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 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 int iser_pi_guard;
117 module_param_named(pi_guard, iser_pi_guard, int, S_IRUGO);
118 MODULE_PARM_DESC(pi_guard, "T10-PI guard_type [deprecated]");
119 
120 static int iscsi_iser_set(const char *val, const struct kernel_param *kp)
121 {
122 	int ret;
123 	unsigned int n = 0;
124 
125 	ret = kstrtouint(val, 10, &n);
126 	if (ret != 0 || n == 0)
127 		return -EINVAL;
128 
129 	return param_set_uint(val, kp);
130 }
131 
132 /*
133  * iscsi_iser_recv() - Process a successful recv completion
134  * @conn:         iscsi connection
135  * @hdr:          iscsi header
136  * @rx_data:      buffer containing receive data payload
137  * @rx_data_len:  length of rx_data
138  *
139  * Notes: In case of data length errors or iscsi PDU completion failures
140  *        this routine will signal iscsi layer of connection failure.
141  */
142 void
143 iscsi_iser_recv(struct iscsi_conn *conn, struct iscsi_hdr *hdr,
144 		char *rx_data, int rx_data_len)
145 {
146 	int rc = 0;
147 	int datalen;
148 
149 	/* verify PDU length */
150 	datalen = ntoh24(hdr->dlength);
151 	if (datalen > rx_data_len || (datalen + 4) < rx_data_len) {
152 		iser_err("wrong datalen %d (hdr), %d (IB)\n",
153 			datalen, rx_data_len);
154 		rc = ISCSI_ERR_DATALEN;
155 		goto error;
156 	}
157 
158 	if (datalen != rx_data_len)
159 		iser_dbg("aligned datalen (%d) hdr, %d (IB)\n",
160 			datalen, rx_data_len);
161 
162 	rc = iscsi_complete_pdu(conn, hdr, rx_data, rx_data_len);
163 	if (rc && rc != ISCSI_ERR_NO_SCSI_CMD)
164 		goto error;
165 
166 	return;
167 error:
168 	iscsi_conn_failure(conn, rc);
169 }
170 
171 /**
172  * iscsi_iser_pdu_alloc() - allocate an iscsi-iser PDU
173  * @task:     iscsi task
174  * @opcode:   iscsi command opcode
175  *
176  * Netes: This routine can't fail, just assign iscsi task
177  *        hdr and max hdr size.
178  */
179 static int
180 iscsi_iser_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
181 {
182 	struct iscsi_iser_task *iser_task = task->dd_data;
183 
184 	task->hdr = (struct iscsi_hdr *)&iser_task->desc.iscsi_header;
185 	task->hdr_max = sizeof(iser_task->desc.iscsi_header);
186 
187 	return 0;
188 }
189 
190 /**
191  * iser_initialize_task_headers() - Initialize task headers
192  * @task:       iscsi task
193  * @tx_desc:    iser tx descriptor
194  *
195  * Notes:
196  * This routine may race with iser teardown flow for scsi
197  * error handling TMFs. So for TMF we should acquire the
198  * state mutex to avoid dereferencing the IB device which
199  * may have already been terminated.
200  */
201 int
202 iser_initialize_task_headers(struct iscsi_task *task,
203 			     struct iser_tx_desc *tx_desc)
204 {
205 	struct iser_conn *iser_conn = task->conn->dd_data;
206 	struct iser_device *device = iser_conn->ib_conn.device;
207 	struct iscsi_iser_task *iser_task = task->dd_data;
208 	u64 dma_addr;
209 
210 	if (unlikely(iser_conn->state != ISER_CONN_UP))
211 		return -ENODEV;
212 
213 	dma_addr = ib_dma_map_single(device->ib_device, (void *)tx_desc,
214 				ISER_HEADERS_LEN, DMA_TO_DEVICE);
215 	if (ib_dma_mapping_error(device->ib_device, dma_addr))
216 		return -ENOMEM;
217 
218 	tx_desc->inv_wr.next = NULL;
219 	tx_desc->reg_wr.wr.next = NULL;
220 	tx_desc->mapped = true;
221 	tx_desc->dma_addr = dma_addr;
222 	tx_desc->tx_sg[0].addr   = tx_desc->dma_addr;
223 	tx_desc->tx_sg[0].length = ISER_HEADERS_LEN;
224 	tx_desc->tx_sg[0].lkey   = device->pd->local_dma_lkey;
225 
226 	iser_task->iser_conn = iser_conn;
227 
228 	return 0;
229 }
230 
231 /**
232  * iscsi_iser_task_init() - Initialize iscsi-iser task
233  * @task: iscsi task
234  *
235  * Initialize the task for the scsi command or mgmt command.
236  *
237  * Return: Returns zero on success or -ENOMEM when failing
238  *         to init task headers (dma mapping error).
239  */
240 static int
241 iscsi_iser_task_init(struct iscsi_task *task)
242 {
243 	struct iscsi_iser_task *iser_task = task->dd_data;
244 	int ret;
245 
246 	ret = iser_initialize_task_headers(task, &iser_task->desc);
247 	if (ret) {
248 		iser_err("Failed to init task %p, err = %d\n",
249 			 iser_task, ret);
250 		return ret;
251 	}
252 
253 	/* mgmt task */
254 	if (!task->sc)
255 		return 0;
256 
257 	iser_task->command_sent = 0;
258 	iser_task_rdma_init(iser_task);
259 	iser_task->sc = task->sc;
260 
261 	return 0;
262 }
263 
264 /**
265  * iscsi_iser_mtask_xmit() - xmit management (immediate) task
266  * @conn: iscsi connection
267  * @task: task management task
268  *
269  * Notes:
270  *	The function can return -EAGAIN in which case caller must
271  *	call it again later, or recover. '0' return code means successful
272  *	xmit.
273  *
274  **/
275 static int
276 iscsi_iser_mtask_xmit(struct iscsi_conn *conn, struct iscsi_task *task)
277 {
278 	int error = 0;
279 
280 	iser_dbg("mtask xmit [cid %d itt 0x%x]\n", conn->id, task->itt);
281 
282 	error = iser_send_control(conn, task);
283 
284 	/* since iser xmits control with zero copy, tasks can not be recycled
285 	 * right after sending them.
286 	 * The recycling scheme is based on whether a response is expected
287 	 * - if yes, the task is recycled at iscsi_complete_pdu
288 	 * - if no,  the task is recycled at iser_snd_completion
289 	 */
290 	return error;
291 }
292 
293 static int
294 iscsi_iser_task_xmit_unsol_data(struct iscsi_conn *conn,
295 				 struct iscsi_task *task)
296 {
297 	struct iscsi_r2t_info *r2t = &task->unsol_r2t;
298 	struct iscsi_data hdr;
299 	int error = 0;
300 
301 	/* Send data-out PDUs while there's still unsolicited data to send */
302 	while (iscsi_task_has_unsol_data(task)) {
303 		iscsi_prep_data_out_pdu(task, r2t, &hdr);
304 		iser_dbg("Sending data-out: itt 0x%x, data count %d\n",
305 			   hdr.itt, r2t->data_count);
306 
307 		/* the buffer description has been passed with the command */
308 		/* Send the command */
309 		error = iser_send_data_out(conn, task, &hdr);
310 		if (error) {
311 			r2t->datasn--;
312 			goto iscsi_iser_task_xmit_unsol_data_exit;
313 		}
314 		r2t->sent += r2t->data_count;
315 		iser_dbg("Need to send %d more as data-out PDUs\n",
316 			   r2t->data_length - r2t->sent);
317 	}
318 
319 iscsi_iser_task_xmit_unsol_data_exit:
320 	return error;
321 }
322 
323 /**
324  * iscsi_iser_task_xmit() - xmit iscsi-iser task
325  * @task: iscsi task
326  *
327  * Return: zero on success or escalates $error on failure.
328  */
329 static int
330 iscsi_iser_task_xmit(struct iscsi_task *task)
331 {
332 	struct iscsi_conn *conn = task->conn;
333 	struct iscsi_iser_task *iser_task = task->dd_data;
334 	int error = 0;
335 
336 	if (!task->sc)
337 		return iscsi_iser_mtask_xmit(conn, task);
338 
339 	if (task->sc->sc_data_direction == DMA_TO_DEVICE) {
340 		BUG_ON(scsi_bufflen(task->sc) == 0);
341 
342 		iser_dbg("cmd [itt %x total %d imm %d unsol_data %d\n",
343 			   task->itt, scsi_bufflen(task->sc),
344 			   task->imm_count, task->unsol_r2t.data_length);
345 	}
346 
347 	iser_dbg("ctask xmit [cid %d itt 0x%x]\n",
348 		   conn->id, task->itt);
349 
350 	/* Send the cmd PDU */
351 	if (!iser_task->command_sent) {
352 		error = iser_send_command(conn, task);
353 		if (error)
354 			goto iscsi_iser_task_xmit_exit;
355 		iser_task->command_sent = 1;
356 	}
357 
358 	/* Send unsolicited data-out PDU(s) if necessary */
359 	if (iscsi_task_has_unsol_data(task))
360 		error = iscsi_iser_task_xmit_unsol_data(conn, task);
361 
362  iscsi_iser_task_xmit_exit:
363 	return error;
364 }
365 
366 /**
367  * iscsi_iser_cleanup_task() - cleanup an iscsi-iser task
368  * @task: iscsi task
369  *
370  * Notes: In case the RDMA device is already NULL (might have
371  *        been removed in DEVICE_REMOVAL CM event it will bail-out
372  *        without doing dma unmapping.
373  */
374 static void iscsi_iser_cleanup_task(struct iscsi_task *task)
375 {
376 	struct iscsi_iser_task *iser_task = task->dd_data;
377 	struct iser_tx_desc *tx_desc = &iser_task->desc;
378 	struct iser_conn *iser_conn = task->conn->dd_data;
379 	struct iser_device *device = iser_conn->ib_conn.device;
380 
381 	/* DEVICE_REMOVAL event might have already released the device */
382 	if (!device)
383 		return;
384 
385 	if (likely(tx_desc->mapped)) {
386 		ib_dma_unmap_single(device->ib_device, tx_desc->dma_addr,
387 				    ISER_HEADERS_LEN, DMA_TO_DEVICE);
388 		tx_desc->mapped = false;
389 	}
390 
391 	/* mgmt tasks do not need special cleanup */
392 	if (!task->sc)
393 		return;
394 
395 	if (iser_task->status == ISER_TASK_STATUS_STARTED) {
396 		iser_task->status = ISER_TASK_STATUS_COMPLETED;
397 		iser_task_rdma_finalize(iser_task);
398 	}
399 }
400 
401 /**
402  * iscsi_iser_check_protection() - check protection information status of task.
403  * @task:     iscsi task
404  * @sector:   error sector if exsists (output)
405  *
406  * Return: zero if no data-integrity errors have occured
407  *         0x1: data-integrity error occured in the guard-block
408  *         0x2: data-integrity error occured in the reference tag
409  *         0x3: data-integrity error occured in the application tag
410  *
411  *         In addition the error sector is marked.
412  */
413 static u8
414 iscsi_iser_check_protection(struct iscsi_task *task, sector_t *sector)
415 {
416 	struct iscsi_iser_task *iser_task = task->dd_data;
417 	enum iser_data_dir dir = iser_task->dir[ISER_DIR_IN] ?
418 					ISER_DIR_IN : ISER_DIR_OUT;
419 
420 	return iser_check_task_pi_status(iser_task, dir, sector);
421 }
422 
423 /**
424  * iscsi_iser_conn_create() - create a new iscsi-iser connection
425  * @cls_session: iscsi class connection
426  * @conn_idx:    connection index within the session (for MCS)
427  *
428  * Return: iscsi_cls_conn when iscsi_conn_setup succeeds or NULL
429  *         otherwise.
430  */
431 static struct iscsi_cls_conn *
432 iscsi_iser_conn_create(struct iscsi_cls_session *cls_session,
433 		       uint32_t conn_idx)
434 {
435 	struct iscsi_conn *conn;
436 	struct iscsi_cls_conn *cls_conn;
437 
438 	cls_conn = iscsi_conn_setup(cls_session, 0, conn_idx);
439 	if (!cls_conn)
440 		return NULL;
441 	conn = cls_conn->dd_data;
442 
443 	/*
444 	 * due to issues with the login code re iser sematics
445 	 * this not set in iscsi_conn_setup - FIXME
446 	 */
447 	conn->max_recv_dlength = ISER_RECV_DATA_SEG_LEN;
448 
449 	return cls_conn;
450 }
451 
452 /**
453  * iscsi_iser_conn_bind() - bind iscsi and iser connection structures
454  * @cls_session:     iscsi class session
455  * @cls_conn:        iscsi class connection
456  * @transport_eph:   transport end-point handle
457  * @is_leading:      indicate if this is the session leading connection (MCS)
458  *
459  * Return: zero on success, $error if iscsi_conn_bind fails and
460  *         -EINVAL in case end-point doesn't exsits anymore or iser connection
461  *         state is not UP (teardown already started).
462  */
463 static int
464 iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session,
465 		     struct iscsi_cls_conn *cls_conn,
466 		     uint64_t transport_eph,
467 		     int is_leading)
468 {
469 	struct iscsi_conn *conn = cls_conn->dd_data;
470 	struct iser_conn *iser_conn;
471 	struct iscsi_endpoint *ep;
472 	int error;
473 
474 	error = iscsi_conn_bind(cls_session, cls_conn, is_leading);
475 	if (error)
476 		return error;
477 
478 	/* the transport ep handle comes from user space so it must be
479 	 * verified against the global ib connections list */
480 	ep = iscsi_lookup_endpoint(transport_eph);
481 	if (!ep) {
482 		iser_err("can't bind eph %llx\n",
483 			 (unsigned long long)transport_eph);
484 		return -EINVAL;
485 	}
486 	iser_conn = ep->dd_data;
487 
488 	mutex_lock(&iser_conn->state_mutex);
489 	if (iser_conn->state != ISER_CONN_UP) {
490 		error = -EINVAL;
491 		iser_err("iser_conn %p state is %d, teardown started\n",
492 			 iser_conn, iser_conn->state);
493 		goto out;
494 	}
495 
496 	error = iser_alloc_rx_descriptors(iser_conn, conn->session);
497 	if (error)
498 		goto out;
499 
500 	/* binds the iSER connection retrieved from the previously
501 	 * connected ep_handle to the iSCSI layer connection. exchanges
502 	 * connection pointers */
503 	iser_info("binding iscsi conn %p to iser_conn %p\n", conn, iser_conn);
504 
505 	conn->dd_data = iser_conn;
506 	iser_conn->iscsi_conn = conn;
507 
508 out:
509 	mutex_unlock(&iser_conn->state_mutex);
510 	return error;
511 }
512 
513 /**
514  * iscsi_iser_conn_start() - start iscsi-iser connection
515  * @cls_conn: iscsi class connection
516  *
517  * Notes: Here iser intialize (or re-initialize) stop_completion as
518  *        from this point iscsi must call conn_stop in session/connection
519  *        teardown so iser transport must wait for it.
520  */
521 static int
522 iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn)
523 {
524 	struct iscsi_conn *iscsi_conn;
525 	struct iser_conn *iser_conn;
526 
527 	iscsi_conn = cls_conn->dd_data;
528 	iser_conn = iscsi_conn->dd_data;
529 	reinit_completion(&iser_conn->stop_completion);
530 
531 	return iscsi_conn_start(cls_conn);
532 }
533 
534 /**
535  * iscsi_iser_conn_stop() - stop iscsi-iser connection
536  * @cls_conn:  iscsi class connection
537  * @flag:      indicate if recover or terminate (passed as is)
538  *
539  * Notes: Calling iscsi_conn_stop might theoretically race with
540  *        DEVICE_REMOVAL event and dereference a previously freed RDMA device
541  *        handle, so we call it under iser the state lock to protect against
542  *        this kind of race.
543  */
544 static void
545 iscsi_iser_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
546 {
547 	struct iscsi_conn *conn = cls_conn->dd_data;
548 	struct iser_conn *iser_conn = conn->dd_data;
549 
550 	iser_info("stopping iscsi_conn: %p, iser_conn: %p\n", conn, iser_conn);
551 
552 	/*
553 	 * Userspace may have goofed up and not bound the connection or
554 	 * might have only partially setup the connection.
555 	 */
556 	if (iser_conn) {
557 		mutex_lock(&iser_conn->state_mutex);
558 		mutex_lock(&unbind_iser_conn_mutex);
559 		iser_conn_terminate(iser_conn);
560 		iscsi_conn_stop(cls_conn, flag);
561 
562 		/* unbind */
563 		iser_conn->iscsi_conn = NULL;
564 		conn->dd_data = NULL;
565 		mutex_unlock(&unbind_iser_conn_mutex);
566 
567 		complete(&iser_conn->stop_completion);
568 		mutex_unlock(&iser_conn->state_mutex);
569 	} else {
570 		iscsi_conn_stop(cls_conn, flag);
571 	}
572 }
573 
574 /**
575  * iscsi_iser_session_destroy() - destroy iscsi-iser session
576  * @cls_session: iscsi class session
577  *
578  * Removes and free iscsi host.
579  */
580 static void
581 iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session)
582 {
583 	struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
584 
585 	iscsi_session_teardown(cls_session);
586 	iscsi_host_remove(shost);
587 	iscsi_host_free(shost);
588 }
589 
590 static inline unsigned int
591 iser_dif_prot_caps(int prot_caps)
592 {
593 	int ret = 0;
594 
595 	if (prot_caps & IB_PROT_T10DIF_TYPE_1)
596 		ret |= SHOST_DIF_TYPE1_PROTECTION |
597 		       SHOST_DIX_TYPE0_PROTECTION |
598 		       SHOST_DIX_TYPE1_PROTECTION;
599 	if (prot_caps & IB_PROT_T10DIF_TYPE_2)
600 		ret |= SHOST_DIF_TYPE2_PROTECTION |
601 		       SHOST_DIX_TYPE2_PROTECTION;
602 	if (prot_caps & IB_PROT_T10DIF_TYPE_3)
603 		ret |= SHOST_DIF_TYPE3_PROTECTION |
604 		       SHOST_DIX_TYPE3_PROTECTION;
605 
606 	return ret;
607 }
608 
609 /**
610  * iscsi_iser_session_create() - create an iscsi-iser session
611  * @ep:             iscsi end-point handle
612  * @cmds_max:       maximum commands in this session
613  * @qdepth:         session command queue depth
614  * @initial_cmdsn:  initiator command sequnce number
615  *
616  * Allocates and adds a scsi host, expose DIF supprot if
617  * exists, and sets up an iscsi session.
618  */
619 static struct iscsi_cls_session *
620 iscsi_iser_session_create(struct iscsi_endpoint *ep,
621 			  uint16_t cmds_max, uint16_t qdepth,
622 			  uint32_t initial_cmdsn)
623 {
624 	struct iscsi_cls_session *cls_session;
625 	struct Scsi_Host *shost;
626 	struct iser_conn *iser_conn = NULL;
627 	struct ib_conn *ib_conn;
628 	struct ib_device *ib_dev;
629 	u32 max_fr_sectors;
630 
631 	shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 0);
632 	if (!shost)
633 		return NULL;
634 	shost->transportt = iscsi_iser_scsi_transport;
635 	shost->cmd_per_lun = qdepth;
636 	shost->max_lun = iscsi_max_lun;
637 	shost->max_id = 0;
638 	shost->max_channel = 0;
639 	shost->max_cmd_len = 16;
640 
641 	/*
642 	 * older userspace tools (before 2.0-870) did not pass us
643 	 * the leading conn's ep so this will be NULL;
644 	 */
645 	if (ep) {
646 		iser_conn = ep->dd_data;
647 		shost->sg_tablesize = iser_conn->scsi_sg_tablesize;
648 		shost->can_queue = min_t(u16, cmds_max, iser_conn->max_cmds);
649 
650 		mutex_lock(&iser_conn->state_mutex);
651 		if (iser_conn->state != ISER_CONN_UP) {
652 			iser_err("iser conn %p already started teardown\n",
653 				 iser_conn);
654 			mutex_unlock(&iser_conn->state_mutex);
655 			goto free_host;
656 		}
657 
658 		ib_conn = &iser_conn->ib_conn;
659 		ib_dev = ib_conn->device->ib_device;
660 		if (ib_conn->pi_support) {
661 			u32 sig_caps = ib_dev->attrs.sig_prot_cap;
662 
663 			shost->sg_prot_tablesize = shost->sg_tablesize;
664 			scsi_host_set_prot(shost, iser_dif_prot_caps(sig_caps));
665 			scsi_host_set_guard(shost, SHOST_DIX_GUARD_IP |
666 						   SHOST_DIX_GUARD_CRC);
667 		}
668 
669 		if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_SG_GAPS_REG))
670 			shost->virt_boundary_mask = SZ_4K - 1;
671 
672 		if (iscsi_host_add(shost, ib_dev->dev.parent)) {
673 			mutex_unlock(&iser_conn->state_mutex);
674 			goto free_host;
675 		}
676 		mutex_unlock(&iser_conn->state_mutex);
677 	} else {
678 		shost->can_queue = min_t(u16, cmds_max, ISER_DEF_XMIT_CMDS_MAX);
679 		if (iscsi_host_add(shost, NULL))
680 			goto free_host;
681 	}
682 
683 	max_fr_sectors = (shost->sg_tablesize * PAGE_SIZE) >> 9;
684 	shost->max_sectors = min(iser_max_sectors, max_fr_sectors);
685 
686 	iser_dbg("iser_conn %p, sg_tablesize %u, max_sectors %u\n",
687 		 iser_conn, shost->sg_tablesize,
688 		 shost->max_sectors);
689 
690 	if (shost->max_sectors < iser_max_sectors)
691 		iser_warn("max_sectors was reduced from %u to %u\n",
692 			  iser_max_sectors, shost->max_sectors);
693 
694 	cls_session = iscsi_session_setup(&iscsi_iser_transport, shost,
695 					  shost->can_queue, 0,
696 					  sizeof(struct iscsi_iser_task),
697 					  initial_cmdsn, 0);
698 	if (!cls_session)
699 		goto remove_host;
700 
701 	return cls_session;
702 
703 remove_host:
704 	iscsi_host_remove(shost);
705 free_host:
706 	iscsi_host_free(shost);
707 	return NULL;
708 }
709 
710 static int
711 iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn,
712 		     enum iscsi_param param, char *buf, int buflen)
713 {
714 	int value;
715 
716 	switch (param) {
717 	case ISCSI_PARAM_MAX_RECV_DLENGTH:
718 		/* TBD */
719 		break;
720 	case ISCSI_PARAM_HDRDGST_EN:
721 		sscanf(buf, "%d", &value);
722 		if (value) {
723 			iser_err("DataDigest wasn't negotiated to None\n");
724 			return -EPROTO;
725 		}
726 		break;
727 	case ISCSI_PARAM_DATADGST_EN:
728 		sscanf(buf, "%d", &value);
729 		if (value) {
730 			iser_err("DataDigest wasn't negotiated to None\n");
731 			return -EPROTO;
732 		}
733 		break;
734 	case ISCSI_PARAM_IFMARKER_EN:
735 		sscanf(buf, "%d", &value);
736 		if (value) {
737 			iser_err("IFMarker wasn't negotiated to No\n");
738 			return -EPROTO;
739 		}
740 		break;
741 	case ISCSI_PARAM_OFMARKER_EN:
742 		sscanf(buf, "%d", &value);
743 		if (value) {
744 			iser_err("OFMarker wasn't negotiated to No\n");
745 			return -EPROTO;
746 		}
747 		break;
748 	default:
749 		return iscsi_set_param(cls_conn, param, buf, buflen);
750 	}
751 
752 	return 0;
753 }
754 
755 /**
756  * iscsi_iser_conn_get_stats() - get iscsi connection statistics
757  * @cls_conn:    iscsi class connection
758  * @stats:       iscsi stats to output
759  *
760  * Output connection statistics.
761  */
762 static void
763 iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
764 {
765 	struct iscsi_conn *conn = cls_conn->dd_data;
766 
767 	stats->txdata_octets = conn->txdata_octets;
768 	stats->rxdata_octets = conn->rxdata_octets;
769 	stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
770 	stats->dataout_pdus = conn->dataout_pdus_cnt;
771 	stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
772 	stats->datain_pdus = conn->datain_pdus_cnt; /* always 0 */
773 	stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */
774 	stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
775 	stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
776 	stats->custom_length = 0;
777 }
778 
779 static int iscsi_iser_get_ep_param(struct iscsi_endpoint *ep,
780 				   enum iscsi_param param, char *buf)
781 {
782 	struct iser_conn *iser_conn = ep->dd_data;
783 
784 	switch (param) {
785 	case ISCSI_PARAM_CONN_PORT:
786 	case ISCSI_PARAM_CONN_ADDRESS:
787 		if (!iser_conn || !iser_conn->ib_conn.cma_id)
788 			return -ENOTCONN;
789 
790 		return iscsi_conn_get_addr_param((struct sockaddr_storage *)
791 				&iser_conn->ib_conn.cma_id->route.addr.dst_addr,
792 				param, buf);
793 	default:
794 		break;
795 	}
796 	return -ENOSYS;
797 }
798 
799 /**
800  * iscsi_iser_ep_connect() - Initiate iSER connection establishment
801  * @shost:          scsi_host
802  * @dst_addr:       destination address
803  * @non_blocking:   indicate if routine can block
804  *
805  * Allocate an iscsi endpoint, an iser_conn structure and bind them.
806  * After that start RDMA connection establishment via rdma_cm. We
807  * don't allocate iser_conn embedded in iscsi_endpoint since in teardown
808  * the endpoint will be destroyed at ep_disconnect while iser_conn will
809  * cleanup its resources asynchronuously.
810  *
811  * Return: iscsi_endpoint created by iscsi layer or ERR_PTR(error)
812  *         if fails.
813  */
814 static struct iscsi_endpoint *
815 iscsi_iser_ep_connect(struct Scsi_Host *shost, struct sockaddr *dst_addr,
816 		      int non_blocking)
817 {
818 	int err;
819 	struct iser_conn *iser_conn;
820 	struct iscsi_endpoint *ep;
821 
822 	ep = iscsi_create_endpoint(0);
823 	if (!ep)
824 		return ERR_PTR(-ENOMEM);
825 
826 	iser_conn = kzalloc(sizeof(*iser_conn), GFP_KERNEL);
827 	if (!iser_conn) {
828 		err = -ENOMEM;
829 		goto failure;
830 	}
831 
832 	ep->dd_data = iser_conn;
833 	iser_conn->ep = ep;
834 	iser_conn_init(iser_conn);
835 
836 	err = iser_connect(iser_conn, NULL, dst_addr, non_blocking);
837 	if (err)
838 		goto failure;
839 
840 	return ep;
841 failure:
842 	iscsi_destroy_endpoint(ep);
843 	return ERR_PTR(err);
844 }
845 
846 /**
847  * iscsi_iser_ep_poll() - poll for iser connection establishment to complete
848  * @ep:            iscsi endpoint (created at ep_connect)
849  * @timeout_ms:    polling timeout allowed in ms.
850  *
851  * This routine boils down to waiting for up_completion signaling
852  * that cma_id got CONNECTED event.
853  *
854  * Return: 1 if succeeded in connection establishment, 0 if timeout expired
855  *         (libiscsi will retry will kick in) or -1 if interrupted by signal
856  *         or more likely iser connection state transitioned to TEMINATING or
857  *         DOWN during the wait period.
858  */
859 static int
860 iscsi_iser_ep_poll(struct iscsi_endpoint *ep, int timeout_ms)
861 {
862 	struct iser_conn *iser_conn = ep->dd_data;
863 	int rc;
864 
865 	rc = wait_for_completion_interruptible_timeout(&iser_conn->up_completion,
866 						       msecs_to_jiffies(timeout_ms));
867 	/* if conn establishment failed, return error code to iscsi */
868 	if (rc == 0) {
869 		mutex_lock(&iser_conn->state_mutex);
870 		if (iser_conn->state == ISER_CONN_TERMINATING ||
871 		    iser_conn->state == ISER_CONN_DOWN)
872 			rc = -1;
873 		mutex_unlock(&iser_conn->state_mutex);
874 	}
875 
876 	iser_info("iser conn %p rc = %d\n", iser_conn, rc);
877 
878 	if (rc > 0)
879 		return 1; /* success, this is the equivalent of EPOLLOUT */
880 	else if (!rc)
881 		return 0; /* timeout */
882 	else
883 		return rc; /* signal */
884 }
885 
886 /**
887  * iscsi_iser_ep_disconnect() - Initiate connection teardown process
888  * @ep:    iscsi endpoint handle
889  *
890  * This routine is not blocked by iser and RDMA termination process
891  * completion as we queue a deffered work for iser/RDMA destruction
892  * and cleanup or actually call it immediately in case we didn't pass
893  * iscsi conn bind/start stage, thus it is safe.
894  */
895 static void
896 iscsi_iser_ep_disconnect(struct iscsi_endpoint *ep)
897 {
898 	struct iser_conn *iser_conn = ep->dd_data;
899 
900 	iser_info("ep %p iser conn %p\n", ep, iser_conn);
901 
902 	mutex_lock(&iser_conn->state_mutex);
903 	iser_conn_terminate(iser_conn);
904 
905 	/*
906 	 * if iser_conn and iscsi_conn are bound, we must wait for
907 	 * iscsi_conn_stop and flush errors completion before freeing
908 	 * the iser resources. Otherwise we are safe to free resources
909 	 * immediately.
910 	 */
911 	if (iser_conn->iscsi_conn) {
912 		INIT_WORK(&iser_conn->release_work, iser_release_work);
913 		queue_work(release_wq, &iser_conn->release_work);
914 		mutex_unlock(&iser_conn->state_mutex);
915 	} else {
916 		iser_conn->state = ISER_CONN_DOWN;
917 		mutex_unlock(&iser_conn->state_mutex);
918 		iser_conn_release(iser_conn);
919 	}
920 
921 	iscsi_destroy_endpoint(ep);
922 }
923 
924 static umode_t iser_attr_is_visible(int param_type, int param)
925 {
926 	switch (param_type) {
927 	case ISCSI_HOST_PARAM:
928 		switch (param) {
929 		case ISCSI_HOST_PARAM_NETDEV_NAME:
930 		case ISCSI_HOST_PARAM_HWADDRESS:
931 		case ISCSI_HOST_PARAM_INITIATOR_NAME:
932 			return S_IRUGO;
933 		default:
934 			return 0;
935 		}
936 	case ISCSI_PARAM:
937 		switch (param) {
938 		case ISCSI_PARAM_MAX_RECV_DLENGTH:
939 		case ISCSI_PARAM_MAX_XMIT_DLENGTH:
940 		case ISCSI_PARAM_HDRDGST_EN:
941 		case ISCSI_PARAM_DATADGST_EN:
942 		case ISCSI_PARAM_CONN_ADDRESS:
943 		case ISCSI_PARAM_CONN_PORT:
944 		case ISCSI_PARAM_EXP_STATSN:
945 		case ISCSI_PARAM_PERSISTENT_ADDRESS:
946 		case ISCSI_PARAM_PERSISTENT_PORT:
947 		case ISCSI_PARAM_PING_TMO:
948 		case ISCSI_PARAM_RECV_TMO:
949 		case ISCSI_PARAM_INITIAL_R2T_EN:
950 		case ISCSI_PARAM_MAX_R2T:
951 		case ISCSI_PARAM_IMM_DATA_EN:
952 		case ISCSI_PARAM_FIRST_BURST:
953 		case ISCSI_PARAM_MAX_BURST:
954 		case ISCSI_PARAM_PDU_INORDER_EN:
955 		case ISCSI_PARAM_DATASEQ_INORDER_EN:
956 		case ISCSI_PARAM_TARGET_NAME:
957 		case ISCSI_PARAM_TPGT:
958 		case ISCSI_PARAM_USERNAME:
959 		case ISCSI_PARAM_PASSWORD:
960 		case ISCSI_PARAM_USERNAME_IN:
961 		case ISCSI_PARAM_PASSWORD_IN:
962 		case ISCSI_PARAM_FAST_ABORT:
963 		case ISCSI_PARAM_ABORT_TMO:
964 		case ISCSI_PARAM_LU_RESET_TMO:
965 		case ISCSI_PARAM_TGT_RESET_TMO:
966 		case ISCSI_PARAM_IFACE_NAME:
967 		case ISCSI_PARAM_INITIATOR_NAME:
968 		case ISCSI_PARAM_DISCOVERY_SESS:
969 			return S_IRUGO;
970 		default:
971 			return 0;
972 		}
973 	}
974 
975 	return 0;
976 }
977 
978 static struct scsi_host_template iscsi_iser_sht = {
979 	.module                 = THIS_MODULE,
980 	.name                   = "iSCSI Initiator over iSER",
981 	.queuecommand           = iscsi_queuecommand,
982 	.change_queue_depth	= scsi_change_queue_depth,
983 	.sg_tablesize           = ISCSI_ISER_DEF_SG_TABLESIZE,
984 	.cmd_per_lun            = ISER_DEF_CMD_PER_LUN,
985 	.eh_timed_out		= iscsi_eh_cmd_timed_out,
986 	.eh_abort_handler       = iscsi_eh_abort,
987 	.eh_device_reset_handler= iscsi_eh_device_reset,
988 	.eh_target_reset_handler = iscsi_eh_recover_target,
989 	.target_alloc		= iscsi_target_alloc,
990 	.proc_name              = "iscsi_iser",
991 	.this_id                = -1,
992 	.track_queue_depth	= 1,
993 };
994 
995 static struct iscsi_transport iscsi_iser_transport = {
996 	.owner                  = THIS_MODULE,
997 	.name                   = "iser",
998 	.caps                   = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_TEXT_NEGO,
999 	/* session management */
1000 	.create_session         = iscsi_iser_session_create,
1001 	.destroy_session        = iscsi_iser_session_destroy,
1002 	/* connection management */
1003 	.create_conn            = iscsi_iser_conn_create,
1004 	.bind_conn              = iscsi_iser_conn_bind,
1005 	.destroy_conn           = iscsi_conn_teardown,
1006 	.attr_is_visible	= iser_attr_is_visible,
1007 	.set_param              = iscsi_iser_set_param,
1008 	.get_conn_param		= iscsi_conn_get_param,
1009 	.get_ep_param		= iscsi_iser_get_ep_param,
1010 	.get_session_param	= iscsi_session_get_param,
1011 	.start_conn             = iscsi_iser_conn_start,
1012 	.stop_conn              = iscsi_iser_conn_stop,
1013 	/* iscsi host params */
1014 	.get_host_param		= iscsi_host_get_param,
1015 	.set_host_param		= iscsi_host_set_param,
1016 	/* IO */
1017 	.send_pdu		= iscsi_conn_send_pdu,
1018 	.get_stats		= iscsi_iser_conn_get_stats,
1019 	.init_task		= iscsi_iser_task_init,
1020 	.xmit_task		= iscsi_iser_task_xmit,
1021 	.cleanup_task		= iscsi_iser_cleanup_task,
1022 	.alloc_pdu		= iscsi_iser_pdu_alloc,
1023 	.check_protection	= iscsi_iser_check_protection,
1024 	/* recovery */
1025 	.session_recovery_timedout = iscsi_session_recovery_timedout,
1026 
1027 	.ep_connect             = iscsi_iser_ep_connect,
1028 	.ep_poll                = iscsi_iser_ep_poll,
1029 	.ep_disconnect          = iscsi_iser_ep_disconnect
1030 };
1031 
1032 static int __init iser_init(void)
1033 {
1034 	int err;
1035 
1036 	iser_dbg("Starting iSER datamover...\n");
1037 
1038 	memset(&ig, 0, sizeof(struct iser_global));
1039 
1040 	ig.desc_cache = kmem_cache_create("iser_descriptors",
1041 					  sizeof(struct iser_tx_desc),
1042 					  0, SLAB_HWCACHE_ALIGN,
1043 					  NULL);
1044 	if (ig.desc_cache == NULL)
1045 		return -ENOMEM;
1046 
1047 	/* device init is called only after the first addr resolution */
1048 	mutex_init(&ig.device_list_mutex);
1049 	INIT_LIST_HEAD(&ig.device_list);
1050 	mutex_init(&ig.connlist_mutex);
1051 	INIT_LIST_HEAD(&ig.connlist);
1052 
1053 	release_wq = alloc_workqueue("release workqueue", 0, 0);
1054 	if (!release_wq) {
1055 		iser_err("failed to allocate release workqueue\n");
1056 		err = -ENOMEM;
1057 		goto err_alloc_wq;
1058 	}
1059 
1060 	iscsi_iser_scsi_transport = iscsi_register_transport(
1061 							&iscsi_iser_transport);
1062 	if (!iscsi_iser_scsi_transport) {
1063 		iser_err("iscsi_register_transport failed\n");
1064 		err = -EINVAL;
1065 		goto err_reg;
1066 	}
1067 
1068 	return 0;
1069 
1070 err_reg:
1071 	destroy_workqueue(release_wq);
1072 err_alloc_wq:
1073 	kmem_cache_destroy(ig.desc_cache);
1074 
1075 	return err;
1076 }
1077 
1078 static void __exit iser_exit(void)
1079 {
1080 	struct iser_conn *iser_conn, *n;
1081 	int connlist_empty;
1082 
1083 	iser_dbg("Removing iSER datamover...\n");
1084 	destroy_workqueue(release_wq);
1085 
1086 	mutex_lock(&ig.connlist_mutex);
1087 	connlist_empty = list_empty(&ig.connlist);
1088 	mutex_unlock(&ig.connlist_mutex);
1089 
1090 	if (!connlist_empty) {
1091 		iser_err("Error cleanup stage completed but we still have iser "
1092 			 "connections, destroying them anyway\n");
1093 		list_for_each_entry_safe(iser_conn, n, &ig.connlist,
1094 					 conn_list) {
1095 			iser_conn_release(iser_conn);
1096 		}
1097 	}
1098 
1099 	iscsi_unregister_transport(&iscsi_iser_transport);
1100 	kmem_cache_destroy(ig.desc_cache);
1101 }
1102 
1103 module_init(iser_init);
1104 module_exit(iser_exit);
1105