xref: /linux/drivers/infiniband/ulp/iser/iscsi_iser.c (revision c411ed854584a71b0e86ac3019b60e4789d88086)
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 MODULE_VERSION(DRV_VER);
81 
82 static struct scsi_host_template iscsi_iser_sht;
83 static struct iscsi_transport iscsi_iser_transport;
84 static struct scsi_transport_template *iscsi_iser_scsi_transport;
85 static struct workqueue_struct *release_wq;
86 static DEFINE_MUTEX(unbind_iser_conn_mutex);
87 struct iser_global ig;
88 
89 int iser_debug_level = 0;
90 module_param_named(debug_level, iser_debug_level, int, S_IRUGO | S_IWUSR);
91 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0 (default:disabled)");
92 
93 static unsigned int iscsi_max_lun = 512;
94 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
95 MODULE_PARM_DESC(max_lun, "Max LUNs to allow per session (default:512");
96 
97 unsigned int iser_max_sectors = ISER_DEF_MAX_SECTORS;
98 module_param_named(max_sectors, iser_max_sectors, uint, S_IRUGO | S_IWUSR);
99 MODULE_PARM_DESC(max_sectors, "Max number of sectors in a single scsi command (default:1024");
100 
101 bool iser_always_reg = true;
102 module_param_named(always_register, iser_always_reg, bool, S_IRUGO);
103 MODULE_PARM_DESC(always_register,
104 		 "Always register memory, even for continuous memory regions (default:true)");
105 
106 bool iser_pi_enable = false;
107 module_param_named(pi_enable, iser_pi_enable, bool, S_IRUGO);
108 MODULE_PARM_DESC(pi_enable, "Enable T10-PI offload support (default:disabled)");
109 
110 int iser_pi_guard;
111 module_param_named(pi_guard, iser_pi_guard, int, S_IRUGO);
112 MODULE_PARM_DESC(pi_guard, "T10-PI guard_type [deprecated]");
113 
114 /*
115  * iscsi_iser_recv() - Process a successful recv completion
116  * @conn:         iscsi connection
117  * @hdr:          iscsi header
118  * @rx_data:      buffer containing receive data payload
119  * @rx_data_len:  length of rx_data
120  *
121  * Notes: In case of data length errors or iscsi PDU completion failures
122  *        this routine will signal iscsi layer of connection failure.
123  */
124 void
125 iscsi_iser_recv(struct iscsi_conn *conn, struct iscsi_hdr *hdr,
126 		char *rx_data, int rx_data_len)
127 {
128 	int rc = 0;
129 	int datalen;
130 
131 	/* verify PDU length */
132 	datalen = ntoh24(hdr->dlength);
133 	if (datalen > rx_data_len || (datalen + 4) < rx_data_len) {
134 		iser_err("wrong datalen %d (hdr), %d (IB)\n",
135 			datalen, rx_data_len);
136 		rc = ISCSI_ERR_DATALEN;
137 		goto error;
138 	}
139 
140 	if (datalen != rx_data_len)
141 		iser_dbg("aligned datalen (%d) hdr, %d (IB)\n",
142 			datalen, rx_data_len);
143 
144 	rc = iscsi_complete_pdu(conn, hdr, rx_data, rx_data_len);
145 	if (rc && rc != ISCSI_ERR_NO_SCSI_CMD)
146 		goto error;
147 
148 	return;
149 error:
150 	iscsi_conn_failure(conn, rc);
151 }
152 
153 /**
154  * iscsi_iser_pdu_alloc() - allocate an iscsi-iser PDU
155  * @task:     iscsi task
156  * @opcode:   iscsi command opcode
157  *
158  * Netes: This routine can't fail, just assign iscsi task
159  *        hdr and max hdr size.
160  */
161 static int
162 iscsi_iser_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
163 {
164 	struct iscsi_iser_task *iser_task = task->dd_data;
165 
166 	task->hdr = (struct iscsi_hdr *)&iser_task->desc.iscsi_header;
167 	task->hdr_max = sizeof(iser_task->desc.iscsi_header);
168 
169 	return 0;
170 }
171 
172 /**
173  * iser_initialize_task_headers() - Initialize task headers
174  * @task:       iscsi task
175  * @tx_desc:    iser tx descriptor
176  *
177  * Notes:
178  * This routine may race with iser teardown flow for scsi
179  * error handling TMFs. So for TMF we should acquire the
180  * state mutex to avoid dereferencing the IB device which
181  * may have already been terminated.
182  */
183 int
184 iser_initialize_task_headers(struct iscsi_task *task,
185 			     struct iser_tx_desc *tx_desc)
186 {
187 	struct iser_conn *iser_conn = task->conn->dd_data;
188 	struct iser_device *device = iser_conn->ib_conn.device;
189 	struct iscsi_iser_task *iser_task = task->dd_data;
190 	u64 dma_addr;
191 	const bool mgmt_task = !task->sc && !in_interrupt();
192 	int ret = 0;
193 
194 	if (unlikely(mgmt_task))
195 		mutex_lock(&iser_conn->state_mutex);
196 
197 	if (unlikely(iser_conn->state != ISER_CONN_UP)) {
198 		ret = -ENODEV;
199 		goto out;
200 	}
201 
202 	dma_addr = ib_dma_map_single(device->ib_device, (void *)tx_desc,
203 				ISER_HEADERS_LEN, DMA_TO_DEVICE);
204 	if (ib_dma_mapping_error(device->ib_device, dma_addr)) {
205 		ret = -ENOMEM;
206 		goto out;
207 	}
208 
209 	tx_desc->wr_idx = 0;
210 	tx_desc->mapped = true;
211 	tx_desc->dma_addr = dma_addr;
212 	tx_desc->tx_sg[0].addr   = tx_desc->dma_addr;
213 	tx_desc->tx_sg[0].length = ISER_HEADERS_LEN;
214 	tx_desc->tx_sg[0].lkey   = device->pd->local_dma_lkey;
215 
216 	iser_task->iser_conn = iser_conn;
217 out:
218 	if (unlikely(mgmt_task))
219 		mutex_unlock(&iser_conn->state_mutex);
220 
221 	return ret;
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
234 iscsi_iser_task_init(struct iscsi_task *task)
235 {
236 	struct iscsi_iser_task *iser_task = task->dd_data;
237 	int ret;
238 
239 	ret = iser_initialize_task_headers(task, &iser_task->desc);
240 	if (ret) {
241 		iser_err("Failed to init task %p, err = %d\n",
242 			 iser_task, ret);
243 		return ret;
244 	}
245 
246 	/* mgmt task */
247 	if (!task->sc)
248 		return 0;
249 
250 	iser_task->command_sent = 0;
251 	iser_task_rdma_init(iser_task);
252 	iser_task->sc = task->sc;
253 
254 	return 0;
255 }
256 
257 /**
258  * iscsi_iser_mtask_xmit() - xmit management (immediate) task
259  * @conn: iscsi connection
260  * @task: task management task
261  *
262  * Notes:
263  *	The function can return -EAGAIN in which case caller must
264  *	call it again later, or recover. '0' return code means successful
265  *	xmit.
266  *
267  **/
268 static int
269 iscsi_iser_mtask_xmit(struct iscsi_conn *conn, struct iscsi_task *task)
270 {
271 	int error = 0;
272 
273 	iser_dbg("mtask xmit [cid %d itt 0x%x]\n", conn->id, task->itt);
274 
275 	error = iser_send_control(conn, task);
276 
277 	/* since iser xmits control with zero copy, tasks can not be recycled
278 	 * right after sending them.
279 	 * The recycling scheme is based on whether a response is expected
280 	 * - if yes, the task is recycled at iscsi_complete_pdu
281 	 * - if no,  the task is recycled at iser_snd_completion
282 	 */
283 	return error;
284 }
285 
286 static int
287 iscsi_iser_task_xmit_unsol_data(struct iscsi_conn *conn,
288 				 struct iscsi_task *task)
289 {
290 	struct iscsi_r2t_info *r2t = &task->unsol_r2t;
291 	struct iscsi_data hdr;
292 	int error = 0;
293 
294 	/* Send data-out PDUs while there's still unsolicited data to send */
295 	while (iscsi_task_has_unsol_data(task)) {
296 		iscsi_prep_data_out_pdu(task, r2t, &hdr);
297 		iser_dbg("Sending data-out: itt 0x%x, data count %d\n",
298 			   hdr.itt, r2t->data_count);
299 
300 		/* the buffer description has been passed with the command */
301 		/* Send the command */
302 		error = iser_send_data_out(conn, task, &hdr);
303 		if (error) {
304 			r2t->datasn--;
305 			goto iscsi_iser_task_xmit_unsol_data_exit;
306 		}
307 		r2t->sent += r2t->data_count;
308 		iser_dbg("Need to send %d more as data-out PDUs\n",
309 			   r2t->data_length - r2t->sent);
310 	}
311 
312 iscsi_iser_task_xmit_unsol_data_exit:
313 	return error;
314 }
315 
316 /**
317  * iscsi_iser_task_xmit() - xmit iscsi-iser task
318  * @task: iscsi task
319  *
320  * Return: zero on success or escalates $error on failure.
321  */
322 static int
323 iscsi_iser_task_xmit(struct iscsi_task *task)
324 {
325 	struct iscsi_conn *conn = task->conn;
326 	struct iscsi_iser_task *iser_task = task->dd_data;
327 	int error = 0;
328 
329 	if (!task->sc)
330 		return iscsi_iser_mtask_xmit(conn, task);
331 
332 	if (task->sc->sc_data_direction == DMA_TO_DEVICE) {
333 		BUG_ON(scsi_bufflen(task->sc) == 0);
334 
335 		iser_dbg("cmd [itt %x total %d imm %d unsol_data %d\n",
336 			   task->itt, scsi_bufflen(task->sc),
337 			   task->imm_count, task->unsol_r2t.data_length);
338 	}
339 
340 	iser_dbg("ctask xmit [cid %d itt 0x%x]\n",
341 		   conn->id, task->itt);
342 
343 	/* Send the cmd PDU */
344 	if (!iser_task->command_sent) {
345 		error = iser_send_command(conn, task);
346 		if (error)
347 			goto iscsi_iser_task_xmit_exit;
348 		iser_task->command_sent = 1;
349 	}
350 
351 	/* Send unsolicited data-out PDU(s) if necessary */
352 	if (iscsi_task_has_unsol_data(task))
353 		error = iscsi_iser_task_xmit_unsol_data(conn, task);
354 
355  iscsi_iser_task_xmit_exit:
356 	return error;
357 }
358 
359 /**
360  * iscsi_iser_cleanup_task() - cleanup an iscsi-iser task
361  * @task: iscsi task
362  *
363  * Notes: In case the RDMA device is already NULL (might have
364  *        been removed in DEVICE_REMOVAL CM event it will bail-out
365  *        without doing dma unmapping.
366  */
367 static void iscsi_iser_cleanup_task(struct iscsi_task *task)
368 {
369 	struct iscsi_iser_task *iser_task = task->dd_data;
370 	struct iser_tx_desc *tx_desc = &iser_task->desc;
371 	struct iser_conn *iser_conn = task->conn->dd_data;
372 	struct iser_device *device = iser_conn->ib_conn.device;
373 
374 	/* DEVICE_REMOVAL event might have already released the device */
375 	if (!device)
376 		return;
377 
378 	if (likely(tx_desc->mapped)) {
379 		ib_dma_unmap_single(device->ib_device, tx_desc->dma_addr,
380 				    ISER_HEADERS_LEN, DMA_TO_DEVICE);
381 		tx_desc->mapped = false;
382 	}
383 
384 	/* mgmt tasks do not need special cleanup */
385 	if (!task->sc)
386 		return;
387 
388 	if (iser_task->status == ISER_TASK_STATUS_STARTED) {
389 		iser_task->status = ISER_TASK_STATUS_COMPLETED;
390 		iser_task_rdma_finalize(iser_task);
391 	}
392 }
393 
394 /**
395  * iscsi_iser_check_protection() - check protection information status of task.
396  * @task:     iscsi task
397  * @sector:   error sector if exsists (output)
398  *
399  * Return: zero if no data-integrity errors have occured
400  *         0x1: data-integrity error occured in the guard-block
401  *         0x2: data-integrity error occured in the reference tag
402  *         0x3: data-integrity error occured in the application tag
403  *
404  *         In addition the error sector is marked.
405  */
406 static u8
407 iscsi_iser_check_protection(struct iscsi_task *task, sector_t *sector)
408 {
409 	struct iscsi_iser_task *iser_task = task->dd_data;
410 
411 	if (iser_task->dir[ISER_DIR_IN])
412 		return iser_check_task_pi_status(iser_task, ISER_DIR_IN,
413 						 sector);
414 	else
415 		return iser_check_task_pi_status(iser_task, ISER_DIR_OUT,
416 						 sector);
417 }
418 
419 /**
420  * iscsi_iser_conn_create() - create a new iscsi-iser connection
421  * @cls_session: iscsi class connection
422  * @conn_idx:    connection index within the session (for MCS)
423  *
424  * Return: iscsi_cls_conn when iscsi_conn_setup succeeds or NULL
425  *         otherwise.
426  */
427 static struct iscsi_cls_conn *
428 iscsi_iser_conn_create(struct iscsi_cls_session *cls_session,
429 		       uint32_t conn_idx)
430 {
431 	struct iscsi_conn *conn;
432 	struct iscsi_cls_conn *cls_conn;
433 
434 	cls_conn = iscsi_conn_setup(cls_session, 0, conn_idx);
435 	if (!cls_conn)
436 		return NULL;
437 	conn = cls_conn->dd_data;
438 
439 	/*
440 	 * due to issues with the login code re iser sematics
441 	 * this not set in iscsi_conn_setup - FIXME
442 	 */
443 	conn->max_recv_dlength = ISER_RECV_DATA_SEG_LEN;
444 
445 	return cls_conn;
446 }
447 
448 /**
449  * iscsi_iser_conn_bind() - bind iscsi and iser connection structures
450  * @cls_session:     iscsi class session
451  * @cls_conn:        iscsi class connection
452  * @transport_eph:   transport end-point handle
453  * @is_leading:      indicate if this is the session leading connection (MCS)
454  *
455  * Return: zero on success, $error if iscsi_conn_bind fails and
456  *         -EINVAL in case end-point doesn't exsits anymore or iser connection
457  *         state is not UP (teardown already started).
458  */
459 static int
460 iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session,
461 		     struct iscsi_cls_conn *cls_conn,
462 		     uint64_t transport_eph,
463 		     int is_leading)
464 {
465 	struct iscsi_conn *conn = cls_conn->dd_data;
466 	struct iser_conn *iser_conn;
467 	struct iscsi_endpoint *ep;
468 	int error;
469 
470 	error = iscsi_conn_bind(cls_session, cls_conn, is_leading);
471 	if (error)
472 		return error;
473 
474 	/* the transport ep handle comes from user space so it must be
475 	 * verified against the global ib connections list */
476 	ep = iscsi_lookup_endpoint(transport_eph);
477 	if (!ep) {
478 		iser_err("can't bind eph %llx\n",
479 			 (unsigned long long)transport_eph);
480 		return -EINVAL;
481 	}
482 	iser_conn = ep->dd_data;
483 
484 	mutex_lock(&iser_conn->state_mutex);
485 	if (iser_conn->state != ISER_CONN_UP) {
486 		error = -EINVAL;
487 		iser_err("iser_conn %p state is %d, teardown started\n",
488 			 iser_conn, iser_conn->state);
489 		goto out;
490 	}
491 
492 	error = iser_alloc_rx_descriptors(iser_conn, conn->session);
493 	if (error)
494 		goto out;
495 
496 	/* binds the iSER connection retrieved from the previously
497 	 * connected ep_handle to the iSCSI layer connection. exchanges
498 	 * connection pointers */
499 	iser_info("binding iscsi conn %p to iser_conn %p\n", conn, iser_conn);
500 
501 	conn->dd_data = iser_conn;
502 	iser_conn->iscsi_conn = conn;
503 
504 out:
505 	mutex_unlock(&iser_conn->state_mutex);
506 	return error;
507 }
508 
509 /**
510  * iscsi_iser_conn_start() - start iscsi-iser connection
511  * @cls_conn: iscsi class connection
512  *
513  * Notes: Here iser intialize (or re-initialize) stop_completion as
514  *        from this point iscsi must call conn_stop in session/connection
515  *        teardown so iser transport must wait for it.
516  */
517 static int
518 iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn)
519 {
520 	struct iscsi_conn *iscsi_conn;
521 	struct iser_conn *iser_conn;
522 
523 	iscsi_conn = cls_conn->dd_data;
524 	iser_conn = iscsi_conn->dd_data;
525 	reinit_completion(&iser_conn->stop_completion);
526 
527 	return iscsi_conn_start(cls_conn);
528 }
529 
530 /**
531  * iscsi_iser_conn_stop() - stop iscsi-iser connection
532  * @cls_conn:  iscsi class connection
533  * @flag:      indicate if recover or terminate (passed as is)
534  *
535  * Notes: Calling iscsi_conn_stop might theoretically race with
536  *        DEVICE_REMOVAL event and dereference a previously freed RDMA device
537  *        handle, so we call it under iser the state lock to protect against
538  *        this kind of race.
539  */
540 static void
541 iscsi_iser_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
542 {
543 	struct iscsi_conn *conn = cls_conn->dd_data;
544 	struct iser_conn *iser_conn = conn->dd_data;
545 
546 	iser_info("stopping iscsi_conn: %p, iser_conn: %p\n", conn, iser_conn);
547 
548 	/*
549 	 * Userspace may have goofed up and not bound the connection or
550 	 * might have only partially setup the connection.
551 	 */
552 	if (iser_conn) {
553 		mutex_lock(&iser_conn->state_mutex);
554 		mutex_lock(&unbind_iser_conn_mutex);
555 		iser_conn_terminate(iser_conn);
556 		iscsi_conn_stop(cls_conn, flag);
557 
558 		/* unbind */
559 		iser_conn->iscsi_conn = NULL;
560 		conn->dd_data = NULL;
561 		mutex_unlock(&unbind_iser_conn_mutex);
562 
563 		complete(&iser_conn->stop_completion);
564 		mutex_unlock(&iser_conn->state_mutex);
565 	} else {
566 		iscsi_conn_stop(cls_conn, flag);
567 	}
568 }
569 
570 /**
571  * iscsi_iser_session_destroy() - destroy iscsi-iser session
572  * @cls_session: iscsi class session
573  *
574  * Removes and free iscsi host.
575  */
576 static void
577 iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session)
578 {
579 	struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
580 
581 	iscsi_session_teardown(cls_session);
582 	iscsi_host_remove(shost);
583 	iscsi_host_free(shost);
584 }
585 
586 static inline unsigned int
587 iser_dif_prot_caps(int prot_caps)
588 {
589 	return ((prot_caps & IB_PROT_T10DIF_TYPE_1) ?
590 		SHOST_DIF_TYPE1_PROTECTION | SHOST_DIX_TYPE0_PROTECTION |
591 		SHOST_DIX_TYPE1_PROTECTION : 0) |
592 	       ((prot_caps & IB_PROT_T10DIF_TYPE_2) ?
593 		SHOST_DIF_TYPE2_PROTECTION | SHOST_DIX_TYPE2_PROTECTION : 0) |
594 	       ((prot_caps & IB_PROT_T10DIF_TYPE_3) ?
595 		SHOST_DIF_TYPE3_PROTECTION | SHOST_DIX_TYPE3_PROTECTION : 0);
596 }
597 
598 /**
599  * iscsi_iser_session_create() - create an iscsi-iser session
600  * @ep:             iscsi end-point handle
601  * @cmds_max:       maximum commands in this session
602  * @qdepth:         session command queue depth
603  * @initial_cmdsn:  initiator command sequnce number
604  *
605  * Allocates and adds a scsi host, expose DIF supprot if
606  * exists, and sets up an iscsi session.
607  */
608 static struct iscsi_cls_session *
609 iscsi_iser_session_create(struct iscsi_endpoint *ep,
610 			  uint16_t cmds_max, uint16_t qdepth,
611 			  uint32_t initial_cmdsn)
612 {
613 	struct iscsi_cls_session *cls_session;
614 	struct iscsi_session *session;
615 	struct Scsi_Host *shost;
616 	struct iser_conn *iser_conn = NULL;
617 	struct ib_conn *ib_conn;
618 	u32 max_fr_sectors;
619 	u16 max_cmds;
620 
621 	shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 0);
622 	if (!shost)
623 		return NULL;
624 	shost->transportt = iscsi_iser_scsi_transport;
625 	shost->cmd_per_lun = qdepth;
626 	shost->max_lun = iscsi_max_lun;
627 	shost->max_id = 0;
628 	shost->max_channel = 0;
629 	shost->max_cmd_len = 16;
630 
631 	/*
632 	 * older userspace tools (before 2.0-870) did not pass us
633 	 * the leading conn's ep so this will be NULL;
634 	 */
635 	if (ep) {
636 		iser_conn = ep->dd_data;
637 		max_cmds = iser_conn->max_cmds;
638 		shost->sg_tablesize = iser_conn->scsi_sg_tablesize;
639 
640 		mutex_lock(&iser_conn->state_mutex);
641 		if (iser_conn->state != ISER_CONN_UP) {
642 			iser_err("iser conn %p already started teardown\n",
643 				 iser_conn);
644 			mutex_unlock(&iser_conn->state_mutex);
645 			goto free_host;
646 		}
647 
648 		ib_conn = &iser_conn->ib_conn;
649 		if (ib_conn->pi_support) {
650 			u32 sig_caps = ib_conn->device->ib_device->attrs.sig_prot_cap;
651 
652 			scsi_host_set_prot(shost, iser_dif_prot_caps(sig_caps));
653 			scsi_host_set_guard(shost, SHOST_DIX_GUARD_IP |
654 						   SHOST_DIX_GUARD_CRC);
655 		}
656 
657 		if (iscsi_host_add(shost,
658 				   ib_conn->device->ib_device->dev.parent)) {
659 			mutex_unlock(&iser_conn->state_mutex);
660 			goto free_host;
661 		}
662 		mutex_unlock(&iser_conn->state_mutex);
663 	} else {
664 		max_cmds = ISER_DEF_XMIT_CMDS_MAX;
665 		if (iscsi_host_add(shost, NULL))
666 			goto free_host;
667 	}
668 
669 	/*
670 	 * FRs or FMRs can only map up to a (device) page per entry, but if the
671 	 * first entry is misaligned we'll end up using using two entries
672 	 * (head and tail) for a single page worth data, so we have to drop
673 	 * one segment from the calculation.
674 	 */
675 	max_fr_sectors = ((shost->sg_tablesize - 1) * PAGE_SIZE) >> 9;
676 	shost->max_sectors = min(iser_max_sectors, max_fr_sectors);
677 
678 	iser_dbg("iser_conn %p, sg_tablesize %u, max_sectors %u\n",
679 		 iser_conn, shost->sg_tablesize,
680 		 shost->max_sectors);
681 
682 	if (cmds_max > max_cmds) {
683 		iser_info("cmds_max changed from %u to %u\n",
684 			  cmds_max, max_cmds);
685 		cmds_max = max_cmds;
686 	}
687 
688 	cls_session = iscsi_session_setup(&iscsi_iser_transport, shost,
689 					  cmds_max, 0,
690 					  sizeof(struct iscsi_iser_task),
691 					  initial_cmdsn, 0);
692 	if (!cls_session)
693 		goto remove_host;
694 	session = cls_session->dd_data;
695 
696 	shost->can_queue = session->scsi_cmds_max;
697 	return cls_session;
698 
699 remove_host:
700 	iscsi_host_remove(shost);
701 free_host:
702 	iscsi_host_free(shost);
703 	return NULL;
704 }
705 
706 static int
707 iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn,
708 		     enum iscsi_param param, char *buf, int buflen)
709 {
710 	int value;
711 
712 	switch (param) {
713 	case ISCSI_PARAM_MAX_RECV_DLENGTH:
714 		/* TBD */
715 		break;
716 	case ISCSI_PARAM_HDRDGST_EN:
717 		sscanf(buf, "%d", &value);
718 		if (value) {
719 			iser_err("DataDigest wasn't negotiated to None\n");
720 			return -EPROTO;
721 		}
722 		break;
723 	case ISCSI_PARAM_DATADGST_EN:
724 		sscanf(buf, "%d", &value);
725 		if (value) {
726 			iser_err("DataDigest wasn't negotiated to None\n");
727 			return -EPROTO;
728 		}
729 		break;
730 	case ISCSI_PARAM_IFMARKER_EN:
731 		sscanf(buf, "%d", &value);
732 		if (value) {
733 			iser_err("IFMarker wasn't negotiated to No\n");
734 			return -EPROTO;
735 		}
736 		break;
737 	case ISCSI_PARAM_OFMARKER_EN:
738 		sscanf(buf, "%d", &value);
739 		if (value) {
740 			iser_err("OFMarker wasn't negotiated to No\n");
741 			return -EPROTO;
742 		}
743 		break;
744 	default:
745 		return iscsi_set_param(cls_conn, param, buf, buflen);
746 	}
747 
748 	return 0;
749 }
750 
751 /**
752  * iscsi_iser_set_param() - set class connection parameter
753  * @cls_conn:    iscsi class connection
754  * @stats:       iscsi stats to output
755  *
756  * Output connection statistics.
757  */
758 static void
759 iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
760 {
761 	struct iscsi_conn *conn = cls_conn->dd_data;
762 
763 	stats->txdata_octets = conn->txdata_octets;
764 	stats->rxdata_octets = conn->rxdata_octets;
765 	stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
766 	stats->dataout_pdus = conn->dataout_pdus_cnt;
767 	stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
768 	stats->datain_pdus = conn->datain_pdus_cnt; /* always 0 */
769 	stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */
770 	stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
771 	stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
772 	stats->custom_length = 0;
773 }
774 
775 static int iscsi_iser_get_ep_param(struct iscsi_endpoint *ep,
776 				   enum iscsi_param param, char *buf)
777 {
778 	struct iser_conn *iser_conn = ep->dd_data;
779 	int len;
780 
781 	switch (param) {
782 	case ISCSI_PARAM_CONN_PORT:
783 	case ISCSI_PARAM_CONN_ADDRESS:
784 		if (!iser_conn || !iser_conn->ib_conn.cma_id)
785 			return -ENOTCONN;
786 
787 		return iscsi_conn_get_addr_param((struct sockaddr_storage *)
788 				&iser_conn->ib_conn.cma_id->route.addr.dst_addr,
789 				param, buf);
790 		break;
791 	default:
792 		return -ENOSYS;
793 	}
794 
795 	return len;
796 }
797 
798 /**
799  * iscsi_iser_ep_connect() - Initiate iSER connection establishment
800  * @shost:          scsi_host
801  * @dst_addr:       destination address
802  * @non-blocking:   indicate if routine can block
803  *
804  * Allocate an iscsi endpoint, an iser_conn structure and bind them.
805  * After that start RDMA connection establishment via rdma_cm. We
806  * don't allocate iser_conn embedded in iscsi_endpoint since in teardown
807  * the endpoint will be destroyed at ep_disconnect while iser_conn will
808  * cleanup its resources asynchronuously.
809  *
810  * Return: iscsi_endpoint created by iscsi layer or ERR_PTR(error)
811  *         if fails.
812  */
813 static struct iscsi_endpoint *
814 iscsi_iser_ep_connect(struct Scsi_Host *shost, struct sockaddr *dst_addr,
815 		      int non_blocking)
816 {
817 	int err;
818 	struct iser_conn *iser_conn;
819 	struct iscsi_endpoint *ep;
820 
821 	ep = iscsi_create_endpoint(0);
822 	if (!ep)
823 		return ERR_PTR(-ENOMEM);
824 
825 	iser_conn = kzalloc(sizeof(*iser_conn), GFP_KERNEL);
826 	if (!iser_conn) {
827 		err = -ENOMEM;
828 		goto failure;
829 	}
830 
831 	ep->dd_data = iser_conn;
832 	iser_conn->ep = ep;
833 	iser_conn_init(iser_conn);
834 
835 	err = iser_connect(iser_conn, NULL, dst_addr, non_blocking);
836 	if (err)
837 		goto failure;
838 
839 	return ep;
840 failure:
841 	iscsi_destroy_endpoint(ep);
842 	return ERR_PTR(err);
843 }
844 
845 /**
846  * iscsi_iser_ep_poll() - poll for iser connection establishment to complete
847  * @ep:            iscsi endpoint (created at ep_connect)
848  * @timeout_ms:    polling timeout allowed in ms.
849  *
850  * This routine boils down to waiting for up_completion signaling
851  * that cma_id got CONNECTED event.
852  *
853  * Return: 1 if succeeded in connection establishment, 0 if timeout expired
854  *         (libiscsi will retry will kick in) or -1 if interrupted by signal
855  *         or more likely iser connection state transitioned to TEMINATING or
856  *         DOWN during the wait period.
857  */
858 static int
859 iscsi_iser_ep_poll(struct iscsi_endpoint *ep, int timeout_ms)
860 {
861 	struct iser_conn *iser_conn = ep->dd_data;
862 	int rc;
863 
864 	rc = wait_for_completion_interruptible_timeout(&iser_conn->up_completion,
865 						       msecs_to_jiffies(timeout_ms));
866 	/* if conn establishment failed, return error code to iscsi */
867 	if (rc == 0) {
868 		mutex_lock(&iser_conn->state_mutex);
869 		if (iser_conn->state == ISER_CONN_TERMINATING ||
870 		    iser_conn->state == ISER_CONN_DOWN)
871 			rc = -1;
872 		mutex_unlock(&iser_conn->state_mutex);
873 	}
874 
875 	iser_info("iser conn %p rc = %d\n", iser_conn, rc);
876 
877 	if (rc > 0)
878 		return 1; /* success, this is the equivalent of POLLOUT */
879 	else if (!rc)
880 		return 0; /* timeout */
881 	else
882 		return rc; /* signal */
883 }
884 
885 /**
886  * iscsi_iser_ep_disconnect() - Initiate connection teardown process
887  * @ep:    iscsi endpoint handle
888  *
889  * This routine is not blocked by iser and RDMA termination process
890  * completion as we queue a deffered work for iser/RDMA destruction
891  * and cleanup or actually call it immediately in case we didn't pass
892  * iscsi conn bind/start stage, thus it is safe.
893  */
894 static void
895 iscsi_iser_ep_disconnect(struct iscsi_endpoint *ep)
896 {
897 	struct iser_conn *iser_conn = ep->dd_data;
898 
899 	iser_info("ep %p iser conn %p\n", ep, iser_conn);
900 
901 	mutex_lock(&iser_conn->state_mutex);
902 	iser_conn_terminate(iser_conn);
903 
904 	/*
905 	 * if iser_conn and iscsi_conn are bound, we must wait for
906 	 * iscsi_conn_stop and flush errors completion before freeing
907 	 * the iser resources. Otherwise we are safe to free resources
908 	 * immediately.
909 	 */
910 	if (iser_conn->iscsi_conn) {
911 		INIT_WORK(&iser_conn->release_work, iser_release_work);
912 		queue_work(release_wq, &iser_conn->release_work);
913 		mutex_unlock(&iser_conn->state_mutex);
914 	} else {
915 		iser_conn->state = ISER_CONN_DOWN;
916 		mutex_unlock(&iser_conn->state_mutex);
917 		iser_conn_release(iser_conn);
918 	}
919 
920 	iscsi_destroy_endpoint(ep);
921 }
922 
923 static umode_t iser_attr_is_visible(int param_type, int param)
924 {
925 	switch (param_type) {
926 	case ISCSI_HOST_PARAM:
927 		switch (param) {
928 		case ISCSI_HOST_PARAM_NETDEV_NAME:
929 		case ISCSI_HOST_PARAM_HWADDRESS:
930 		case ISCSI_HOST_PARAM_INITIATOR_NAME:
931 			return S_IRUGO;
932 		default:
933 			return 0;
934 		}
935 	case ISCSI_PARAM:
936 		switch (param) {
937 		case ISCSI_PARAM_MAX_RECV_DLENGTH:
938 		case ISCSI_PARAM_MAX_XMIT_DLENGTH:
939 		case ISCSI_PARAM_HDRDGST_EN:
940 		case ISCSI_PARAM_DATADGST_EN:
941 		case ISCSI_PARAM_CONN_ADDRESS:
942 		case ISCSI_PARAM_CONN_PORT:
943 		case ISCSI_PARAM_EXP_STATSN:
944 		case ISCSI_PARAM_PERSISTENT_ADDRESS:
945 		case ISCSI_PARAM_PERSISTENT_PORT:
946 		case ISCSI_PARAM_PING_TMO:
947 		case ISCSI_PARAM_RECV_TMO:
948 		case ISCSI_PARAM_INITIAL_R2T_EN:
949 		case ISCSI_PARAM_MAX_R2T:
950 		case ISCSI_PARAM_IMM_DATA_EN:
951 		case ISCSI_PARAM_FIRST_BURST:
952 		case ISCSI_PARAM_MAX_BURST:
953 		case ISCSI_PARAM_PDU_INORDER_EN:
954 		case ISCSI_PARAM_DATASEQ_INORDER_EN:
955 		case ISCSI_PARAM_TARGET_NAME:
956 		case ISCSI_PARAM_TPGT:
957 		case ISCSI_PARAM_USERNAME:
958 		case ISCSI_PARAM_PASSWORD:
959 		case ISCSI_PARAM_USERNAME_IN:
960 		case ISCSI_PARAM_PASSWORD_IN:
961 		case ISCSI_PARAM_FAST_ABORT:
962 		case ISCSI_PARAM_ABORT_TMO:
963 		case ISCSI_PARAM_LU_RESET_TMO:
964 		case ISCSI_PARAM_TGT_RESET_TMO:
965 		case ISCSI_PARAM_IFACE_NAME:
966 		case ISCSI_PARAM_INITIATOR_NAME:
967 		case ISCSI_PARAM_DISCOVERY_SESS:
968 			return S_IRUGO;
969 		default:
970 			return 0;
971 		}
972 	}
973 
974 	return 0;
975 }
976 
977 static int iscsi_iser_slave_alloc(struct scsi_device *sdev)
978 {
979 	struct iscsi_session *session;
980 	struct iser_conn *iser_conn;
981 	struct ib_device *ib_dev;
982 
983 	mutex_lock(&unbind_iser_conn_mutex);
984 
985 	session = starget_to_session(scsi_target(sdev))->dd_data;
986 	iser_conn = session->leadconn->dd_data;
987 	if (!iser_conn) {
988 		mutex_unlock(&unbind_iser_conn_mutex);
989 		return -ENOTCONN;
990 	}
991 	ib_dev = iser_conn->ib_conn.device->ib_device;
992 
993 	if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_SG_GAPS_REG))
994 		blk_queue_virt_boundary(sdev->request_queue, ~MASK_4K);
995 
996 	mutex_unlock(&unbind_iser_conn_mutex);
997 
998 	return 0;
999 }
1000 
1001 static struct scsi_host_template iscsi_iser_sht = {
1002 	.module                 = THIS_MODULE,
1003 	.name                   = "iSCSI Initiator over iSER",
1004 	.queuecommand           = iscsi_queuecommand,
1005 	.change_queue_depth	= scsi_change_queue_depth,
1006 	.sg_tablesize           = ISCSI_ISER_DEF_SG_TABLESIZE,
1007 	.cmd_per_lun            = ISER_DEF_CMD_PER_LUN,
1008 	.eh_timed_out		= iscsi_eh_cmd_timed_out,
1009 	.eh_abort_handler       = iscsi_eh_abort,
1010 	.eh_device_reset_handler= iscsi_eh_device_reset,
1011 	.eh_target_reset_handler = iscsi_eh_recover_target,
1012 	.target_alloc		= iscsi_target_alloc,
1013 	.use_clustering         = ENABLE_CLUSTERING,
1014 	.slave_alloc            = iscsi_iser_slave_alloc,
1015 	.proc_name              = "iscsi_iser",
1016 	.this_id                = -1,
1017 	.track_queue_depth	= 1,
1018 };
1019 
1020 static struct iscsi_transport iscsi_iser_transport = {
1021 	.owner                  = THIS_MODULE,
1022 	.name                   = "iser",
1023 	.caps                   = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_TEXT_NEGO,
1024 	/* session management */
1025 	.create_session         = iscsi_iser_session_create,
1026 	.destroy_session        = iscsi_iser_session_destroy,
1027 	/* connection management */
1028 	.create_conn            = iscsi_iser_conn_create,
1029 	.bind_conn              = iscsi_iser_conn_bind,
1030 	.destroy_conn           = iscsi_conn_teardown,
1031 	.attr_is_visible	= iser_attr_is_visible,
1032 	.set_param              = iscsi_iser_set_param,
1033 	.get_conn_param		= iscsi_conn_get_param,
1034 	.get_ep_param		= iscsi_iser_get_ep_param,
1035 	.get_session_param	= iscsi_session_get_param,
1036 	.start_conn             = iscsi_iser_conn_start,
1037 	.stop_conn              = iscsi_iser_conn_stop,
1038 	/* iscsi host params */
1039 	.get_host_param		= iscsi_host_get_param,
1040 	.set_host_param		= iscsi_host_set_param,
1041 	/* IO */
1042 	.send_pdu		= iscsi_conn_send_pdu,
1043 	.get_stats		= iscsi_iser_conn_get_stats,
1044 	.init_task		= iscsi_iser_task_init,
1045 	.xmit_task		= iscsi_iser_task_xmit,
1046 	.cleanup_task		= iscsi_iser_cleanup_task,
1047 	.alloc_pdu		= iscsi_iser_pdu_alloc,
1048 	.check_protection	= iscsi_iser_check_protection,
1049 	/* recovery */
1050 	.session_recovery_timedout = iscsi_session_recovery_timedout,
1051 
1052 	.ep_connect             = iscsi_iser_ep_connect,
1053 	.ep_poll                = iscsi_iser_ep_poll,
1054 	.ep_disconnect          = iscsi_iser_ep_disconnect
1055 };
1056 
1057 static int __init iser_init(void)
1058 {
1059 	int err;
1060 
1061 	iser_dbg("Starting iSER datamover...\n");
1062 
1063 	if (iscsi_max_lun < 1) {
1064 		iser_err("Invalid max_lun value of %u\n", iscsi_max_lun);
1065 		return -EINVAL;
1066 	}
1067 
1068 	memset(&ig, 0, sizeof(struct iser_global));
1069 
1070 	ig.desc_cache = kmem_cache_create("iser_descriptors",
1071 					  sizeof(struct iser_tx_desc),
1072 					  0, SLAB_HWCACHE_ALIGN,
1073 					  NULL);
1074 	if (ig.desc_cache == NULL)
1075 		return -ENOMEM;
1076 
1077 	/* device init is called only after the first addr resolution */
1078 	mutex_init(&ig.device_list_mutex);
1079 	INIT_LIST_HEAD(&ig.device_list);
1080 	mutex_init(&ig.connlist_mutex);
1081 	INIT_LIST_HEAD(&ig.connlist);
1082 
1083 	release_wq = alloc_workqueue("release workqueue", 0, 0);
1084 	if (!release_wq) {
1085 		iser_err("failed to allocate release workqueue\n");
1086 		err = -ENOMEM;
1087 		goto err_alloc_wq;
1088 	}
1089 
1090 	iscsi_iser_scsi_transport = iscsi_register_transport(
1091 							&iscsi_iser_transport);
1092 	if (!iscsi_iser_scsi_transport) {
1093 		iser_err("iscsi_register_transport failed\n");
1094 		err = -EINVAL;
1095 		goto err_reg;
1096 	}
1097 
1098 	return 0;
1099 
1100 err_reg:
1101 	destroy_workqueue(release_wq);
1102 err_alloc_wq:
1103 	kmem_cache_destroy(ig.desc_cache);
1104 
1105 	return err;
1106 }
1107 
1108 static void __exit iser_exit(void)
1109 {
1110 	struct iser_conn *iser_conn, *n;
1111 	int connlist_empty;
1112 
1113 	iser_dbg("Removing iSER datamover...\n");
1114 	destroy_workqueue(release_wq);
1115 
1116 	mutex_lock(&ig.connlist_mutex);
1117 	connlist_empty = list_empty(&ig.connlist);
1118 	mutex_unlock(&ig.connlist_mutex);
1119 
1120 	if (!connlist_empty) {
1121 		iser_err("Error cleanup stage completed but we still have iser "
1122 			 "connections, destroying them anyway\n");
1123 		list_for_each_entry_safe(iser_conn, n, &ig.connlist,
1124 					 conn_list) {
1125 			iser_conn_release(iser_conn);
1126 		}
1127 	}
1128 
1129 	iscsi_unregister_transport(&iscsi_iser_transport);
1130 	kmem_cache_destroy(ig.desc_cache);
1131 }
1132 
1133 module_init(iser_init);
1134 module_exit(iser_exit);
1135