transport = {
978 	.owner                  = THIS_MODULE,
979 	.name                   = "iser",
980 	.caps                   = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_TEXT_NEGO,
981 	/* session management */
982 	.create_session         = iscsi_iser_session_create,
983 	.destroy_session        = iscsi_iser_session_destroy,
984 	/* connection management */
985 	.create_conn            = iscsi_iser_conn_create,
986 	.bind_conn              = iscsi_iser_conn_bind,
987 	.unbind_conn		= iscsi_conn_unbind,
988 	.destroy_conn           = iscsi_conn_teardown,
989 	.attr_is_visible	= iser_attr_is_visible,
990 	.set_param              = iscsi_iser_set_param,
991 	.get_conn_param		= iscsi_conn_get_param,
992 	.get_ep_param		= iscsi_iser_get_ep_param,
993 	.get_session_param	= iscsi_session_get_param,
994 	.start_conn             = iscsi_iser_conn_start,
995 	.stop_conn              = iscsi_iser_conn_stop,
996 	/* iscsi host params */
997 	.get_host_param		= iscsi_host_get_param,
998 	.set_host_param		= iscsi_host_set_param,
999 	/* IO */
1000 	.send_pdu		= iscsi_conn_send_pdu,
1001 	.get_stats		= iscsi_iser_conn_get_stats,
1002 	.init_task		= iscsi_iser_task_init,
1003 	.xmit_task		= iscsi_iser_task_xmit,
1004 	.cleanup_task		= iscsi_iser_cleanup_task,
1005 	.alloc_pdu		= iscsi_iser_pdu_alloc,
1006 	.check_protection	= iscsi_iser_check_protection,
1007 	/* recovery */
1008 	.session_recovery_timedout = iscsi_session_recovery_timedout,
1009 
1010 	.ep_connect             = iscsi_iser_ep_connect,
1011 	.ep_poll                = iscsi_iser_ep_poll,
1012 	.ep_disconnect          = iscsi_iser_ep_disconnect
1013 };
1014 
1015 static int __init iser_init(void)
1016 {
1017 	int err;
1018 
1019 	iser_dbg("Starting iSER datamover...\n");
1020 
1021 	memset(&ig, 0, sizeof(struct iser_global));
1022 
1023 	ig.desc_cache = kmem_cache_create("iser_descriptors",
1024 					  sizeof(struct iser_tx_desc),
1025 					  0, SLAB_HWCACHE_ALIGN,
1026 					  NULL);
1027 	if (ig.desc_cache == NULL)
1028 		return -ENOMEM;
1029 
1030 	/* device init is called only after the first addr resolution */
1031 	mutex_init(&ig.device_list_mutex);
1032 	INIT_LIST_HEAD(&ig.device_list);
1033 	mutex_init(&ig.connlist_mutex);
1034 	INIT_LIST_HEAD(&ig.connlist);
1035 
1036 	release_wq = alloc_workqueue("release workqueue", 0, 0);
1037 	if (!release_wq) {
1038 		iser_err("failed to allocate release workqueue\n");
1039 		err = -ENOMEM;
1040 		goto err_alloc_wq;
1041 	}
1042 
1043 	iscsi_iser_scsi_transport = iscsi_register_transport(
1044 							&iscsi_iser_transport);
1045 	if (!iscsi_iser_scsi_transport) {
1046 		iser_err("iscsi_register_transport failed\n");
1047 		err = -EINVAL;
1048 		goto err_reg;
1049 	}
1050 
1051 	return 0;
1052 
1053 err_reg:
1054 	destroy_workqueue(release_wq);
1055 err_alloc_wq:
1056 	kmem_cache_destroy(ig.desc_cache);
1057 
1058 	return err;
1059 }
1060 
1061 static void __exit iser_exit(void)
1062 {
1063 	struct iser_conn *iser_conn, *n;
1064 	int connlist_empty;
1065 
1066 	iser_dbg("Removing iSER datamover...\n");
1067 	destroy_workqueue(release_wq);
1068 
1069 	mutex_lock(&ig.connlist_mutex);
1070 	connlist_empty = list_empty(&ig.connlist);
1071 	mutex_unlock(&ig.connlist_mutex);
1072 
1073 	if (!connlist_empty) {
1074 		iser_err("Error cleanup stage completed but we still have iser "
1075 			 "connections, destroying them anyway\n");
1076 		list_for_each_entry_safe(iser_conn, n, &ig.connlist,
1077 					 conn_list) {
1078 			iser_conn_release(iser_conn);
1079 		}
1080 	}
1081 
1082 	iscsi_unregister_transport(&iscsi_iser_transport);
1083 	kmem_cache_destroy(ig.desc_cache);
1084 }
1085 
1086 module_init(iser_init);
1087 module_exit(iser_exit);
1088