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, 0:CRC|1:IP_CSUM (default:IP_CSUM)");
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
iscsi_iser_recv(struct iscsi_conn * conn,struct iscsi_hdr * hdr,char * rx_data,int rx_data_len)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
iscsi_iser_pdu_alloc(struct iscsi_task * task,uint8_t opcode)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
iser_initialize_task_headers(struct iscsi_task * task,struct iser_tx_desc * tx_desc)167 int iser_initialize_task_headers(struct iscsi_task *task,
168 struct iser_tx_desc *tx_desc)
169 {
170 struct iser_conn *iser_conn = task->conn->dd_data;
171 struct iser_device *device = iser_conn->ib_conn.device;
172 struct iscsi_iser_task *iser_task = task->dd_data;
173 u64 dma_addr;
174
175 dma_addr = ib_dma_map_single(device->ib_device, (void *)tx_desc,
176 ISER_HEADERS_LEN, DMA_TO_DEVICE);
177 if (ib_dma_mapping_error(device->ib_device, dma_addr))
178 return -ENOMEM;
179
180 tx_desc->dma_addr = dma_addr;
181 tx_desc->tx_sg[0].addr = tx_desc->dma_addr;
182 tx_desc->tx_sg[0].length = ISER_HEADERS_LEN;
183 tx_desc->tx_sg[0].lkey = device->mr->lkey;
184
185 iser_task->iser_conn = iser_conn;
186 return 0;
187 }
188
189 /**
190 * iscsi_iser_task_init() - Initialize iscsi-iser task
191 * @task: iscsi task
192 *
193 * Initialize the task for the scsi command or mgmt command.
194 *
195 * Return: Returns zero on success or -ENOMEM when failing
196 * to init task headers (dma mapping error).
197 */
198 static int
iscsi_iser_task_init(struct iscsi_task * task)199 iscsi_iser_task_init(struct iscsi_task *task)
200 {
201 struct iscsi_iser_task *iser_task = task->dd_data;
202
203 if (iser_initialize_task_headers(task, &iser_task->desc))
204 return -ENOMEM;
205
206 /* mgmt task */
207 if (!task->sc)
208 return 0;
209
210 iser_task->command_sent = 0;
211 iser_task_rdma_init(iser_task);
212 iser_task->sc = task->sc;
213
214 return 0;
215 }
216
217 /**
218 * iscsi_iser_mtask_xmit() - xmit management (immediate) task
219 * @conn: iscsi connection
220 * @task: task management task
221 *
222 * Notes:
223 * The function can return -EAGAIN in which case caller must
224 * call it again later, or recover. '0' return code means successful
225 * xmit.
226 *
227 **/
228 static int
iscsi_iser_mtask_xmit(struct iscsi_conn * conn,struct iscsi_task * task)229 iscsi_iser_mtask_xmit(struct iscsi_conn *conn, struct iscsi_task *task)
230 {
231 int error = 0;
232
233 iser_dbg("mtask xmit [cid %d itt 0x%x]\n", conn->id, task->itt);
234
235 error = iser_send_control(conn, task);
236
237 /* since iser xmits control with zero copy, tasks can not be recycled
238 * right after sending them.
239 * The recycling scheme is based on whether a response is expected
240 * - if yes, the task is recycled at iscsi_complete_pdu
241 * - if no, the task is recycled at iser_snd_completion
242 */
243 return error;
244 }
245
246 static int
iscsi_iser_task_xmit_unsol_data(struct iscsi_conn * conn,struct iscsi_task * task)247 iscsi_iser_task_xmit_unsol_data(struct iscsi_conn *conn,
248 struct iscsi_task *task)
249 {
250 struct iscsi_r2t_info *r2t = &task->unsol_r2t;
251 struct iscsi_data hdr;
252 int error = 0;
253
254 /* Send data-out PDUs while there's still unsolicited data to send */
255 while (iscsi_task_has_unsol_data(task)) {
256 iscsi_prep_data_out_pdu(task, r2t, &hdr);
257 iser_dbg("Sending data-out: itt 0x%x, data count %d\n",
258 hdr.itt, r2t->data_count);
259
260 /* the buffer description has been passed with the command */
261 /* Send the command */
262 error = iser_send_data_out(conn, task, &hdr);
263 if (error) {
264 r2t->datasn--;
265 goto iscsi_iser_task_xmit_unsol_data_exit;
266 }
267 r2t->sent += r2t->data_count;
268 iser_dbg("Need to send %d more as data-out PDUs\n",
269 r2t->data_length - r2t->sent);
270 }
271
272 iscsi_iser_task_xmit_unsol_data_exit:
273 return error;
274 }
275
276 /**
277 * iscsi_iser_task_xmit() - xmit iscsi-iser task
278 * @task: iscsi task
279 *
280 * Return: zero on success or escalates $error on failure.
281 */
282 static int
iscsi_iser_task_xmit(struct iscsi_task * task)283 iscsi_iser_task_xmit(struct iscsi_task *task)
284 {
285 struct iscsi_conn *conn = task->conn;
286 struct iscsi_iser_task *iser_task = task->dd_data;
287 int error = 0;
288
289 if (!task->sc)
290 return iscsi_iser_mtask_xmit(conn, task);
291
292 if (task->sc->sc_data_direction == DMA_TO_DEVICE) {
293 BUG_ON(scsi_bufflen(task->sc) == 0);
294
295 iser_dbg("cmd [itt %x total %d imm %d unsol_data %d\n",
296 task->itt, scsi_bufflen(task->sc),
297 task->imm_count, task->unsol_r2t.data_length);
298 }
299
300 iser_dbg("ctask xmit [cid %d itt 0x%x]\n",
301 conn->id, task->itt);
302
303 /* Send the cmd PDU */
304 if (!iser_task->command_sent) {
305 error = iser_send_command(conn, task);
306 if (error)
307 goto iscsi_iser_task_xmit_exit;
308 iser_task->command_sent = 1;
309 }
310
311 /* Send unsolicited data-out PDU(s) if necessary */
312 if (iscsi_task_has_unsol_data(task))
313 error = iscsi_iser_task_xmit_unsol_data(conn, task);
314
315 iscsi_iser_task_xmit_exit:
316 return error;
317 }
318
319 /**
320 * iscsi_iser_cleanup_task() - cleanup an iscsi-iser task
321 * @task: iscsi task
322 *
323 * Notes: In case the RDMA device is already NULL (might have
324 * been removed in DEVICE_REMOVAL CM event it will bail-out
325 * without doing dma unmapping.
326 */
iscsi_iser_cleanup_task(struct iscsi_task * task)327 static void iscsi_iser_cleanup_task(struct iscsi_task *task)
328 {
329 struct iscsi_iser_task *iser_task = task->dd_data;
330 struct iser_tx_desc *tx_desc = &iser_task->desc;
331 struct iser_conn *iser_conn = task->conn->dd_data;
332 struct iser_device *device = iser_conn->ib_conn.device;
333
334 /* DEVICE_REMOVAL event might have already released the device */
335 if (!device)
336 return;
337
338 ib_dma_unmap_single(device->ib_device,
339 tx_desc->dma_addr, ISER_HEADERS_LEN, DMA_TO_DEVICE);
340
341 /* mgmt tasks do not need special cleanup */
342 if (!task->sc)
343 return;
344
345 if (iser_task->status == ISER_TASK_STATUS_STARTED) {
346 iser_task->status = ISER_TASK_STATUS_COMPLETED;
347 iser_task_rdma_finalize(iser_task);
348 }
349 }
350
351 /**
352 * iscsi_iser_check_protection() - check protection information status of task.
353 * @task: iscsi task
354 * @sector: error sector if exsists (output)
355 *
356 * Return: zero if no data-integrity errors have occured
357 * 0x1: data-integrity error occured in the guard-block
358 * 0x2: data-integrity error occured in the reference tag
359 * 0x3: data-integrity error occured in the application tag
360 *
361 * In addition the error sector is marked.
362 */
363 static u8
iscsi_iser_check_protection(struct iscsi_task * task,sector_t * sector)364 iscsi_iser_check_protection(struct iscsi_task *task, sector_t *sector)
365 {
366 struct iscsi_iser_task *iser_task = task->dd_data;
367
368 if (iser_task->dir[ISER_DIR_IN])
369 return iser_check_task_pi_status(iser_task, ISER_DIR_IN,
370 sector);
371 else
372 return iser_check_task_pi_status(iser_task, ISER_DIR_OUT,
373 sector);
374 }
375
376 /**
377 * iscsi_iser_conn_create() - create a new iscsi-iser connection
378 * @cls_session: iscsi class connection
379 * @conn_idx: connection index within the session (for MCS)
380 *
381 * Return: iscsi_cls_conn when iscsi_conn_setup succeeds or NULL
382 * otherwise.
383 */
384 static struct iscsi_cls_conn *
iscsi_iser_conn_create(struct iscsi_cls_session * cls_session,uint32_t conn_idx)385 iscsi_iser_conn_create(struct iscsi_cls_session *cls_session,
386 uint32_t conn_idx)
387 {
388 struct iscsi_conn *conn;
389 struct iscsi_cls_conn *cls_conn;
390
391 cls_conn = iscsi_conn_setup(cls_session, 0, conn_idx);
392 if (!cls_conn)
393 return NULL;
394 conn = cls_conn->dd_data;
395
396 /*
397 * due to issues with the login code re iser sematics
398 * this not set in iscsi_conn_setup - FIXME
399 */
400 conn->max_recv_dlength = ISER_RECV_DATA_SEG_LEN;
401
402 return cls_conn;
403 }
404
405 /**
406 * iscsi_iser_conn_bind() - bind iscsi and iser connection structures
407 * @cls_session: iscsi class session
408 * @cls_conn: iscsi class connection
409 * @transport_eph: transport end-point handle
410 * @is_leading: indicate if this is the session leading connection (MCS)
411 *
412 * Return: zero on success, $error if iscsi_conn_bind fails and
413 * -EINVAL in case end-point doesn't exsits anymore or iser connection
414 * state is not UP (teardown already started).
415 */
416 static int
iscsi_iser_conn_bind(struct iscsi_cls_session * cls_session,struct iscsi_cls_conn * cls_conn,uint64_t transport_eph,int is_leading)417 iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session,
418 struct iscsi_cls_conn *cls_conn,
419 uint64_t transport_eph,
420 int is_leading)
421 {
422 struct iscsi_conn *conn = cls_conn->dd_data;
423 struct iser_conn *iser_conn;
424 struct iscsi_endpoint *ep;
425 int error;
426
427 error = iscsi_conn_bind(cls_session, cls_conn, is_leading);
428 if (error)
429 return error;
430
431 /* the transport ep handle comes from user space so it must be
432 * verified against the global ib connections list */
433 ep = iscsi_lookup_endpoint(transport_eph);
434 if (!ep) {
435 iser_err("can't bind eph %llx\n",
436 (unsigned long long)transport_eph);
437 return -EINVAL;
438 }
439 iser_conn = ep->dd_data;
440
441 mutex_lock(&iser_conn->state_mutex);
442 if (iser_conn->state != ISER_CONN_UP) {
443 error = -EINVAL;
444 iser_err("iser_conn %p state is %d, teardown started\n",
445 iser_conn, iser_conn->state);
446 goto out;
447 }
448
449 error = iser_alloc_rx_descriptors(iser_conn, conn->session);
450 if (error)
451 goto out;
452
453 /* binds the iSER connection retrieved from the previously
454 * connected ep_handle to the iSCSI layer connection. exchanges
455 * connection pointers */
456 iser_info("binding iscsi conn %p to iser_conn %p\n", conn, iser_conn);
457
458 conn->dd_data = iser_conn;
459 iser_conn->iscsi_conn = conn;
460
461 out:
462 mutex_unlock(&iser_conn->state_mutex);
463 return error;
464 }
465
466 /**
467 * iscsi_iser_conn_start() - start iscsi-iser connection
468 * @cls_conn: iscsi class connection
469 *
470 * Notes: Here iser intialize (or re-initialize) stop_completion as
471 * from this point iscsi must call conn_stop in session/connection
472 * teardown so iser transport must wait for it.
473 */
474 static int
iscsi_iser_conn_start(struct iscsi_cls_conn * cls_conn)475 iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn)
476 {
477 struct iscsi_conn *iscsi_conn;
478 struct iser_conn *iser_conn;
479
480 iscsi_conn = cls_conn->dd_data;
481 iser_conn = iscsi_conn->dd_data;
482 reinit_completion(&iser_conn->stop_completion);
483
484 return iscsi_conn_start(cls_conn);
485 }
486
487 /**
488 * iscsi_iser_conn_stop() - stop iscsi-iser connection
489 * @cls_conn: iscsi class connection
490 * @flag: indicate if recover or terminate (passed as is)
491 *
492 * Notes: Calling iscsi_conn_stop might theoretically race with
493 * DEVICE_REMOVAL event and dereference a previously freed RDMA device
494 * handle, so we call it under iser the state lock to protect against
495 * this kind of race.
496 */
497 static void
iscsi_iser_conn_stop(struct iscsi_cls_conn * cls_conn,int flag)498 iscsi_iser_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
499 {
500 struct iscsi_conn *conn = cls_conn->dd_data;
501 struct iser_conn *iser_conn = conn->dd_data;
502
503 iser_info("stopping iscsi_conn: %p, iser_conn: %p\n", conn, iser_conn);
504
505 /*
506 * Userspace may have goofed up and not bound the connection or
507 * might have only partially setup the connection.
508 */
509 if (iser_conn) {
510 mutex_lock(&iser_conn->state_mutex);
511 iscsi_conn_stop(cls_conn, flag);
512 iser_conn_terminate(iser_conn);
513
514 /* unbind */
515 iser_conn->iscsi_conn = NULL;
516 conn->dd_data = NULL;
517
518 complete(&iser_conn->stop_completion);
519 mutex_unlock(&iser_conn->state_mutex);
520 } else {
521 iscsi_conn_stop(cls_conn, flag);
522 }
523 }
524
525 /**
526 * iscsi_iser_session_destroy() - destroy iscsi-iser session
527 * @cls_session: iscsi class session
528 *
529 * Removes and free iscsi host.
530 */
531 static void
iscsi_iser_session_destroy(struct iscsi_cls_session * cls_session)532 iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session)
533 {
534 struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
535
536 iscsi_session_teardown(cls_session);
537 iscsi_host_remove(shost);
538 iscsi_host_free(shost);
539 }
540
541 static inline unsigned int
iser_dif_prot_caps(int prot_caps)542 iser_dif_prot_caps(int prot_caps)
543 {
544 return ((prot_caps & IB_PROT_T10DIF_TYPE_1) ? SHOST_DIF_TYPE1_PROTECTION |
545 SHOST_DIX_TYPE1_PROTECTION : 0) |
546 ((prot_caps & IB_PROT_T10DIF_TYPE_2) ? SHOST_DIF_TYPE2_PROTECTION |
547 SHOST_DIX_TYPE2_PROTECTION : 0) |
548 ((prot_caps & IB_PROT_T10DIF_TYPE_3) ? SHOST_DIF_TYPE3_PROTECTION |
549 SHOST_DIX_TYPE3_PROTECTION : 0);
550 }
551
552 /**
553 * iscsi_iser_session_create() - create an iscsi-iser session
554 * @ep: iscsi end-point handle
555 * @cmds_max: maximum commands in this session
556 * @qdepth: session command queue depth
557 * @initial_cmdsn: initiator command sequnce number
558 *
559 * Allocates and adds a scsi host, expose DIF supprot if
560 * exists, and sets up an iscsi session.
561 */
562 static struct iscsi_cls_session *
iscsi_iser_session_create(struct iscsi_endpoint * ep,uint16_t cmds_max,uint16_t qdepth,uint32_t initial_cmdsn)563 iscsi_iser_session_create(struct iscsi_endpoint *ep,
564 uint16_t cmds_max, uint16_t qdepth,
565 uint32_t initial_cmdsn)
566 {
567 struct iscsi_cls_session *cls_session;
568 struct iscsi_session *session;
569 struct Scsi_Host *shost;
570 struct iser_conn *iser_conn = NULL;
571 struct ib_conn *ib_conn;
572
573 shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 0);
574 if (!shost)
575 return NULL;
576 shost->transportt = iscsi_iser_scsi_transport;
577 shost->cmd_per_lun = qdepth;
578 shost->max_lun = iscsi_max_lun;
579 shost->max_id = 0;
580 shost->max_channel = 0;
581 shost->max_cmd_len = 16;
582
583 /*
584 * older userspace tools (before 2.0-870) did not pass us
585 * the leading conn's ep so this will be NULL;
586 */
587 if (ep) {
588 iser_conn = ep->dd_data;
589 ib_conn = &iser_conn->ib_conn;
590 if (ib_conn->pi_support) {
591 u32 sig_caps = ib_conn->device->dev_attr.sig_prot_cap;
592
593 scsi_host_set_prot(shost, iser_dif_prot_caps(sig_caps));
594 if (iser_pi_guard)
595 scsi_host_set_guard(shost, SHOST_DIX_GUARD_IP);
596 else
597 scsi_host_set_guard(shost, SHOST_DIX_GUARD_CRC);
598 }
599 }
600
601 if (iscsi_host_add(shost, ep ?
602 ib_conn->device->ib_device->dma_device : NULL))
603 goto free_host;
604
605 if (cmds_max > ISER_DEF_XMIT_CMDS_MAX) {
606 iser_info("cmds_max changed from %u to %u\n",
607 cmds_max, ISER_DEF_XMIT_CMDS_MAX);
608 cmds_max = ISER_DEF_XMIT_CMDS_MAX;
609 }
610
611 cls_session = iscsi_session_setup(&iscsi_iser_transport, shost,
612 cmds_max, 0,
613 sizeof(struct iscsi_iser_task),
614 initial_cmdsn, 0);
615 if (!cls_session)
616 goto remove_host;
617 session = cls_session->dd_data;
618
619 shost->can_queue = session->scsi_cmds_max;
620 return cls_session;
621
622 remove_host:
623 iscsi_host_remove(shost);
624 free_host:
625 iscsi_host_free(shost);
626 return NULL;
627 }
628
629 static int
iscsi_iser_set_param(struct iscsi_cls_conn * cls_conn,enum iscsi_param param,char * buf,int buflen)630 iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn,
631 enum iscsi_param param, char *buf, int buflen)
632 {
633 int value;
634
635 switch (param) {
636 case ISCSI_PARAM_MAX_RECV_DLENGTH:
637 /* TBD */
638 break;
639 case ISCSI_PARAM_HDRDGST_EN:
640 sscanf(buf, "%d", &value);
641 if (value) {
642 iser_err("DataDigest wasn't negotiated to None\n");
643 return -EPROTO;
644 }
645 break;
646 case ISCSI_PARAM_DATADGST_EN:
647 sscanf(buf, "%d", &value);
648 if (value) {
649 iser_err("DataDigest wasn't negotiated to None\n");
650 return -EPROTO;
651 }
652 break;
653 case ISCSI_PARAM_IFMARKER_EN:
654 sscanf(buf, "%d", &value);
655 if (value) {
656 iser_err("IFMarker wasn't negotiated to No\n");
657 return -EPROTO;
658 }
659 break;
660 case ISCSI_PARAM_OFMARKER_EN:
661 sscanf(buf, "%d", &value);
662 if (value) {
663 iser_err("OFMarker wasn't negotiated to No\n");
664 return -EPROTO;
665 }
666 break;
667 default:
668 return iscsi_set_param(cls_conn, param, buf, buflen);
669 }
670
671 return 0;
672 }
673
674 /**
675 * iscsi_iser_set_param() - set class connection parameter
676 * @cls_conn: iscsi class connection
677 * @stats: iscsi stats to output
678 *
679 * Output connection statistics.
680 */
681 static void
iscsi_iser_conn_get_stats(struct iscsi_cls_conn * cls_conn,struct iscsi_stats * stats)682 iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
683 {
684 struct iscsi_conn *conn = cls_conn->dd_data;
685
686 stats->txdata_octets = conn->txdata_octets;
687 stats->rxdata_octets = conn->rxdata_octets;
688 stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
689 stats->dataout_pdus = conn->dataout_pdus_cnt;
690 stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
691 stats->datain_pdus = conn->datain_pdus_cnt; /* always 0 */
692 stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */
693 stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
694 stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
695 stats->custom_length = 4;
696 strcpy(stats->custom[0].desc, "qp_tx_queue_full");
697 stats->custom[0].value = 0; /* TB iser_conn->qp_tx_queue_full; */
698 strcpy(stats->custom[1].desc, "fmr_map_not_avail");
699 stats->custom[1].value = 0; /* TB iser_conn->fmr_map_not_avail */;
700 strcpy(stats->custom[2].desc, "eh_abort_cnt");
701 stats->custom[2].value = conn->eh_abort_cnt;
702 strcpy(stats->custom[3].desc, "fmr_unalign_cnt");
703 stats->custom[3].value = conn->fmr_unalign_cnt;
704 }
705
iscsi_iser_get_ep_param(struct iscsi_endpoint * ep,enum iscsi_param param,char * buf)706 static int iscsi_iser_get_ep_param(struct iscsi_endpoint *ep,
707 enum iscsi_param param, char *buf)
708 {
709 struct iser_conn *iser_conn = ep->dd_data;
710 int len;
711
712 switch (param) {
713 case ISCSI_PARAM_CONN_PORT:
714 case ISCSI_PARAM_CONN_ADDRESS:
715 if (!iser_conn || !iser_conn->ib_conn.cma_id)
716 return -ENOTCONN;
717
718 return iscsi_conn_get_addr_param((struct sockaddr_storage *)
719 &iser_conn->ib_conn.cma_id->route.addr.dst_addr,
720 param, buf);
721 break;
722 default:
723 return -ENOSYS;
724 }
725
726 return len;
727 }
728
729 /**
730 * iscsi_iser_ep_connect() - Initiate iSER connection establishment
731 * @shost: scsi_host
732 * @dst_addr: destination address
733 * @non-blocking: indicate if routine can block
734 *
735 * Allocate an iscsi endpoint, an iser_conn structure and bind them.
736 * After that start RDMA connection establishment via rdma_cm. We
737 * don't allocate iser_conn embedded in iscsi_endpoint since in teardown
738 * the endpoint will be destroyed at ep_disconnect while iser_conn will
739 * cleanup its resources asynchronuously.
740 *
741 * Return: iscsi_endpoint created by iscsi layer or ERR_PTR(error)
742 * if fails.
743 */
744 static struct iscsi_endpoint *
iscsi_iser_ep_connect(struct Scsi_Host * shost,struct sockaddr * dst_addr,int non_blocking)745 iscsi_iser_ep_connect(struct Scsi_Host *shost, struct sockaddr *dst_addr,
746 int non_blocking)
747 {
748 int err;
749 struct iser_conn *iser_conn;
750 struct iscsi_endpoint *ep;
751
752 ep = iscsi_create_endpoint(0);
753 if (!ep)
754 return ERR_PTR(-ENOMEM);
755
756 iser_conn = kzalloc(sizeof(*iser_conn), GFP_KERNEL);
757 if (!iser_conn) {
758 err = -ENOMEM;
759 goto failure;
760 }
761
762 ep->dd_data = iser_conn;
763 iser_conn->ep = ep;
764 iser_conn_init(iser_conn);
765
766 err = iser_connect(iser_conn, NULL, dst_addr, non_blocking);
767 if (err)
768 goto failure;
769
770 return ep;
771 failure:
772 iscsi_destroy_endpoint(ep);
773 return ERR_PTR(err);
774 }
775
776 /**
777 * iscsi_iser_ep_poll() - poll for iser connection establishment to complete
778 * @ep: iscsi endpoint (created at ep_connect)
779 * @timeout_ms: polling timeout allowed in ms.
780 *
781 * This routine boils down to waiting for up_completion signaling
782 * that cma_id got CONNECTED event.
783 *
784 * Return: 1 if succeeded in connection establishment, 0 if timeout expired
785 * (libiscsi will retry will kick in) or -1 if interrupted by signal
786 * or more likely iser connection state transitioned to TEMINATING or
787 * DOWN during the wait period.
788 */
789 static int
iscsi_iser_ep_poll(struct iscsi_endpoint * ep,int timeout_ms)790 iscsi_iser_ep_poll(struct iscsi_endpoint *ep, int timeout_ms)
791 {
792 struct iser_conn *iser_conn;
793 int rc;
794
795 iser_conn = ep->dd_data;
796 rc = wait_for_completion_interruptible_timeout(&iser_conn->up_completion,
797 msecs_to_jiffies(timeout_ms));
798 /* if conn establishment failed, return error code to iscsi */
799 if (rc == 0) {
800 mutex_lock(&iser_conn->state_mutex);
801 if (iser_conn->state == ISER_CONN_TERMINATING ||
802 iser_conn->state == ISER_CONN_DOWN)
803 rc = -1;
804 mutex_unlock(&iser_conn->state_mutex);
805 }
806
807 iser_info("ib conn %p rc = %d\n", iser_conn, rc);
808
809 if (rc > 0)
810 return 1; /* success, this is the equivalent of POLLOUT */
811 else if (!rc)
812 return 0; /* timeout */
813 else
814 return rc; /* signal */
815 }
816
817 /**
818 * iscsi_iser_ep_disconnect() - Initiate connection teardown process
819 * @ep: iscsi endpoint handle
820 *
821 * This routine is not blocked by iser and RDMA termination process
822 * completion as we queue a deffered work for iser/RDMA destruction
823 * and cleanup or actually call it immediately in case we didn't pass
824 * iscsi conn bind/start stage, thus it is safe.
825 */
826 static void
iscsi_iser_ep_disconnect(struct iscsi_endpoint * ep)827 iscsi_iser_ep_disconnect(struct iscsi_endpoint *ep)
828 {
829 struct iser_conn *iser_conn;
830
831 iser_conn = ep->dd_data;
832 iser_info("ep %p iser conn %p state %d\n",
833 ep, iser_conn, iser_conn->state);
834
835 mutex_lock(&iser_conn->state_mutex);
836 iser_conn_terminate(iser_conn);
837
838 /*
839 * if iser_conn and iscsi_conn are bound, we must wait for
840 * iscsi_conn_stop and flush errors completion before freeing
841 * the iser resources. Otherwise we are safe to free resources
842 * immediately.
843 */
844 if (iser_conn->iscsi_conn) {
845 INIT_WORK(&iser_conn->release_work, iser_release_work);
846 queue_work(release_wq, &iser_conn->release_work);
847 mutex_unlock(&iser_conn->state_mutex);
848 } else {
849 iser_conn->state = ISER_CONN_DOWN;
850 mutex_unlock(&iser_conn->state_mutex);
851 iser_conn_release(iser_conn);
852 }
853 iscsi_destroy_endpoint(ep);
854 }
855
iser_attr_is_visible(int param_type,int param)856 static umode_t iser_attr_is_visible(int param_type, int param)
857 {
858 switch (param_type) {
859 case ISCSI_HOST_PARAM:
860 switch (param) {
861 case ISCSI_HOST_PARAM_NETDEV_NAME:
862 case ISCSI_HOST_PARAM_HWADDRESS:
863 case ISCSI_HOST_PARAM_INITIATOR_NAME:
864 return S_IRUGO;
865 default:
866 return 0;
867 }
868 case ISCSI_PARAM:
869 switch (param) {
870 case ISCSI_PARAM_MAX_RECV_DLENGTH:
871 case ISCSI_PARAM_MAX_XMIT_DLENGTH:
872 case ISCSI_PARAM_HDRDGST_EN:
873 case ISCSI_PARAM_DATADGST_EN:
874 case ISCSI_PARAM_CONN_ADDRESS:
875 case ISCSI_PARAM_CONN_PORT:
876 case ISCSI_PARAM_EXP_STATSN:
877 case ISCSI_PARAM_PERSISTENT_ADDRESS:
878 case ISCSI_PARAM_PERSISTENT_PORT:
879 case ISCSI_PARAM_PING_TMO:
880 case ISCSI_PARAM_RECV_TMO:
881 case ISCSI_PARAM_INITIAL_R2T_EN:
882 case ISCSI_PARAM_MAX_R2T:
883 case ISCSI_PARAM_IMM_DATA_EN:
884 case ISCSI_PARAM_FIRST_BURST:
885 case ISCSI_PARAM_MAX_BURST:
886 case ISCSI_PARAM_PDU_INORDER_EN:
887 case ISCSI_PARAM_DATASEQ_INORDER_EN:
888 case ISCSI_PARAM_TARGET_NAME:
889 case ISCSI_PARAM_TPGT:
890 case ISCSI_PARAM_USERNAME:
891 case ISCSI_PARAM_PASSWORD:
892 case ISCSI_PARAM_USERNAME_IN:
893 case ISCSI_PARAM_PASSWORD_IN:
894 case ISCSI_PARAM_FAST_ABORT:
895 case ISCSI_PARAM_ABORT_TMO:
896 case ISCSI_PARAM_LU_RESET_TMO:
897 case ISCSI_PARAM_TGT_RESET_TMO:
898 case ISCSI_PARAM_IFACE_NAME:
899 case ISCSI_PARAM_INITIATOR_NAME:
900 case ISCSI_PARAM_DISCOVERY_SESS:
901 return S_IRUGO;
902 default:
903 return 0;
904 }
905 }
906
907 return 0;
908 }
909
910 static struct scsi_host_template iscsi_iser_sht = {
911 .module = THIS_MODULE,
912 .name = "iSCSI Initiator over iSER",
913 .queuecommand = iscsi_queuecommand,
914 .change_queue_depth = iscsi_change_queue_depth,
915 .sg_tablesize = ISCSI_ISER_SG_TABLESIZE,
916 .max_sectors = 1024,
917 .cmd_per_lun = ISER_DEF_CMD_PER_LUN,
918 .eh_abort_handler = iscsi_eh_abort,
919 .eh_device_reset_handler= iscsi_eh_device_reset,
920 .eh_target_reset_handler = iscsi_eh_recover_target,
921 .target_alloc = iscsi_target_alloc,
922 .use_clustering = DISABLE_CLUSTERING,
923 .proc_name = "iscsi_iser",
924 .this_id = -1,
925 };
926
927 static struct iscsi_transport iscsi_iser_transport = {
928 .owner = THIS_MODULE,
929 .name = "iser",
930 .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_TEXT_NEGO,
931 /* session management */
932 .create_session = iscsi_iser_session_create,
933 .destroy_session = iscsi_iser_session_destroy,
934 /* connection management */
935 .create_conn = iscsi_iser_conn_create,
936 .bind_conn = iscsi_iser_conn_bind,
937 .destroy_conn = iscsi_conn_teardown,
938 .attr_is_visible = iser_attr_is_visible,
939 .set_param = iscsi_iser_set_param,
940 .get_conn_param = iscsi_conn_get_param,
941 .get_ep_param = iscsi_iser_get_ep_param,
942 .get_session_param = iscsi_session_get_param,
943 .start_conn = iscsi_iser_conn_start,
944 .stop_conn = iscsi_iser_conn_stop,
945 /* iscsi host params */
946 .get_host_param = iscsi_host_get_param,
947 .set_host_param = iscsi_host_set_param,
948 /* IO */
949 .send_pdu = iscsi_conn_send_pdu,
950 .get_stats = iscsi_iser_conn_get_stats,
951 .init_task = iscsi_iser_task_init,
952 .xmit_task = iscsi_iser_task_xmit,
953 .cleanup_task = iscsi_iser_cleanup_task,
954 .alloc_pdu = iscsi_iser_pdu_alloc,
955 .check_protection = iscsi_iser_check_protection,
956 /* recovery */
957 .session_recovery_timedout = iscsi_session_recovery_timedout,
958
959 .ep_connect = iscsi_iser_ep_connect,
960 .ep_poll = iscsi_iser_ep_poll,
961 .ep_disconnect = iscsi_iser_ep_disconnect
962 };
963
iser_init(void)964 static int __init iser_init(void)
965 {
966 int err;
967
968 iser_dbg("Starting iSER datamover...\n");
969
970 if (iscsi_max_lun < 1) {
971 iser_err("Invalid max_lun value of %u\n", iscsi_max_lun);
972 return -EINVAL;
973 }
974
975 memset(&ig, 0, sizeof(struct iser_global));
976
977 ig.desc_cache = kmem_cache_create("iser_descriptors",
978 sizeof(struct iser_tx_desc),
979 0, SLAB_HWCACHE_ALIGN,
980 NULL);
981 if (ig.desc_cache == NULL)
982 return -ENOMEM;
983
984 /* device init is called only after the first addr resolution */
985 mutex_init(&ig.device_list_mutex);
986 INIT_LIST_HEAD(&ig.device_list);
987 mutex_init(&ig.connlist_mutex);
988 INIT_LIST_HEAD(&ig.connlist);
989
990 release_wq = alloc_workqueue("release workqueue", 0, 0);
991 if (!release_wq) {
992 iser_err("failed to allocate release workqueue\n");
993 return -ENOMEM;
994 }
995
996 iscsi_iser_scsi_transport = iscsi_register_transport(
997 &iscsi_iser_transport);
998 if (!iscsi_iser_scsi_transport) {
999 iser_err("iscsi_register_transport failed\n");
1000 err = -EINVAL;
1001 goto register_transport_failure;
1002 }
1003
1004 return 0;
1005
1006 register_transport_failure:
1007 kmem_cache_destroy(ig.desc_cache);
1008
1009 return err;
1010 }
1011
iser_exit(void)1012 static void __exit iser_exit(void)
1013 {
1014 struct iser_conn *iser_conn, *n;
1015 int connlist_empty;
1016
1017 iser_dbg("Removing iSER datamover...\n");
1018 destroy_workqueue(release_wq);
1019
1020 mutex_lock(&ig.connlist_mutex);
1021 connlist_empty = list_empty(&ig.connlist);
1022 mutex_unlock(&ig.connlist_mutex);
1023
1024 if (!connlist_empty) {
1025 iser_err("Error cleanup stage completed but we still have iser "
1026 "connections, destroying them anyway.\n");
1027 list_for_each_entry_safe(iser_conn, n, &ig.connlist,
1028 conn_list) {
1029 iser_conn_release(iser_conn);
1030 }
1031 }
1032
1033 iscsi_unregister_transport(&iscsi_iser_transport);
1034 kmem_cache_destroy(ig.desc_cache);
1035 }
1036
1037 module_init(iser_init);
1038 module_exit(iser_exit);
1039