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 <asm/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 static struct scsi_host_template iscsi_iser_sht;
78 static struct iscsi_transport iscsi_iser_transport;
79 static struct scsi_transport_template *iscsi_iser_scsi_transport;
80 
81 static unsigned int iscsi_max_lun = 512;
82 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
83 
84 int iser_debug_level = 0;
85 bool iser_pi_enable = false;
86 int iser_pi_guard = 1;
87 
88 MODULE_DESCRIPTION("iSER (iSCSI Extensions for RDMA) Datamover");
89 MODULE_LICENSE("Dual BSD/GPL");
90 MODULE_AUTHOR("Alex Nezhinsky, Dan Bar Dov, Or Gerlitz");
91 MODULE_VERSION(DRV_VER);
92 
93 module_param_named(debug_level, iser_debug_level, int, 0644);
94 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0 (default:disabled)");
95 
96 module_param_named(pi_enable, iser_pi_enable, bool, 0644);
97 MODULE_PARM_DESC(pi_enable, "Enable T10-PI offload support (default:disabled)");
98 
99 module_param_named(pi_guard, iser_pi_guard, int, 0644);
100 MODULE_PARM_DESC(pi_guard, "T10-PI guard_type [deprecated]");
101 
102 static struct workqueue_struct *release_wq;
103 struct iser_global ig;
104 
105 /*
106  * iscsi_iser_recv() - Process a successfull recv completion
107  * @conn:         iscsi connection
108  * @hdr:          iscsi header
109  * @rx_data:      buffer containing receive data payload
110  * @rx_data_len:  length of rx_data
111  *
112  * Notes: In case of data length errors or iscsi PDU completion failures
113  *        this routine will signal iscsi layer of connection failure.
114  */
115 void
116 iscsi_iser_recv(struct iscsi_conn *conn, struct iscsi_hdr *hdr,
117 		char *rx_data, int rx_data_len)
118 {
119 	int rc = 0;
120 	int datalen;
121 	int ahslen;
122 
123 	/* verify PDU length */
124 	datalen = ntoh24(hdr->dlength);
125 	if (datalen > rx_data_len || (datalen + 4) < rx_data_len) {
126 		iser_err("wrong datalen %d (hdr), %d (IB)\n",
127 			datalen, rx_data_len);
128 		rc = ISCSI_ERR_DATALEN;
129 		goto error;
130 	}
131 
132 	if (datalen != rx_data_len)
133 		iser_dbg("aligned datalen (%d) hdr, %d (IB)\n",
134 			datalen, rx_data_len);
135 
136 	/* read AHS */
137 	ahslen = hdr->hlength * 4;
138 
139 	rc = iscsi_complete_pdu(conn, hdr, rx_data, rx_data_len);
140 	if (rc && rc != ISCSI_ERR_NO_SCSI_CMD)
141 		goto error;
142 
143 	return;
144 error:
145 	iscsi_conn_failure(conn, rc);
146 }
147 
148 /**
149  * iscsi_iser_pdu_alloc() - allocate an iscsi-iser PDU
150  * @task:     iscsi task
151  * @opcode:   iscsi command opcode
152  *
153  * Netes: This routine can't fail, just assign iscsi task
154  *        hdr and max hdr size.
155  */
156 static int
157 iscsi_iser_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
158 {
159 	struct iscsi_iser_task *iser_task = task->dd_data;
160 
161 	task->hdr = (struct iscsi_hdr *)&iser_task->desc.iscsi_header;
162 	task->hdr_max = sizeof(iser_task->desc.iscsi_header);
163 
164 	return 0;
165 }
166 
167 /**
168  * iser_initialize_task_headers() - Initialize task headers
169  * @task:       iscsi task
170  * @tx_desc:    iser tx descriptor
171  *
172  * Notes:
173  * This routine may race with iser teardown flow for scsi
174  * error handling TMFs. So for TMF we should acquire the
175  * state mutex to avoid dereferencing the IB device which
176  * may have already been terminated.
177  */
178 int
179 iser_initialize_task_headers(struct iscsi_task *task,
180 			     struct iser_tx_desc *tx_desc)
181 {
182 	struct iser_conn *iser_conn = task->conn->dd_data;
183 	struct iser_device *device = iser_conn->ib_conn.device;
184 	struct iscsi_iser_task *iser_task = task->dd_data;
185 	u64 dma_addr;
186 	const bool mgmt_task = !task->sc && !in_interrupt();
187 	int ret = 0;
188 
189 	if (unlikely(mgmt_task))
190 		mutex_lock(&iser_conn->state_mutex);
191 
192 	if (unlikely(iser_conn->state != ISER_CONN_UP)) {
193 		ret = -ENODEV;
194 		goto out;
195 	}
196 
197 	dma_addr = ib_dma_map_single(device->ib_device, (void *)tx_desc,
198 				ISER_HEADERS_LEN, DMA_TO_DEVICE);
199 	if (ib_dma_mapping_error(device->ib_device, dma_addr)) {
200 		ret = -ENOMEM;
201 		goto out;
202 	}
203 
204 	tx_desc->dma_addr = dma_addr;
205 	tx_desc->tx_sg[0].addr   = tx_desc->dma_addr;
206 	tx_desc->tx_sg[0].length = ISER_HEADERS_LEN;
207 	tx_desc->tx_sg[0].lkey   = device->mr->lkey;
208 
209 	iser_task->iser_conn = iser_conn;
210 out:
211 	if (unlikely(mgmt_task))
212 		mutex_unlock(&iser_conn->state_mutex);
213 
214 	return ret;
215 }
216 
217 /**
218  * iscsi_iser_task_init() - Initialize iscsi-iser task
219  * @task: iscsi task
220  *
221  * Initialize the task for the scsi command or mgmt command.
222  *
223  * Return: Returns zero on success or -ENOMEM when failing
224  *         to init task headers (dma mapping error).
225  */
226 static int
227 iscsi_iser_task_init(struct iscsi_task *task)
228 {
229 	struct iscsi_iser_task *iser_task = task->dd_data;
230 	int ret;
231 
232 	ret = iser_initialize_task_headers(task, &iser_task->desc);
233 	if (ret) {
234 		iser_err("Failed to init task %p, err = %d\n",
235 			 iser_task, ret);
236 		return ret;
237 	}
238 
239 	/* mgmt task */
240 	if (!task->sc)
241 		return 0;
242 
243 	iser_task->command_sent = 0;
244 	iser_task_rdma_init(iser_task);
245 	iser_task->sc = task->sc;
246 
247 	return 0;
248 }
249 
250 /**
251  * iscsi_iser_mtask_xmit() - xmit management (immediate) task
252  * @conn: iscsi connection
253  * @task: task management task
254  *
255  * Notes:
256  *	The function can return -EAGAIN in which case caller must
257  *	call it again later, or recover. '0' return code means successful
258  *	xmit.
259  *
260  **/
261 static int
262 iscsi_iser_mtask_xmit(struct iscsi_conn *conn, struct iscsi_task *task)
263 {
264 	int error = 0;
265 
266 	iser_dbg("mtask xmit [cid %d itt 0x%x]\n", conn->id, task->itt);
267 
268 	error = iser_send_control(conn, task);
269 
270 	/* since iser xmits control with zero copy, tasks can not be recycled
271 	 * right after sending them.
272 	 * The recycling scheme is based on whether a response is expected
273 	 * - if yes, the task is recycled at iscsi_complete_pdu
274 	 * - if no,  the task is recycled at iser_snd_completion
275 	 */
276 	return error;
277 }
278 
279 static int
280 iscsi_iser_task_xmit_unsol_data(struct iscsi_conn *conn,
281 				 struct iscsi_task *task)
282 {
283 	struct iscsi_r2t_info *r2t = &task->unsol_r2t;
284 	struct iscsi_data hdr;
285 	int error = 0;
286 
287 	/* Send data-out PDUs while there's still unsolicited data to send */
288 	while (iscsi_task_has_unsol_data(task)) {
289 		iscsi_prep_data_out_pdu(task, r2t, &hdr);
290 		iser_dbg("Sending data-out: itt 0x%x, data count %d\n",
291 			   hdr.itt, r2t->data_count);
292 
293 		/* the buffer description has been passed with the command */
294 		/* Send the command */
295 		error = iser_send_data_out(conn, task, &hdr);
296 		if (error) {
297 			r2t->datasn--;
298 			goto iscsi_iser_task_xmit_unsol_data_exit;
299 		}
300 		r2t->sent += r2t->data_count;
301 		iser_dbg("Need to send %d more as data-out PDUs\n",
302 			   r2t->data_length - r2t->sent);
303 	}
304 
305 iscsi_iser_task_xmit_unsol_data_exit:
306 	return error;
307 }
308 
309 /**
310  * iscsi_iser_task_xmit() - xmit iscsi-iser task
311  * @task: iscsi task
312  *
313  * Return: zero on success or escalates $error on failure.
314  */
315 static int
316 iscsi_iser_task_xmit(struct iscsi_task *task)
317 {
318 	struct iscsi_conn *conn = task->conn;
319 	struct iscsi_iser_task *iser_task = task->dd_data;
320 	int error = 0;
321 
322 	if (!task->sc)
323 		return iscsi_iser_mtask_xmit(conn, task);
324 
325 	if (task->sc->sc_data_direction == DMA_TO_DEVICE) {
326 		BUG_ON(scsi_bufflen(task->sc) == 0);
327 
328 		iser_dbg("cmd [itt %x total %d imm %d unsol_data %d\n",
329 			   task->itt, scsi_bufflen(task->sc),
330 			   task->imm_count, task->unsol_r2t.data_length);
331 	}
332 
333 	iser_dbg("ctask xmit [cid %d itt 0x%x]\n",
334 		   conn->id, task->itt);
335 
336 	/* Send the cmd PDU */
337 	if (!iser_task->command_sent) {
338 		error = iser_send_command(conn, task);
339 		if (error)
340 			goto iscsi_iser_task_xmit_exit;
341 		iser_task->command_sent = 1;
342 	}
343 
344 	/* Send unsolicited data-out PDU(s) if necessary */
345 	if (iscsi_task_has_unsol_data(task))
346 		error = iscsi_iser_task_xmit_unsol_data(conn, task);
347 
348  iscsi_iser_task_xmit_exit:
349 	return error;
350 }
351 
352 /**
353  * iscsi_iser_cleanup_task() - cleanup an iscsi-iser task
354  * @task: iscsi task
355  *
356  * Notes: In case the RDMA device is already NULL (might have
357  *        been removed in DEVICE_REMOVAL CM event it will bail-out
358  *        without doing dma unmapping.
359  */
360 static void iscsi_iser_cleanup_task(struct iscsi_task *task)
361 {
362 	struct iscsi_iser_task *iser_task = task->dd_data;
363 	struct iser_tx_desc    *tx_desc   = &iser_task->desc;
364 	struct iser_conn       *iser_conn	  = task->conn->dd_data;
365 	struct iser_device *device = iser_conn->ib_conn.device;
366 
367 	/* DEVICE_REMOVAL event might have already released the device */
368 	if (!device)
369 		return;
370 
371 	ib_dma_unmap_single(device->ib_device,
372 		tx_desc->dma_addr, ISER_HEADERS_LEN, DMA_TO_DEVICE);
373 
374 	/* mgmt tasks do not need special cleanup */
375 	if (!task->sc)
376 		return;
377 
378 	if (iser_task->status == ISER_TASK_STATUS_STARTED) {
379 		iser_task->status = ISER_TASK_STATUS_COMPLETED;
380 		iser_task_rdma_finalize(iser_task);
381 	}
382 }
383 
384 /**
385  * iscsi_iser_check_protection() - check protection information status of task.
386  * @task:     iscsi task
387  * @sector:   error sector if exsists (output)
388  *
389  * Return: zero if no data-integrity errors have occured
390  *         0x1: data-integrity error occured in the guard-block
391  *         0x2: data-integrity error occured in the reference tag
392  *         0x3: data-integrity error occured in the application tag
393  *
394  *         In addition the error sector is marked.
395  */
396 static u8
397 iscsi_iser_check_protection(struct iscsi_task *task, sector_t *sector)
398 {
399 	struct iscsi_iser_task *iser_task = task->dd_data;
400 
401 	if (iser_task->dir[ISER_DIR_IN])
402 		return iser_check_task_pi_status(iser_task, ISER_DIR_IN,
403 						 sector);
404 	else
405 		return iser_check_task_pi_status(iser_task, ISER_DIR_OUT,
406 						 sector);
407 }
408 
409 /**
410  * iscsi_iser_conn_create() - create a new iscsi-iser connection
411  * @cls_session: iscsi class connection
412  * @conn_idx:    connection index within the session (for MCS)
413  *
414  * Return: iscsi_cls_conn when iscsi_conn_setup succeeds or NULL
415  *         otherwise.
416  */
417 static struct iscsi_cls_conn *
418 iscsi_iser_conn_create(struct iscsi_cls_session *cls_session,
419 		       uint32_t conn_idx)
420 {
421 	struct iscsi_conn *conn;
422 	struct iscsi_cls_conn *cls_conn;
423 
424 	cls_conn = iscsi_conn_setup(cls_session, 0, conn_idx);
425 	if (!cls_conn)
426 		return NULL;
427 	conn = cls_conn->dd_data;
428 
429 	/*
430 	 * due to issues with the login code re iser sematics
431 	 * this not set in iscsi_conn_setup - FIXME
432 	 */
433 	conn->max_recv_dlength = ISER_RECV_DATA_SEG_LEN;
434 
435 	return cls_conn;
436 }
437 
438 /**
439  * iscsi_iser_conn_bind() - bind iscsi and iser connection structures
440  * @cls_session:     iscsi class session
441  * @cls_conn:        iscsi class connection
442  * @transport_eph:   transport end-point handle
443  * @is_leading:      indicate if this is the session leading connection (MCS)
444  *
445  * Return: zero on success, $error if iscsi_conn_bind fails and
446  *         -EINVAL in case end-point doesn't exsits anymore or iser connection
447  *         state is not UP (teardown already started).
448  */
449 static int
450 iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session,
451 		     struct iscsi_cls_conn *cls_conn,
452 		     uint64_t transport_eph,
453 		     int is_leading)
454 {
455 	struct iscsi_conn *conn = cls_conn->dd_data;
456 	struct iser_conn *iser_conn;
457 	struct iscsi_endpoint *ep;
458 	int error;
459 
460 	error = iscsi_conn_bind(cls_session, cls_conn, is_leading);
461 	if (error)
462 		return error;
463 
464 	/* the transport ep handle comes from user space so it must be
465 	 * verified against the global ib connections list */
466 	ep = iscsi_lookup_endpoint(transport_eph);
467 	if (!ep) {
468 		iser_err("can't bind eph %llx\n",
469 			 (unsigned long long)transport_eph);
470 		return -EINVAL;
471 	}
472 	iser_conn = ep->dd_data;
473 
474 	mutex_lock(&iser_conn->state_mutex);
475 	if (iser_conn->state != ISER_CONN_UP) {
476 		error = -EINVAL;
477 		iser_err("iser_conn %p state is %d, teardown started\n",
478 			 iser_conn, iser_conn->state);
479 		goto out;
480 	}
481 
482 	error = iser_alloc_rx_descriptors(iser_conn, conn->session);
483 	if (error)
484 		goto out;
485 
486 	/* binds the iSER connection retrieved from the previously
487 	 * connected ep_handle to the iSCSI layer connection. exchanges
488 	 * connection pointers */
489 	iser_info("binding iscsi conn %p to iser_conn %p\n", conn, iser_conn);
490 
491 	conn->dd_data = iser_conn;
492 	iser_conn->iscsi_conn = conn;
493 
494 out:
495 	mutex_unlock(&iser_conn->state_mutex);
496 	return error;
497 }
498 
499 /**
500  * iscsi_iser_conn_start() - start iscsi-iser connection
501  * @cls_conn: iscsi class connection
502  *
503  * Notes: Here iser intialize (or re-initialize) stop_completion as
504  *        from this point iscsi must call conn_stop in session/connection
505  *        teardown so iser transport must wait for it.
506  */
507 static int
508 iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn)
509 {
510 	struct iscsi_conn *iscsi_conn;
511 	struct iser_conn *iser_conn;
512 
513 	iscsi_conn = cls_conn->dd_data;
514 	iser_conn = iscsi_conn->dd_data;
515 	reinit_completion(&iser_conn->stop_completion);
516 
517 	return iscsi_conn_start(cls_conn);
518 }
519 
520 /**
521  * iscsi_iser_conn_stop() - stop iscsi-iser connection
522  * @cls_conn:  iscsi class connection
523  * @flag:      indicate if recover or terminate (passed as is)
524  *
525  * Notes: Calling iscsi_conn_stop might theoretically race with
526  *        DEVICE_REMOVAL event and dereference a previously freed RDMA device
527  *        handle, so we call it under iser the state lock to protect against
528  *        this kind of race.
529  */
530 static void
531 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 		iser_conn_terminate(iser_conn);
545 		iscsi_conn_stop(cls_conn, flag);
546 
547 		/* unbind */
548 		iser_conn->iscsi_conn = NULL;
549 		conn->dd_data = NULL;
550 
551 		complete(&iser_conn->stop_completion);
552 		mutex_unlock(&iser_conn->state_mutex);
553 	} else {
554 		iscsi_conn_stop(cls_conn, flag);
555 	}
556 }
557 
558 /**
559  * iscsi_iser_session_destroy() - destroy iscsi-iser session
560  * @cls_session: iscsi class session
561  *
562  * Removes and free iscsi host.
563  */
564 static void
565 iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session)
566 {
567 	struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
568 
569 	iscsi_session_teardown(cls_session);
570 	iscsi_host_remove(shost);
571 	iscsi_host_free(shost);
572 }
573 
574 static inline unsigned int
575 iser_dif_prot_caps(int prot_caps)
576 {
577 	return ((prot_caps & IB_PROT_T10DIF_TYPE_1) ?
578 		SHOST_DIF_TYPE1_PROTECTION | SHOST_DIX_TYPE0_PROTECTION |
579 		SHOST_DIX_TYPE1_PROTECTION : 0) |
580 	       ((prot_caps & IB_PROT_T10DIF_TYPE_2) ?
581 		SHOST_DIF_TYPE2_PROTECTION | SHOST_DIX_TYPE2_PROTECTION : 0) |
582 	       ((prot_caps & IB_PROT_T10DIF_TYPE_3) ?
583 		SHOST_DIF_TYPE3_PROTECTION | SHOST_DIX_TYPE3_PROTECTION : 0);
584 }
585 
586 /**
587  * iscsi_iser_session_create() - create an iscsi-iser session
588  * @ep:             iscsi end-point handle
589  * @cmds_max:       maximum commands in this session
590  * @qdepth:         session command queue depth
591  * @initial_cmdsn:  initiator command sequnce number
592  *
593  * Allocates and adds a scsi host, expose DIF supprot if
594  * exists, and sets up an iscsi session.
595  */
596 static struct iscsi_cls_session *
597 iscsi_iser_session_create(struct iscsi_endpoint *ep,
598 			  uint16_t cmds_max, uint16_t qdepth,
599 			  uint32_t initial_cmdsn)
600 {
601 	struct iscsi_cls_session *cls_session;
602 	struct iscsi_session *session;
603 	struct Scsi_Host *shost;
604 	struct iser_conn *iser_conn = NULL;
605 	struct ib_conn *ib_conn;
606 	u16 max_cmds;
607 
608 	shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 0);
609 	if (!shost)
610 		return NULL;
611 	shost->transportt = iscsi_iser_scsi_transport;
612 	shost->cmd_per_lun = qdepth;
613 	shost->max_lun = iscsi_max_lun;
614 	shost->max_id = 0;
615 	shost->max_channel = 0;
616 	shost->max_cmd_len = 16;
617 
618 	/*
619 	 * older userspace tools (before 2.0-870) did not pass us
620 	 * the leading conn's ep so this will be NULL;
621 	 */
622 	if (ep) {
623 		iser_conn = ep->dd_data;
624 		max_cmds = iser_conn->max_cmds;
625 
626 		mutex_lock(&iser_conn->state_mutex);
627 		if (iser_conn->state != ISER_CONN_UP) {
628 			iser_err("iser conn %p already started teardown\n",
629 				 iser_conn);
630 			mutex_unlock(&iser_conn->state_mutex);
631 			goto free_host;
632 		}
633 
634 		ib_conn = &iser_conn->ib_conn;
635 		if (ib_conn->pi_support) {
636 			u32 sig_caps = ib_conn->device->dev_attr.sig_prot_cap;
637 
638 			scsi_host_set_prot(shost, iser_dif_prot_caps(sig_caps));
639 			scsi_host_set_guard(shost, SHOST_DIX_GUARD_IP |
640 						   SHOST_DIX_GUARD_CRC);
641 		}
642 
643 		if (iscsi_host_add(shost,
644 				   ib_conn->device->ib_device->dma_device)) {
645 			mutex_unlock(&iser_conn->state_mutex);
646 			goto free_host;
647 		}
648 		mutex_unlock(&iser_conn->state_mutex);
649 	} else {
650 		max_cmds = ISER_DEF_XMIT_CMDS_MAX;
651 		if (iscsi_host_add(shost, NULL))
652 			goto free_host;
653 	}
654 
655 	if (cmds_max > max_cmds) {
656 		iser_info("cmds_max changed from %u to %u\n",
657 			  cmds_max, max_cmds);
658 		cmds_max = max_cmds;
659 	}
660 
661 	cls_session = iscsi_session_setup(&iscsi_iser_transport, shost,
662 					  cmds_max, 0,
663 					  sizeof(struct iscsi_iser_task),
664 					  initial_cmdsn, 0);
665 	if (!cls_session)
666 		goto remove_host;
667 	session = cls_session->dd_data;
668 
669 	shost->can_queue = session->scsi_cmds_max;
670 	return cls_session;
671 
672 remove_host:
673 	iscsi_host_remove(shost);
674 free_host:
675 	iscsi_host_free(shost);
676 	return NULL;
677 }
678 
679 static int
680 iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn,
681 		     enum iscsi_param param, char *buf, int buflen)
682 {
683 	int value;
684 
685 	switch (param) {
686 	case ISCSI_PARAM_MAX_RECV_DLENGTH:
687 		/* TBD */
688 		break;
689 	case ISCSI_PARAM_HDRDGST_EN:
690 		sscanf(buf, "%d", &value);
691 		if (value) {
692 			iser_err("DataDigest wasn't negotiated to None\n");
693 			return -EPROTO;
694 		}
695 		break;
696 	case ISCSI_PARAM_DATADGST_EN:
697 		sscanf(buf, "%d", &value);
698 		if (value) {
699 			iser_err("DataDigest wasn't negotiated to None\n");
700 			return -EPROTO;
701 		}
702 		break;
703 	case ISCSI_PARAM_IFMARKER_EN:
704 		sscanf(buf, "%d", &value);
705 		if (value) {
706 			iser_err("IFMarker wasn't negotiated to No\n");
707 			return -EPROTO;
708 		}
709 		break;
710 	case ISCSI_PARAM_OFMARKER_EN:
711 		sscanf(buf, "%d", &value);
712 		if (value) {
713 			iser_err("OFMarker wasn't negotiated to No\n");
714 			return -EPROTO;
715 		}
716 		break;
717 	default:
718 		return iscsi_set_param(cls_conn, param, buf, buflen);
719 	}
720 
721 	return 0;
722 }
723 
724 /**
725  * iscsi_iser_set_param() - set class connection parameter
726  * @cls_conn:    iscsi class connection
727  * @stats:       iscsi stats to output
728  *
729  * Output connection statistics.
730  */
731 static void
732 iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
733 {
734 	struct iscsi_conn *conn = cls_conn->dd_data;
735 
736 	stats->txdata_octets = conn->txdata_octets;
737 	stats->rxdata_octets = conn->rxdata_octets;
738 	stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
739 	stats->dataout_pdus = conn->dataout_pdus_cnt;
740 	stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
741 	stats->datain_pdus = conn->datain_pdus_cnt; /* always 0 */
742 	stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */
743 	stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
744 	stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
745 	stats->custom_length = 4;
746 	strcpy(stats->custom[0].desc, "qp_tx_queue_full");
747 	stats->custom[0].value = 0; /* TB iser_conn->qp_tx_queue_full; */
748 	strcpy(stats->custom[1].desc, "fmr_map_not_avail");
749 	stats->custom[1].value = 0; /* TB iser_conn->fmr_map_not_avail */;
750 	strcpy(stats->custom[2].desc, "eh_abort_cnt");
751 	stats->custom[2].value = conn->eh_abort_cnt;
752 	strcpy(stats->custom[3].desc, "fmr_unalign_cnt");
753 	stats->custom[3].value = conn->fmr_unalign_cnt;
754 }
755 
756 static int iscsi_iser_get_ep_param(struct iscsi_endpoint *ep,
757 				   enum iscsi_param param, char *buf)
758 {
759 	struct iser_conn *iser_conn = ep->dd_data;
760 	int len;
761 
762 	switch (param) {
763 	case ISCSI_PARAM_CONN_PORT:
764 	case ISCSI_PARAM_CONN_ADDRESS:
765 		if (!iser_conn || !iser_conn->ib_conn.cma_id)
766 			return -ENOTCONN;
767 
768 		return iscsi_conn_get_addr_param((struct sockaddr_storage *)
769 				&iser_conn->ib_conn.cma_id->route.addr.dst_addr,
770 				param, buf);
771 		break;
772 	default:
773 		return -ENOSYS;
774 	}
775 
776 	return len;
777 }
778 
779 /**
780  * iscsi_iser_ep_connect() - Initiate iSER connection establishment
781  * @shost:          scsi_host
782  * @dst_addr:       destination address
783  * @non-blocking:   indicate if routine can block
784  *
785  * Allocate an iscsi endpoint, an iser_conn structure and bind them.
786  * After that start RDMA connection establishment via rdma_cm. We
787  * don't allocate iser_conn embedded in iscsi_endpoint since in teardown
788  * the endpoint will be destroyed at ep_disconnect while iser_conn will
789  * cleanup its resources asynchronuously.
790  *
791  * Return: iscsi_endpoint created by iscsi layer or ERR_PTR(error)
792  *         if fails.
793  */
794 static struct iscsi_endpoint *
795 iscsi_iser_ep_connect(struct Scsi_Host *shost, struct sockaddr *dst_addr,
796 		      int non_blocking)
797 {
798 	int err;
799 	struct iser_conn *iser_conn;
800 	struct iscsi_endpoint *ep;
801 
802 	ep = iscsi_create_endpoint(0);
803 	if (!ep)
804 		return ERR_PTR(-ENOMEM);
805 
806 	iser_conn = kzalloc(sizeof(*iser_conn), GFP_KERNEL);
807 	if (!iser_conn) {
808 		err = -ENOMEM;
809 		goto failure;
810 	}
811 
812 	ep->dd_data = iser_conn;
813 	iser_conn->ep = ep;
814 	iser_conn_init(iser_conn);
815 
816 	err = iser_connect(iser_conn, NULL, dst_addr, non_blocking);
817 	if (err)
818 		goto failure;
819 
820 	return ep;
821 failure:
822 	iscsi_destroy_endpoint(ep);
823 	return ERR_PTR(err);
824 }
825 
826 /**
827  * iscsi_iser_ep_poll() - poll for iser connection establishment to complete
828  * @ep:            iscsi endpoint (created at ep_connect)
829  * @timeout_ms:    polling timeout allowed in ms.
830  *
831  * This routine boils down to waiting for up_completion signaling
832  * that cma_id got CONNECTED event.
833  *
834  * Return: 1 if succeeded in connection establishment, 0 if timeout expired
835  *         (libiscsi will retry will kick in) or -1 if interrupted by signal
836  *         or more likely iser connection state transitioned to TEMINATING or
837  *         DOWN during the wait period.
838  */
839 static int
840 iscsi_iser_ep_poll(struct iscsi_endpoint *ep, int timeout_ms)
841 {
842 	struct iser_conn *iser_conn;
843 	int rc;
844 
845 	iser_conn = ep->dd_data;
846 	rc = wait_for_completion_interruptible_timeout(&iser_conn->up_completion,
847 						       msecs_to_jiffies(timeout_ms));
848 	/* if conn establishment failed, return error code to iscsi */
849 	if (rc == 0) {
850 		mutex_lock(&iser_conn->state_mutex);
851 		if (iser_conn->state == ISER_CONN_TERMINATING ||
852 		    iser_conn->state == ISER_CONN_DOWN)
853 			rc = -1;
854 		mutex_unlock(&iser_conn->state_mutex);
855 	}
856 
857 	iser_info("ib conn %p rc = %d\n", iser_conn, rc);
858 
859 	if (rc > 0)
860 		return 1; /* success, this is the equivalent of POLLOUT */
861 	else if (!rc)
862 		return 0; /* timeout */
863 	else
864 		return rc; /* signal */
865 }
866 
867 /**
868  * iscsi_iser_ep_disconnect() - Initiate connection teardown process
869  * @ep:    iscsi endpoint handle
870  *
871  * This routine is not blocked by iser and RDMA termination process
872  * completion as we queue a deffered work for iser/RDMA destruction
873  * and cleanup or actually call it immediately in case we didn't pass
874  * iscsi conn bind/start stage, thus it is safe.
875  */
876 static void
877 iscsi_iser_ep_disconnect(struct iscsi_endpoint *ep)
878 {
879 	struct iser_conn *iser_conn;
880 
881 	iser_conn = ep->dd_data;
882 	iser_info("ep %p iser conn %p state %d\n",
883 		  ep, iser_conn, iser_conn->state);
884 
885 	mutex_lock(&iser_conn->state_mutex);
886 	iser_conn_terminate(iser_conn);
887 
888 	/*
889 	 * if iser_conn and iscsi_conn are bound, we must wait for
890 	 * iscsi_conn_stop and flush errors completion before freeing
891 	 * the iser resources. Otherwise we are safe to free resources
892 	 * immediately.
893 	 */
894 	if (iser_conn->iscsi_conn) {
895 		INIT_WORK(&iser_conn->release_work, iser_release_work);
896 		queue_work(release_wq, &iser_conn->release_work);
897 		mutex_unlock(&iser_conn->state_mutex);
898 	} else {
899 		iser_conn->state = ISER_CONN_DOWN;
900 		mutex_unlock(&iser_conn->state_mutex);
901 		iser_conn_release(iser_conn);
902 	}
903 	iscsi_destroy_endpoint(ep);
904 }
905 
906 static umode_t iser_attr_is_visible(int param_type, int param)
907 {
908 	switch (param_type) {
909 	case ISCSI_HOST_PARAM:
910 		switch (param) {
911 		case ISCSI_HOST_PARAM_NETDEV_NAME:
912 		case ISCSI_HOST_PARAM_HWADDRESS:
913 		case ISCSI_HOST_PARAM_INITIATOR_NAME:
914 			return S_IRUGO;
915 		default:
916 			return 0;
917 		}
918 	case ISCSI_PARAM:
919 		switch (param) {
920 		case ISCSI_PARAM_MAX_RECV_DLENGTH:
921 		case ISCSI_PARAM_MAX_XMIT_DLENGTH:
922 		case ISCSI_PARAM_HDRDGST_EN:
923 		case ISCSI_PARAM_DATADGST_EN:
924 		case ISCSI_PARAM_CONN_ADDRESS:
925 		case ISCSI_PARAM_CONN_PORT:
926 		case ISCSI_PARAM_EXP_STATSN:
927 		case ISCSI_PARAM_PERSISTENT_ADDRESS:
928 		case ISCSI_PARAM_PERSISTENT_PORT:
929 		case ISCSI_PARAM_PING_TMO:
930 		case ISCSI_PARAM_RECV_TMO:
931 		case ISCSI_PARAM_INITIAL_R2T_EN:
932 		case ISCSI_PARAM_MAX_R2T:
933 		case ISCSI_PARAM_IMM_DATA_EN:
934 		case ISCSI_PARAM_FIRST_BURST:
935 		case ISCSI_PARAM_MAX_BURST:
936 		case ISCSI_PARAM_PDU_INORDER_EN:
937 		case ISCSI_PARAM_DATASEQ_INORDER_EN:
938 		case ISCSI_PARAM_TARGET_NAME:
939 		case ISCSI_PARAM_TPGT:
940 		case ISCSI_PARAM_USERNAME:
941 		case ISCSI_PARAM_PASSWORD:
942 		case ISCSI_PARAM_USERNAME_IN:
943 		case ISCSI_PARAM_PASSWORD_IN:
944 		case ISCSI_PARAM_FAST_ABORT:
945 		case ISCSI_PARAM_ABORT_TMO:
946 		case ISCSI_PARAM_LU_RESET_TMO:
947 		case ISCSI_PARAM_TGT_RESET_TMO:
948 		case ISCSI_PARAM_IFACE_NAME:
949 		case ISCSI_PARAM_INITIATOR_NAME:
950 		case ISCSI_PARAM_DISCOVERY_SESS:
951 			return S_IRUGO;
952 		default:
953 			return 0;
954 		}
955 	}
956 
957 	return 0;
958 }
959 
960 static struct scsi_host_template iscsi_iser_sht = {
961 	.module                 = THIS_MODULE,
962 	.name                   = "iSCSI Initiator over iSER",
963 	.queuecommand           = iscsi_queuecommand,
964 	.change_queue_depth	= scsi_change_queue_depth,
965 	.sg_tablesize           = ISCSI_ISER_SG_TABLESIZE,
966 	.max_sectors		= 1024,
967 	.cmd_per_lun            = ISER_DEF_CMD_PER_LUN,
968 	.eh_abort_handler       = iscsi_eh_abort,
969 	.eh_device_reset_handler= iscsi_eh_device_reset,
970 	.eh_target_reset_handler = iscsi_eh_recover_target,
971 	.target_alloc		= iscsi_target_alloc,
972 	.use_clustering         = DISABLE_CLUSTERING,
973 	.proc_name              = "iscsi_iser",
974 	.this_id                = -1,
975 	.track_queue_depth	= 1,
976 };
977 
978 static struct iscsi_transport iscsi_iser_transport = {
979 	.owner                  = THIS_MODULE,
980 	.name                   = "iser",
981 	.caps                   = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_TEXT_NEGO,
982 	/* session management */
983 	.create_session         = iscsi_iser_session_create,
984 	.destroy_session        = iscsi_iser_session_destroy,
985 	/* connection management */
986 	.create_conn            = iscsi_iser_conn_create,
987 	.bind_conn              = iscsi_iser_conn_bind,
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 	if (iscsi_max_lun < 1) {
1022 		iser_err("Invalid max_lun value of %u\n", iscsi_max_lun);
1023 		return -EINVAL;
1024 	}
1025 
1026 	memset(&ig, 0, sizeof(struct iser_global));
1027 
1028 	ig.desc_cache = kmem_cache_create("iser_descriptors",
1029 					  sizeof(struct iser_tx_desc),
1030 					  0, SLAB_HWCACHE_ALIGN,
1031 					  NULL);
1032 	if (ig.desc_cache == NULL)
1033 		return -ENOMEM;
1034 
1035 	/* device init is called only after the first addr resolution */
1036 	mutex_init(&ig.device_list_mutex);
1037 	INIT_LIST_HEAD(&ig.device_list);
1038 	mutex_init(&ig.connlist_mutex);
1039 	INIT_LIST_HEAD(&ig.connlist);
1040 
1041 	release_wq = alloc_workqueue("release workqueue", 0, 0);
1042 	if (!release_wq) {
1043 		iser_err("failed to allocate release workqueue\n");
1044 		return -ENOMEM;
1045 	}
1046 
1047 	iscsi_iser_scsi_transport = iscsi_register_transport(
1048 							&iscsi_iser_transport);
1049 	if (!iscsi_iser_scsi_transport) {
1050 		iser_err("iscsi_register_transport failed\n");
1051 		err = -EINVAL;
1052 		goto register_transport_failure;
1053 	}
1054 
1055 	return 0;
1056 
1057 register_transport_failure:
1058 	kmem_cache_destroy(ig.desc_cache);
1059 
1060 	return err;
1061 }
1062 
1063 static void __exit iser_exit(void)
1064 {
1065 	struct iser_conn *iser_conn, *n;
1066 	int connlist_empty;
1067 
1068 	iser_dbg("Removing iSER datamover...\n");
1069 	destroy_workqueue(release_wq);
1070 
1071 	mutex_lock(&ig.connlist_mutex);
1072 	connlist_empty = list_empty(&ig.connlist);
1073 	mutex_unlock(&ig.connlist_mutex);
1074 
1075 	if (!connlist_empty) {
1076 		iser_err("Error cleanup stage completed but we still have iser "
1077 			 "connections, destroying them anyway.\n");
1078 		list_for_each_entry_safe(iser_conn, n, &ig.connlist,
1079 					 conn_list) {
1080 			iser_conn_release(iser_conn);
1081 		}
1082 	}
1083 
1084 	iscsi_unregister_transport(&iscsi_iser_transport);
1085 	kmem_cache_destroy(ig.desc_cache);
1086 }
1087 
1088 module_init(iser_init);
1089 module_exit(iser_exit);
1090