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 unsigned int iscsi_max_lun = 512;
93 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
94 MODULE_PARM_DESC(max_lun, "Max LUNs to allow per session (default:512");
95
96 unsigned int iser_max_sectors = ISER_DEF_MAX_SECTORS;
97 module_param_named(max_sectors, iser_max_sectors, uint, S_IRUGO | S_IWUSR);
98 MODULE_PARM_DESC(max_sectors, "Max number of sectors in a single scsi command (default:1024");
99
100 bool iser_always_reg = true;
101 module_param_named(always_register, iser_always_reg, bool, S_IRUGO);
102 MODULE_PARM_DESC(always_register,
103 "Always register memory, even for continuous memory regions (default:true)");
104
105 bool iser_pi_enable = false;
106 module_param_named(pi_enable, iser_pi_enable, bool, S_IRUGO);
107 MODULE_PARM_DESC(pi_enable, "Enable T10-PI offload support (default:disabled)");
108
109 int iser_pi_guard;
110 module_param_named(pi_guard, iser_pi_guard, int, S_IRUGO);
111 MODULE_PARM_DESC(pi_guard, "T10-PI guard_type [deprecated]");
112
113 /*
114 * iscsi_iser_recv() - Process a successful recv completion
115 * @conn: iscsi connection
116 * @hdr: iscsi header
117 * @rx_data: buffer containing receive data payload
118 * @rx_data_len: length of rx_data
119 *
120 * Notes: In case of data length errors or iscsi PDU completion failures
121 * this routine will signal iscsi layer of connection failure.
122 */
123 void
iscsi_iser_recv(struct iscsi_conn * conn,struct iscsi_hdr * hdr,char * rx_data,int rx_data_len)124 iscsi_iser_recv(struct iscsi_conn *conn, struct iscsi_hdr *hdr,
125 char *rx_data, int rx_data_len)
126 {
127 int rc = 0;
128 int datalen;
129
130 /* verify PDU length */
131 datalen = ntoh24(hdr->dlength);
132 if (datalen > rx_data_len || (datalen + 4) < rx_data_len) {
133 iser_err("wrong datalen %d (hdr), %d (IB)\n",
134 datalen, rx_data_len);
135 rc = ISCSI_ERR_DATALEN;
136 goto error;
137 }
138
139 if (datalen != rx_data_len)
140 iser_dbg("aligned datalen (%d) hdr, %d (IB)\n",
141 datalen, rx_data_len);
142
143 rc = iscsi_complete_pdu(conn, hdr, rx_data, rx_data_len);
144 if (rc && rc != ISCSI_ERR_NO_SCSI_CMD)
145 goto error;
146
147 return;
148 error:
149 iscsi_conn_failure(conn, rc);
150 }
151
152 /**
153 * iscsi_iser_pdu_alloc() - allocate an iscsi-iser PDU
154 * @task: iscsi task
155 * @opcode: iscsi command opcode
156 *
157 * Netes: This routine can't fail, just assign iscsi task
158 * hdr and max hdr size.
159 */
160 static int
iscsi_iser_pdu_alloc(struct iscsi_task * task,uint8_t opcode)161 iscsi_iser_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
162 {
163 struct iscsi_iser_task *iser_task = task->dd_data;
164
165 task->hdr = (struct iscsi_hdr *)&iser_task->desc.iscsi_header;
166 task->hdr_max = sizeof(iser_task->desc.iscsi_header);
167
168 return 0;
169 }
170
171 /**
172 * iser_initialize_task_headers() - Initialize task headers
173 * @task: iscsi task
174 * @tx_desc: iser tx descriptor
175 *
176 * Notes:
177 * This routine may race with iser teardown flow for scsi
178 * error handling TMFs. So for TMF we should acquire the
179 * state mutex to avoid dereferencing the IB device which
180 * may have already been terminated.
181 */
182 int
iser_initialize_task_headers(struct iscsi_task * task,struct iser_tx_desc * tx_desc)183 iser_initialize_task_headers(struct iscsi_task *task,
184 struct iser_tx_desc *tx_desc)
185 {
186 struct iser_conn *iser_conn = task->conn->dd_data;
187 struct iser_device *device = iser_conn->ib_conn.device;
188 struct iscsi_iser_task *iser_task = task->dd_data;
189 u64 dma_addr;
190 const bool mgmt_task = !task->sc && !in_interrupt();
191 int ret = 0;
192
193 if (unlikely(mgmt_task))
194 mutex_lock(&iser_conn->state_mutex);
195
196 if (unlikely(iser_conn->state != ISER_CONN_UP)) {
197 ret = -ENODEV;
198 goto out;
199 }
200
201 dma_addr = ib_dma_map_single(device->ib_device, (void *)tx_desc,
202 ISER_HEADERS_LEN, DMA_TO_DEVICE);
203 if (ib_dma_mapping_error(device->ib_device, dma_addr)) {
204 ret = -ENOMEM;
205 goto out;
206 }
207
208 tx_desc->inv_wr.next = NULL;
209 tx_desc->reg_wr.wr.next = NULL;
210 tx_desc->mapped = true;
211 tx_desc->dma_addr = dma_addr;
212 tx_desc->tx_sg[0].addr = tx_desc->dma_addr;
213 tx_desc->tx_sg[0].length = ISER_HEADERS_LEN;
214 tx_desc->tx_sg[0].lkey = device->pd->local_dma_lkey;
215
216 iser_task->iser_conn = iser_conn;
217 out:
218 if (unlikely(mgmt_task))
219 mutex_unlock(&iser_conn->state_mutex);
220
221 return ret;
222 }
223
224 /**
225 * iscsi_iser_task_init() - Initialize iscsi-iser task
226 * @task: iscsi task
227 *
228 * Initialize the task for the scsi command or mgmt command.
229 *
230 * Return: Returns zero on success or -ENOMEM when failing
231 * to init task headers (dma mapping error).
232 */
233 static int
iscsi_iser_task_init(struct iscsi_task * task)234 iscsi_iser_task_init(struct iscsi_task *task)
235 {
236 struct iscsi_iser_task *iser_task = task->dd_data;
237 int ret;
238
239 ret = iser_initialize_task_headers(task, &iser_task->desc);
240 if (ret) {
241 iser_err("Failed to init task %p, err = %d\n",
242 iser_task, ret);
243 return ret;
244 }
245
246 /* mgmt task */
247 if (!task->sc)
248 return 0;
249
250 iser_task->command_sent = 0;
251 iser_task_rdma_init(iser_task);
252 iser_task->sc = task->sc;
253
254 return 0;
255 }
256
257 /**
258 * iscsi_iser_mtask_xmit() - xmit management (immediate) task
259 * @conn: iscsi connection
260 * @task: task management task
261 *
262 * Notes:
263 * The function can return -EAGAIN in which case caller must
264 * call it again later, or recover. '0' return code means successful
265 * xmit.
266 *
267 **/
268 static int
iscsi_iser_mtask_xmit(struct iscsi_conn * conn,struct iscsi_task * task)269 iscsi_iser_mtask_xmit(struct iscsi_conn *conn, struct iscsi_task *task)
270 {
271 int error = 0;
272
273 iser_dbg("mtask xmit [cid %d itt 0x%x]\n", conn->id, task->itt);
274
275 error = iser_send_control(conn, task);
276
277 /* since iser xmits control with zero copy, tasks can not be recycled
278 * right after sending them.
279 * The recycling scheme is based on whether a response is expected
280 * - if yes, the task is recycled at iscsi_complete_pdu
281 * - if no, the task is recycled at iser_snd_completion
282 */
283 return error;
284 }
285
286 static int
iscsi_iser_task_xmit_unsol_data(struct iscsi_conn * conn,struct iscsi_task * task)287 iscsi_iser_task_xmit_unsol_data(struct iscsi_conn *conn,
288 struct iscsi_task *task)
289 {
290 struct iscsi_r2t_info *r2t = &task->unsol_r2t;
291 struct iscsi_data hdr;
292 int error = 0;
293
294 /* Send data-out PDUs while there's still unsolicited data to send */
295 while (iscsi_task_has_unsol_data(task)) {
296 iscsi_prep_data_out_pdu(task, r2t, &hdr);
297 iser_dbg("Sending data-out: itt 0x%x, data count %d\n",
298 hdr.itt, r2t->data_count);
299
300 /* the buffer description has been passed with the command */
301 /* Send the command */
302 error = iser_send_data_out(conn, task, &hdr);
303 if (error) {
304 r2t->datasn--;
305 goto iscsi_iser_task_xmit_unsol_data_exit;
306 }
307 r2t->sent += r2t->data_count;
308 iser_dbg("Need to send %d more as data-out PDUs\n",
309 r2t->data_length - r2t->sent);
310 }
311
312 iscsi_iser_task_xmit_unsol_data_exit:
313 return error;
314 }
315
316 /**
317 * iscsi_iser_task_xmit() - xmit iscsi-iser task
318 * @task: iscsi task
319 *
320 * Return: zero on success or escalates $error on failure.
321 */
322 static int
iscsi_iser_task_xmit(struct iscsi_task * task)323 iscsi_iser_task_xmit(struct iscsi_task *task)
324 {
325 struct iscsi_conn *conn = task->conn;
326 struct iscsi_iser_task *iser_task = task->dd_data;
327 int error = 0;
328
329 if (!task->sc)
330 return iscsi_iser_mtask_xmit(conn, task);
331
332 if (task->sc->sc_data_direction == DMA_TO_DEVICE) {
333 BUG_ON(scsi_bufflen(task->sc) == 0);
334
335 iser_dbg("cmd [itt %x total %d imm %d unsol_data %d\n",
336 task->itt, scsi_bufflen(task->sc),
337 task->imm_count, task->unsol_r2t.data_length);
338 }
339
340 iser_dbg("ctask xmit [cid %d itt 0x%x]\n",
341 conn->id, task->itt);
342
343 /* Send the cmd PDU */
344 if (!iser_task->command_sent) {
345 error = iser_send_command(conn, task);
346 if (error)
347 goto iscsi_iser_task_xmit_exit;
348 iser_task->command_sent = 1;
349 }
350
351 /* Send unsolicited data-out PDU(s) if necessary */
352 if (iscsi_task_has_unsol_data(task))
353 error = iscsi_iser_task_xmit_unsol_data(conn, task);
354
355 iscsi_iser_task_xmit_exit:
356 return error;
357 }
358
359 /**
360 * iscsi_iser_cleanup_task() - cleanup an iscsi-iser task
361 * @task: iscsi task
362 *
363 * Notes: In case the RDMA device is already NULL (might have
364 * been removed in DEVICE_REMOVAL CM event it will bail-out
365 * without doing dma unmapping.
366 */
iscsi_iser_cleanup_task(struct iscsi_task * task)367 static void iscsi_iser_cleanup_task(struct iscsi_task *task)
368 {
369 struct iscsi_iser_task *iser_task = task->dd_data;
370 struct iser_tx_desc *tx_desc = &iser_task->desc;
371 struct iser_conn *iser_conn = task->conn->dd_data;
372 struct iser_device *device = iser_conn->ib_conn.device;
373
374 /* DEVICE_REMOVAL event might have already released the device */
375 if (!device)
376 return;
377
378 if (likely(tx_desc->mapped)) {
379 ib_dma_unmap_single(device->ib_device, tx_desc->dma_addr,
380 ISER_HEADERS_LEN, DMA_TO_DEVICE);
381 tx_desc->mapped = false;
382 }
383
384 /* mgmt tasks do not need special cleanup */
385 if (!task->sc)
386 return;
387
388 if (iser_task->status == ISER_TASK_STATUS_STARTED) {
389 iser_task->status = ISER_TASK_STATUS_COMPLETED;
390 iser_task_rdma_finalize(iser_task);
391 }
392 }
393
394 /**
395 * iscsi_iser_check_protection() - check protection information status of task.
396 * @task: iscsi task
397 * @sector: error sector if exsists (output)
398 *
399 * Return: zero if no data-integrity errors have occured
400 * 0x1: data-integrity error occured in the guard-block
401 * 0x2: data-integrity error occured in the reference tag
402 * 0x3: data-integrity error occured in the application tag
403 *
404 * In addition the error sector is marked.
405 */
406 static u8
iscsi_iser_check_protection(struct iscsi_task * task,sector_t * sector)407 iscsi_iser_check_protection(struct iscsi_task *task, sector_t *sector)
408 {
409 struct iscsi_iser_task *iser_task = task->dd_data;
410 enum iser_data_dir dir = iser_task->dir[ISER_DIR_IN] ?
411 ISER_DIR_IN : ISER_DIR_OUT;
412
413 return iser_check_task_pi_status(iser_task, dir, sector);
414 }
415
416 /**
417 * iscsi_iser_conn_create() - create a new iscsi-iser connection
418 * @cls_session: iscsi class connection
419 * @conn_idx: connection index within the session (for MCS)
420 *
421 * Return: iscsi_cls_conn when iscsi_conn_setup succeeds or NULL
422 * otherwise.
423 */
424 static struct iscsi_cls_conn *
iscsi_iser_conn_create(struct iscsi_cls_session * cls_session,uint32_t conn_idx)425 iscsi_iser_conn_create(struct iscsi_cls_session *cls_session,
426 uint32_t conn_idx)
427 {
428 struct iscsi_conn *conn;
429 struct iscsi_cls_conn *cls_conn;
430
431 cls_conn = iscsi_conn_setup(cls_session, 0, conn_idx);
432 if (!cls_conn)
433 return NULL;
434 conn = cls_conn->dd_data;
435
436 /*
437 * due to issues with the login code re iser sematics
438 * this not set in iscsi_conn_setup - FIXME
439 */
440 conn->max_recv_dlength = ISER_RECV_DATA_SEG_LEN;
441
442 return cls_conn;
443 }
444
445 /**
446 * iscsi_iser_conn_bind() - bind iscsi and iser connection structures
447 * @cls_session: iscsi class session
448 * @cls_conn: iscsi class connection
449 * @transport_eph: transport end-point handle
450 * @is_leading: indicate if this is the session leading connection (MCS)
451 *
452 * Return: zero on success, $error if iscsi_conn_bind fails and
453 * -EINVAL in case end-point doesn't exsits anymore or iser connection
454 * state is not UP (teardown already started).
455 */
456 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)457 iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session,
458 struct iscsi_cls_conn *cls_conn,
459 uint64_t transport_eph,
460 int is_leading)
461 {
462 struct iscsi_conn *conn = cls_conn->dd_data;
463 struct iser_conn *iser_conn;
464 struct iscsi_endpoint *ep;
465 int error;
466
467 error = iscsi_conn_bind(cls_session, cls_conn, is_leading);
468 if (error)
469 return error;
470
471 /* the transport ep handle comes from user space so it must be
472 * verified against the global ib connections list */
473 ep = iscsi_lookup_endpoint(transport_eph);
474 if (!ep) {
475 iser_err("can't bind eph %llx\n",
476 (unsigned long long)transport_eph);
477 return -EINVAL;
478 }
479 iser_conn = ep->dd_data;
480
481 mutex_lock(&iser_conn->state_mutex);
482 if (iser_conn->state != ISER_CONN_UP) {
483 error = -EINVAL;
484 iser_err("iser_conn %p state is %d, teardown started\n",
485 iser_conn, iser_conn->state);
486 goto out;
487 }
488
489 error = iser_alloc_rx_descriptors(iser_conn, conn->session);
490 if (error)
491 goto out;
492
493 /* binds the iSER connection retrieved from the previously
494 * connected ep_handle to the iSCSI layer connection. exchanges
495 * connection pointers */
496 iser_info("binding iscsi conn %p to iser_conn %p\n", conn, iser_conn);
497
498 conn->dd_data = iser_conn;
499 iser_conn->iscsi_conn = conn;
500
501 out:
502 iscsi_put_endpoint(ep);
503 mutex_unlock(&iser_conn->state_mutex);
504 return error;
505 }
506
507 /**
508 * iscsi_iser_conn_start() - start iscsi-iser connection
509 * @cls_conn: iscsi class connection
510 *
511 * Notes: Here iser intialize (or re-initialize) stop_completion as
512 * from this point iscsi must call conn_stop in session/connection
513 * teardown so iser transport must wait for it.
514 */
515 static int
iscsi_iser_conn_start(struct iscsi_cls_conn * cls_conn)516 iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn)
517 {
518 struct iscsi_conn *iscsi_conn;
519 struct iser_conn *iser_conn;
520
521 iscsi_conn = cls_conn->dd_data;
522 iser_conn = iscsi_conn->dd_data;
523 reinit_completion(&iser_conn->stop_completion);
524
525 return iscsi_conn_start(cls_conn);
526 }
527
528 /**
529 * iscsi_iser_conn_stop() - stop iscsi-iser connection
530 * @cls_conn: iscsi class connection
531 * @flag: indicate if recover or terminate (passed as is)
532 *
533 * Notes: Calling iscsi_conn_stop might theoretically race with
534 * DEVICE_REMOVAL event and dereference a previously freed RDMA device
535 * handle, so we call it under iser the state lock to protect against
536 * this kind of race.
537 */
538 static void
iscsi_iser_conn_stop(struct iscsi_cls_conn * cls_conn,int flag)539 iscsi_iser_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
540 {
541 struct iscsi_conn *conn = cls_conn->dd_data;
542 struct iser_conn *iser_conn = conn->dd_data;
543
544 iser_info("stopping iscsi_conn: %p, iser_conn: %p\n", conn, iser_conn);
545
546 /*
547 * Userspace may have goofed up and not bound the connection or
548 * might have only partially setup the connection.
549 */
550 if (iser_conn) {
551 mutex_lock(&iser_conn->state_mutex);
552 mutex_lock(&unbind_iser_conn_mutex);
553 iser_conn_terminate(iser_conn);
554 iscsi_conn_stop(cls_conn, flag);
555
556 /* unbind */
557 iser_conn->iscsi_conn = NULL;
558 conn->dd_data = NULL;
559 mutex_unlock(&unbind_iser_conn_mutex);
560
561 complete(&iser_conn->stop_completion);
562 mutex_unlock(&iser_conn->state_mutex);
563 } else {
564 iscsi_conn_stop(cls_conn, flag);
565 }
566 }
567
568 /**
569 * iscsi_iser_session_destroy() - destroy iscsi-iser session
570 * @cls_session: iscsi class session
571 *
572 * Removes and free iscsi host.
573 */
574 static void
iscsi_iser_session_destroy(struct iscsi_cls_session * cls_session)575 iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session)
576 {
577 struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
578
579 iscsi_session_teardown(cls_session);
580 iscsi_host_remove(shost);
581 iscsi_host_free(shost);
582 }
583
584 static inline unsigned int
iser_dif_prot_caps(int prot_caps)585 iser_dif_prot_caps(int prot_caps)
586 {
587 return ((prot_caps & IB_PROT_T10DIF_TYPE_1) ?
588 SHOST_DIF_TYPE1_PROTECTION | SHOST_DIX_TYPE0_PROTECTION |
589 SHOST_DIX_TYPE1_PROTECTION : 0) |
590 ((prot_caps & IB_PROT_T10DIF_TYPE_2) ?
591 SHOST_DIF_TYPE2_PROTECTION | SHOST_DIX_TYPE2_PROTECTION : 0) |
592 ((prot_caps & IB_PROT_T10DIF_TYPE_3) ?
593 SHOST_DIF_TYPE3_PROTECTION | SHOST_DIX_TYPE3_PROTECTION : 0);
594 }
595
596 /**
597 * iscsi_iser_session_create() - create an iscsi-iser session
598 * @ep: iscsi end-point handle
599 * @cmds_max: maximum commands in this session
600 * @qdepth: session command queue depth
601 * @initial_cmdsn: initiator command sequnce number
602 *
603 * Allocates and adds a scsi host, expose DIF supprot if
604 * exists, and sets up an iscsi session.
605 */
606 static struct iscsi_cls_session *
iscsi_iser_session_create(struct iscsi_endpoint * ep,uint16_t cmds_max,uint16_t qdepth,uint32_t initial_cmdsn)607 iscsi_iser_session_create(struct iscsi_endpoint *ep,
608 uint16_t cmds_max, uint16_t qdepth,
609 uint32_t initial_cmdsn)
610 {
611 struct iscsi_cls_session *cls_session;
612 struct Scsi_Host *shost;
613 struct iser_conn *iser_conn = NULL;
614 struct ib_conn *ib_conn;
615 struct ib_device *ib_dev;
616 u32 max_fr_sectors;
617
618 shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 0);
619 if (!shost)
620 return NULL;
621 shost->transportt = iscsi_iser_scsi_transport;
622 shost->cmd_per_lun = qdepth;
623 shost->max_lun = iscsi_max_lun;
624 shost->max_id = 0;
625 shost->max_channel = 0;
626 shost->max_cmd_len = 16;
627
628 /*
629 * older userspace tools (before 2.0-870) did not pass us
630 * the leading conn's ep so this will be NULL;
631 */
632 if (ep) {
633 iser_conn = ep->dd_data;
634 shost->sg_tablesize = iser_conn->scsi_sg_tablesize;
635 shost->can_queue = min_t(u16, cmds_max, iser_conn->max_cmds);
636
637 mutex_lock(&iser_conn->state_mutex);
638 if (iser_conn->state != ISER_CONN_UP) {
639 iser_err("iser conn %p already started teardown\n",
640 iser_conn);
641 mutex_unlock(&iser_conn->state_mutex);
642 goto free_host;
643 }
644
645 ib_conn = &iser_conn->ib_conn;
646 ib_dev = ib_conn->device->ib_device;
647 if (ib_conn->pi_support) {
648 u32 sig_caps = ib_dev->attrs.sig_prot_cap;
649
650 shost->sg_prot_tablesize = shost->sg_tablesize;
651 scsi_host_set_prot(shost, iser_dif_prot_caps(sig_caps));
652 scsi_host_set_guard(shost, SHOST_DIX_GUARD_IP |
653 SHOST_DIX_GUARD_CRC);
654 }
655
656 if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_SG_GAPS_REG))
657 shost->virt_boundary_mask = SZ_4K - 1;
658
659 if (iscsi_host_add(shost, ib_dev->dev.parent)) {
660 mutex_unlock(&iser_conn->state_mutex);
661 goto free_host;
662 }
663 mutex_unlock(&iser_conn->state_mutex);
664 } else {
665 shost->can_queue = min_t(u16, cmds_max, ISER_DEF_XMIT_CMDS_MAX);
666 if (iscsi_host_add(shost, NULL))
667 goto free_host;
668 }
669
670 max_fr_sectors = (shost->sg_tablesize * PAGE_SIZE) >> 9;
671 shost->max_sectors = min(iser_max_sectors, max_fr_sectors);
672
673 iser_dbg("iser_conn %p, sg_tablesize %u, max_sectors %u\n",
674 iser_conn, shost->sg_tablesize,
675 shost->max_sectors);
676
677 if (shost->max_sectors < iser_max_sectors)
678 iser_warn("max_sectors was reduced from %u to %u\n",
679 iser_max_sectors, shost->max_sectors);
680
681 cls_session = iscsi_session_setup(&iscsi_iser_transport, shost,
682 shost->can_queue, 0,
683 sizeof(struct iscsi_iser_task),
684 initial_cmdsn, 0);
685 if (!cls_session)
686 goto remove_host;
687
688 return cls_session;
689
690 remove_host:
691 iscsi_host_remove(shost);
692 free_host:
693 iscsi_host_free(shost);
694 return NULL;
695 }
696
697 static int
iscsi_iser_set_param(struct iscsi_cls_conn * cls_conn,enum iscsi_param param,char * buf,int buflen)698 iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn,
699 enum iscsi_param param, char *buf, int buflen)
700 {
701 int value;
702
703 switch (param) {
704 case ISCSI_PARAM_MAX_RECV_DLENGTH:
705 /* TBD */
706 break;
707 case ISCSI_PARAM_HDRDGST_EN:
708 sscanf(buf, "%d", &value);
709 if (value) {
710 iser_err("DataDigest wasn't negotiated to None\n");
711 return -EPROTO;
712 }
713 break;
714 case ISCSI_PARAM_DATADGST_EN:
715 sscanf(buf, "%d", &value);
716 if (value) {
717 iser_err("DataDigest wasn't negotiated to None\n");
718 return -EPROTO;
719 }
720 break;
721 case ISCSI_PARAM_IFMARKER_EN:
722 sscanf(buf, "%d", &value);
723 if (value) {
724 iser_err("IFMarker wasn't negotiated to No\n");
725 return -EPROTO;
726 }
727 break;
728 case ISCSI_PARAM_OFMARKER_EN:
729 sscanf(buf, "%d", &value);
730 if (value) {
731 iser_err("OFMarker wasn't negotiated to No\n");
732 return -EPROTO;
733 }
734 break;
735 default:
736 return iscsi_set_param(cls_conn, param, buf, buflen);
737 }
738
739 return 0;
740 }
741
742 /**
743 * iscsi_iser_set_param() - set class connection parameter
744 * @cls_conn: iscsi class connection
745 * @stats: iscsi stats to output
746 *
747 * Output connection statistics.
748 */
749 static void
iscsi_iser_conn_get_stats(struct iscsi_cls_conn * cls_conn,struct iscsi_stats * stats)750 iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
751 {
752 struct iscsi_conn *conn = cls_conn->dd_data;
753
754 stats->txdata_octets = conn->txdata_octets;
755 stats->rxdata_octets = conn->rxdata_octets;
756 stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
757 stats->dataout_pdus = conn->dataout_pdus_cnt;
758 stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
759 stats->datain_pdus = conn->datain_pdus_cnt; /* always 0 */
760 stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */
761 stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
762 stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
763 stats->custom_length = 0;
764 }
765
iscsi_iser_get_ep_param(struct iscsi_endpoint * ep,enum iscsi_param param,char * buf)766 static int iscsi_iser_get_ep_param(struct iscsi_endpoint *ep,
767 enum iscsi_param param, char *buf)
768 {
769 struct iser_conn *iser_conn = ep->dd_data;
770
771 switch (param) {
772 case ISCSI_PARAM_CONN_PORT:
773 case ISCSI_PARAM_CONN_ADDRESS:
774 if (!iser_conn || !iser_conn->ib_conn.cma_id)
775 return -ENOTCONN;
776
777 return iscsi_conn_get_addr_param((struct sockaddr_storage *)
778 &iser_conn->ib_conn.cma_id->route.addr.dst_addr,
779 param, buf);
780 default:
781 break;
782 }
783 return -ENOSYS;
784 }
785
786 /**
787 * iscsi_iser_ep_connect() - Initiate iSER connection establishment
788 * @shost: scsi_host
789 * @dst_addr: destination address
790 * @non_blocking: indicate if routine can block
791 *
792 * Allocate an iscsi endpoint, an iser_conn structure and bind them.
793 * After that start RDMA connection establishment via rdma_cm. We
794 * don't allocate iser_conn embedded in iscsi_endpoint since in teardown
795 * the endpoint will be destroyed at ep_disconnect while iser_conn will
796 * cleanup its resources asynchronuously.
797 *
798 * Return: iscsi_endpoint created by iscsi layer or ERR_PTR(error)
799 * if fails.
800 */
801 static struct iscsi_endpoint *
iscsi_iser_ep_connect(struct Scsi_Host * shost,struct sockaddr * dst_addr,int non_blocking)802 iscsi_iser_ep_connect(struct Scsi_Host *shost, struct sockaddr *dst_addr,
803 int non_blocking)
804 {
805 int err;
806 struct iser_conn *iser_conn;
807 struct iscsi_endpoint *ep;
808
809 ep = iscsi_create_endpoint(0);
810 if (!ep)
811 return ERR_PTR(-ENOMEM);
812
813 iser_conn = kzalloc(sizeof(*iser_conn), GFP_KERNEL);
814 if (!iser_conn) {
815 err = -ENOMEM;
816 goto failure;
817 }
818
819 ep->dd_data = iser_conn;
820 iser_conn->ep = ep;
821 iser_conn_init(iser_conn);
822
823 err = iser_connect(iser_conn, NULL, dst_addr, non_blocking);
824 if (err)
825 goto failure;
826
827 return ep;
828 failure:
829 iscsi_destroy_endpoint(ep);
830 return ERR_PTR(err);
831 }
832
833 /**
834 * iscsi_iser_ep_poll() - poll for iser connection establishment to complete
835 * @ep: iscsi endpoint (created at ep_connect)
836 * @timeout_ms: polling timeout allowed in ms.
837 *
838 * This routine boils down to waiting for up_completion signaling
839 * that cma_id got CONNECTED event.
840 *
841 * Return: 1 if succeeded in connection establishment, 0 if timeout expired
842 * (libiscsi will retry will kick in) or -1 if interrupted by signal
843 * or more likely iser connection state transitioned to TEMINATING or
844 * DOWN during the wait period.
845 */
846 static int
iscsi_iser_ep_poll(struct iscsi_endpoint * ep,int timeout_ms)847 iscsi_iser_ep_poll(struct iscsi_endpoint *ep, int timeout_ms)
848 {
849 struct iser_conn *iser_conn = ep->dd_data;
850 int rc;
851
852 rc = wait_for_completion_interruptible_timeout(&iser_conn->up_completion,
853 msecs_to_jiffies(timeout_ms));
854 /* if conn establishment failed, return error code to iscsi */
855 if (rc == 0) {
856 mutex_lock(&iser_conn->state_mutex);
857 if (iser_conn->state == ISER_CONN_TERMINATING ||
858 iser_conn->state == ISER_CONN_DOWN)
859 rc = -1;
860 mutex_unlock(&iser_conn->state_mutex);
861 }
862
863 iser_info("iser conn %p rc = %d\n", iser_conn, rc);
864
865 if (rc > 0)
866 return 1; /* success, this is the equivalent of EPOLLOUT */
867 else if (!rc)
868 return 0; /* timeout */
869 else
870 return rc; /* signal */
871 }
872
873 /**
874 * iscsi_iser_ep_disconnect() - Initiate connection teardown process
875 * @ep: iscsi endpoint handle
876 *
877 * This routine is not blocked by iser and RDMA termination process
878 * completion as we queue a deffered work for iser/RDMA destruction
879 * and cleanup or actually call it immediately in case we didn't pass
880 * iscsi conn bind/start stage, thus it is safe.
881 */
882 static void
iscsi_iser_ep_disconnect(struct iscsi_endpoint * ep)883 iscsi_iser_ep_disconnect(struct iscsi_endpoint *ep)
884 {
885 struct iser_conn *iser_conn = ep->dd_data;
886
887 iser_info("ep %p iser conn %p\n", ep, iser_conn);
888
889 mutex_lock(&iser_conn->state_mutex);
890 iser_conn_terminate(iser_conn);
891
892 /*
893 * if iser_conn and iscsi_conn are bound, we must wait for
894 * iscsi_conn_stop and flush errors completion before freeing
895 * the iser resources. Otherwise we are safe to free resources
896 * immediately.
897 */
898 if (iser_conn->iscsi_conn) {
899 INIT_WORK(&iser_conn->release_work, iser_release_work);
900 queue_work(release_wq, &iser_conn->release_work);
901 mutex_unlock(&iser_conn->state_mutex);
902 } else {
903 iser_conn->state = ISER_CONN_DOWN;
904 mutex_unlock(&iser_conn->state_mutex);
905 iser_conn_release(iser_conn);
906 }
907
908 iscsi_destroy_endpoint(ep);
909 }
910
iser_attr_is_visible(int param_type,int param)911 static umode_t iser_attr_is_visible(int param_type, int param)
912 {
913 switch (param_type) {
914 case ISCSI_HOST_PARAM:
915 switch (param) {
916 case ISCSI_HOST_PARAM_NETDEV_NAME:
917 case ISCSI_HOST_PARAM_HWADDRESS:
918 case ISCSI_HOST_PARAM_INITIATOR_NAME:
919 return S_IRUGO;
920 default:
921 return 0;
922 }
923 case ISCSI_PARAM:
924 switch (param) {
925 case ISCSI_PARAM_MAX_RECV_DLENGTH:
926 case ISCSI_PARAM_MAX_XMIT_DLENGTH:
927 case ISCSI_PARAM_HDRDGST_EN:
928 case ISCSI_PARAM_DATADGST_EN:
929 case ISCSI_PARAM_CONN_ADDRESS:
930 case ISCSI_PARAM_CONN_PORT:
931 case ISCSI_PARAM_EXP_STATSN:
932 case ISCSI_PARAM_PERSISTENT_ADDRESS:
933 case ISCSI_PARAM_PERSISTENT_PORT:
934 case ISCSI_PARAM_PING_TMO:
935 case ISCSI_PARAM_RECV_TMO:
936 case ISCSI_PARAM_INITIAL_R2T_EN:
937 case ISCSI_PARAM_MAX_R2T:
938 case ISCSI_PARAM_IMM_DATA_EN:
939 case ISCSI_PARAM_FIRST_BURST:
940 case ISCSI_PARAM_MAX_BURST:
941 case ISCSI_PARAM_PDU_INORDER_EN:
942 case ISCSI_PARAM_DATASEQ_INORDER_EN:
943 case ISCSI_PARAM_TARGET_NAME:
944 case ISCSI_PARAM_TPGT:
945 case ISCSI_PARAM_USERNAME:
946 case ISCSI_PARAM_PASSWORD:
947 case ISCSI_PARAM_USERNAME_IN:
948 case ISCSI_PARAM_PASSWORD_IN:
949 case ISCSI_PARAM_FAST_ABORT:
950 case ISCSI_PARAM_ABORT_TMO:
951 case ISCSI_PARAM_LU_RESET_TMO:
952 case ISCSI_PARAM_TGT_RESET_TMO:
953 case ISCSI_PARAM_IFACE_NAME:
954 case ISCSI_PARAM_INITIATOR_NAME:
955 case ISCSI_PARAM_DISCOVERY_SESS:
956 return S_IRUGO;
957 default:
958 return 0;
959 }
960 }
961
962 return 0;
963 }
964
965 static struct scsi_host_template iscsi_iser_sht = {
966 .module = THIS_MODULE,
967 .name = "iSCSI Initiator over iSER",
968 .queuecommand = iscsi_queuecommand,
969 .change_queue_depth = scsi_change_queue_depth,
970 .sg_tablesize = ISCSI_ISER_DEF_SG_TABLESIZE,
971 .cmd_per_lun = ISER_DEF_CMD_PER_LUN,
972 .eh_timed_out = iscsi_eh_cmd_timed_out,
973 .eh_abort_handler = iscsi_eh_abort,
974 .eh_device_reset_handler= iscsi_eh_device_reset,
975 .eh_target_reset_handler = iscsi_eh_recover_target,
976 .target_alloc = iscsi_target_alloc,
977 .proc_name = "iscsi_iser",
978 .this_id = -1,
979 .track_queue_depth = 1,
980 };
981
982 static struct iscsi_transport iscsi_iser_transport = {
983 .owner = THIS_MODULE,
984 .name = "iser",
985 .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_TEXT_NEGO,
986 /* session management */
987 .create_session = iscsi_iser_session_create,
988 .destroy_session = iscsi_iser_session_destroy,
989 /* connection management */
990 .create_conn = iscsi_iser_conn_create,
991 .bind_conn = iscsi_iser_conn_bind,
992 .unbind_conn = iscsi_conn_unbind,
993 .destroy_conn = iscsi_conn_teardown,
994 .attr_is_visible = iser_attr_is_visible,
995 .set_param = iscsi_iser_set_param,
996 .get_conn_param = iscsi_conn_get_param,
997 .get_ep_param = iscsi_iser_get_ep_param,
998 .get_session_param = iscsi_session_get_param,
999 .start_conn = iscsi_iser_conn_start,
1000 .stop_conn = iscsi_iser_conn_stop,
1001 /* iscsi host params */
1002 .get_host_param = iscsi_host_get_param,
1003 .set_host_param = iscsi_host_set_param,
1004 /* IO */
1005 .send_pdu = iscsi_conn_send_pdu,
1006 .get_stats = iscsi_iser_conn_get_stats,
1007 .init_task = iscsi_iser_task_init,
1008 .xmit_task = iscsi_iser_task_xmit,
1009 .cleanup_task = iscsi_iser_cleanup_task,
1010 .alloc_pdu = iscsi_iser_pdu_alloc,
1011 .check_protection = iscsi_iser_check_protection,
1012 /* recovery */
1013 .session_recovery_timedout = iscsi_session_recovery_timedout,
1014
1015 .ep_connect = iscsi_iser_ep_connect,
1016 .ep_poll = iscsi_iser_ep_poll,
1017 .ep_disconnect = iscsi_iser_ep_disconnect
1018 };
1019
iser_init(void)1020 static int __init iser_init(void)
1021 {
1022 int err;
1023
1024 iser_dbg("Starting iSER datamover...\n");
1025
1026 if (iscsi_max_lun < 1) {
1027 iser_err("Invalid max_lun value of %u\n", iscsi_max_lun);
1028 return -EINVAL;
1029 }
1030
1031 memset(&ig, 0, sizeof(struct iser_global));
1032
1033 ig.desc_cache = kmem_cache_create("iser_descriptors",
1034 sizeof(struct iser_tx_desc),
1035 0, SLAB_HWCACHE_ALIGN,
1036 NULL);
1037 if (ig.desc_cache == NULL)
1038 return -ENOMEM;
1039
1040 /* device init is called only after the first addr resolution */
1041 mutex_init(&ig.device_list_mutex);
1042 INIT_LIST_HEAD(&ig.device_list);
1043 mutex_init(&ig.connlist_mutex);
1044 INIT_LIST_HEAD(&ig.connlist);
1045
1046 release_wq = alloc_workqueue("release workqueue", 0, 0);
1047 if (!release_wq) {
1048 iser_err("failed to allocate release workqueue\n");
1049 err = -ENOMEM;
1050 goto err_alloc_wq;
1051 }
1052
1053 iscsi_iser_scsi_transport = iscsi_register_transport(
1054 &iscsi_iser_transport);
1055 if (!iscsi_iser_scsi_transport) {
1056 iser_err("iscsi_register_transport failed\n");
1057 err = -EINVAL;
1058 goto err_reg;
1059 }
1060
1061 return 0;
1062
1063 err_reg:
1064 destroy_workqueue(release_wq);
1065 err_alloc_wq:
1066 kmem_cache_destroy(ig.desc_cache);
1067
1068 return err;
1069 }
1070
iser_exit(void)1071 static void __exit iser_exit(void)
1072 {
1073 struct iser_conn *iser_conn, *n;
1074 int connlist_empty;
1075
1076 iser_dbg("Removing iSER datamover...\n");
1077 destroy_workqueue(release_wq);
1078
1079 mutex_lock(&ig.connlist_mutex);
1080 connlist_empty = list_empty(&ig.connlist);
1081 mutex_unlock(&ig.connlist_mutex);
1082
1083 if (!connlist_empty) {
1084 iser_err("Error cleanup stage completed but we still have iser "
1085 "connections, destroying them anyway\n");
1086 list_for_each_entry_safe(iser_conn, n, &ig.connlist,
1087 conn_list) {
1088 iser_conn_release(iser_conn);
1089 }
1090 }
1091
1092 iscsi_unregister_transport(&iscsi_iser_transport);
1093 kmem_cache_destroy(ig.desc_cache);
1094 }
1095
1096 module_init(iser_init);
1097 module_exit(iser_exit);
1098