Loading...
1/*
2 * Generic SCSI-3 ALUA SCSI Device Handler
3 *
4 * Copyright (C) 2007-2010 Hannes Reinecke, SUSE Linux Products GmbH.
5 * All rights reserved.
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
20 *
21 */
22#include <linux/slab.h>
23#include <linux/delay.h>
24#include <scsi/scsi.h>
25#include <scsi/scsi_eh.h>
26#include <scsi/scsi_dh.h>
27
28#define ALUA_DH_NAME "alua"
29#define ALUA_DH_VER "1.3"
30
31#define TPGS_STATE_OPTIMIZED 0x0
32#define TPGS_STATE_NONOPTIMIZED 0x1
33#define TPGS_STATE_STANDBY 0x2
34#define TPGS_STATE_UNAVAILABLE 0x3
35#define TPGS_STATE_LBA_DEPENDENT 0x4
36#define TPGS_STATE_OFFLINE 0xe
37#define TPGS_STATE_TRANSITIONING 0xf
38
39#define TPGS_SUPPORT_NONE 0x00
40#define TPGS_SUPPORT_OPTIMIZED 0x01
41#define TPGS_SUPPORT_NONOPTIMIZED 0x02
42#define TPGS_SUPPORT_STANDBY 0x04
43#define TPGS_SUPPORT_UNAVAILABLE 0x08
44#define TPGS_SUPPORT_LBA_DEPENDENT 0x10
45#define TPGS_SUPPORT_OFFLINE 0x40
46#define TPGS_SUPPORT_TRANSITION 0x80
47
48#define TPGS_MODE_UNINITIALIZED -1
49#define TPGS_MODE_NONE 0x0
50#define TPGS_MODE_IMPLICIT 0x1
51#define TPGS_MODE_EXPLICIT 0x2
52
53#define ALUA_INQUIRY_SIZE 36
54#define ALUA_FAILOVER_TIMEOUT (60 * HZ)
55#define ALUA_FAILOVER_RETRIES 5
56
57struct alua_dh_data {
58 int group_id;
59 int rel_port;
60 int tpgs;
61 int state;
62 unsigned char inq[ALUA_INQUIRY_SIZE];
63 unsigned char *buff;
64 int bufflen;
65 unsigned char sense[SCSI_SENSE_BUFFERSIZE];
66 int senselen;
67 struct scsi_device *sdev;
68 activate_complete callback_fn;
69 void *callback_data;
70};
71
72#define ALUA_POLICY_SWITCH_CURRENT 0
73#define ALUA_POLICY_SWITCH_ALL 1
74
75static char print_alua_state(int);
76static int alua_check_sense(struct scsi_device *, struct scsi_sense_hdr *);
77
78static inline struct alua_dh_data *get_alua_data(struct scsi_device *sdev)
79{
80 struct scsi_dh_data *scsi_dh_data = sdev->scsi_dh_data;
81 BUG_ON(scsi_dh_data == NULL);
82 return ((struct alua_dh_data *) scsi_dh_data->buf);
83}
84
85static int realloc_buffer(struct alua_dh_data *h, unsigned len)
86{
87 if (h->buff && h->buff != h->inq)
88 kfree(h->buff);
89
90 h->buff = kmalloc(len, GFP_NOIO);
91 if (!h->buff) {
92 h->buff = h->inq;
93 h->bufflen = ALUA_INQUIRY_SIZE;
94 return 1;
95 }
96 h->bufflen = len;
97 return 0;
98}
99
100static struct request *get_alua_req(struct scsi_device *sdev,
101 void *buffer, unsigned buflen, int rw)
102{
103 struct request *rq;
104 struct request_queue *q = sdev->request_queue;
105
106 rq = blk_get_request(q, rw, GFP_NOIO);
107
108 if (!rq) {
109 sdev_printk(KERN_INFO, sdev,
110 "%s: blk_get_request failed\n", __func__);
111 return NULL;
112 }
113
114 if (buflen && blk_rq_map_kern(q, rq, buffer, buflen, GFP_NOIO)) {
115 blk_put_request(rq);
116 sdev_printk(KERN_INFO, sdev,
117 "%s: blk_rq_map_kern failed\n", __func__);
118 return NULL;
119 }
120
121 rq->cmd_type = REQ_TYPE_BLOCK_PC;
122 rq->cmd_flags |= REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
123 REQ_FAILFAST_DRIVER;
124 rq->retries = ALUA_FAILOVER_RETRIES;
125 rq->timeout = ALUA_FAILOVER_TIMEOUT;
126
127 return rq;
128}
129
130/*
131 * submit_std_inquiry - Issue a standard INQUIRY command
132 * @sdev: sdev the command should be send to
133 */
134static int submit_std_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
135{
136 struct request *rq;
137 int err = SCSI_DH_RES_TEMP_UNAVAIL;
138
139 rq = get_alua_req(sdev, h->inq, ALUA_INQUIRY_SIZE, READ);
140 if (!rq)
141 goto done;
142
143 /* Prepare the command. */
144 rq->cmd[0] = INQUIRY;
145 rq->cmd[1] = 0;
146 rq->cmd[2] = 0;
147 rq->cmd[4] = ALUA_INQUIRY_SIZE;
148 rq->cmd_len = COMMAND_SIZE(INQUIRY);
149
150 rq->sense = h->sense;
151 memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
152 rq->sense_len = h->senselen = 0;
153
154 err = blk_execute_rq(rq->q, NULL, rq, 1);
155 if (err == -EIO) {
156 sdev_printk(KERN_INFO, sdev,
157 "%s: std inquiry failed with %x\n",
158 ALUA_DH_NAME, rq->errors);
159 h->senselen = rq->sense_len;
160 err = SCSI_DH_IO;
161 }
162 blk_put_request(rq);
163done:
164 return err;
165}
166
167/*
168 * submit_vpd_inquiry - Issue an INQUIRY VPD page 0x83 command
169 * @sdev: sdev the command should be sent to
170 */
171static int submit_vpd_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
172{
173 struct request *rq;
174 int err = SCSI_DH_RES_TEMP_UNAVAIL;
175
176 rq = get_alua_req(sdev, h->buff, h->bufflen, READ);
177 if (!rq)
178 goto done;
179
180 /* Prepare the command. */
181 rq->cmd[0] = INQUIRY;
182 rq->cmd[1] = 1;
183 rq->cmd[2] = 0x83;
184 rq->cmd[4] = h->bufflen;
185 rq->cmd_len = COMMAND_SIZE(INQUIRY);
186
187 rq->sense = h->sense;
188 memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
189 rq->sense_len = h->senselen = 0;
190
191 err = blk_execute_rq(rq->q, NULL, rq, 1);
192 if (err == -EIO) {
193 sdev_printk(KERN_INFO, sdev,
194 "%s: evpd inquiry failed with %x\n",
195 ALUA_DH_NAME, rq->errors);
196 h->senselen = rq->sense_len;
197 err = SCSI_DH_IO;
198 }
199 blk_put_request(rq);
200done:
201 return err;
202}
203
204/*
205 * submit_rtpg - Issue a REPORT TARGET GROUP STATES command
206 * @sdev: sdev the command should be sent to
207 */
208static unsigned submit_rtpg(struct scsi_device *sdev, struct alua_dh_data *h)
209{
210 struct request *rq;
211 int err = SCSI_DH_RES_TEMP_UNAVAIL;
212
213 rq = get_alua_req(sdev, h->buff, h->bufflen, READ);
214 if (!rq)
215 goto done;
216
217 /* Prepare the command. */
218 rq->cmd[0] = MAINTENANCE_IN;
219 rq->cmd[1] = MI_REPORT_TARGET_PGS;
220 rq->cmd[6] = (h->bufflen >> 24) & 0xff;
221 rq->cmd[7] = (h->bufflen >> 16) & 0xff;
222 rq->cmd[8] = (h->bufflen >> 8) & 0xff;
223 rq->cmd[9] = h->bufflen & 0xff;
224 rq->cmd_len = COMMAND_SIZE(MAINTENANCE_IN);
225
226 rq->sense = h->sense;
227 memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
228 rq->sense_len = h->senselen = 0;
229
230 err = blk_execute_rq(rq->q, NULL, rq, 1);
231 if (err == -EIO) {
232 sdev_printk(KERN_INFO, sdev,
233 "%s: rtpg failed with %x\n",
234 ALUA_DH_NAME, rq->errors);
235 h->senselen = rq->sense_len;
236 err = SCSI_DH_IO;
237 }
238 blk_put_request(rq);
239done:
240 return err;
241}
242
243/*
244 * alua_stpg - Evaluate SET TARGET GROUP STATES
245 * @sdev: the device to be evaluated
246 * @state: the new target group state
247 *
248 * Send a SET TARGET GROUP STATES command to the device.
249 * We only have to test here if we should resubmit the command;
250 * any other error is assumed as a failure.
251 */
252static void stpg_endio(struct request *req, int error)
253{
254 struct alua_dh_data *h = req->end_io_data;
255 struct scsi_sense_hdr sense_hdr;
256 unsigned err = SCSI_DH_OK;
257
258 if (error || host_byte(req->errors) != DID_OK ||
259 msg_byte(req->errors) != COMMAND_COMPLETE) {
260 err = SCSI_DH_IO;
261 goto done;
262 }
263
264 if (h->senselen > 0) {
265 err = scsi_normalize_sense(h->sense, SCSI_SENSE_BUFFERSIZE,
266 &sense_hdr);
267 if (!err) {
268 err = SCSI_DH_IO;
269 goto done;
270 }
271 err = alua_check_sense(h->sdev, &sense_hdr);
272 if (err == ADD_TO_MLQUEUE) {
273 err = SCSI_DH_RETRY;
274 goto done;
275 }
276 sdev_printk(KERN_INFO, h->sdev,
277 "%s: stpg sense code: %02x/%02x/%02x\n",
278 ALUA_DH_NAME, sense_hdr.sense_key,
279 sense_hdr.asc, sense_hdr.ascq);
280 err = SCSI_DH_IO;
281 }
282 if (err == SCSI_DH_OK) {
283 h->state = TPGS_STATE_OPTIMIZED;
284 sdev_printk(KERN_INFO, h->sdev,
285 "%s: port group %02x switched to state %c\n",
286 ALUA_DH_NAME, h->group_id,
287 print_alua_state(h->state));
288 }
289done:
290 req->end_io_data = NULL;
291 __blk_put_request(req->q, req);
292 if (h->callback_fn) {
293 h->callback_fn(h->callback_data, err);
294 h->callback_fn = h->callback_data = NULL;
295 }
296 return;
297}
298
299/*
300 * submit_stpg - Issue a SET TARGET GROUP STATES command
301 *
302 * Currently we're only setting the current target port group state
303 * to 'active/optimized' and let the array firmware figure out
304 * the states of the remaining groups.
305 */
306static unsigned submit_stpg(struct alua_dh_data *h)
307{
308 struct request *rq;
309 int stpg_len = 8;
310 struct scsi_device *sdev = h->sdev;
311
312 /* Prepare the data buffer */
313 memset(h->buff, 0, stpg_len);
314 h->buff[4] = TPGS_STATE_OPTIMIZED & 0x0f;
315 h->buff[6] = (h->group_id >> 8) & 0xff;
316 h->buff[7] = h->group_id & 0xff;
317
318 rq = get_alua_req(sdev, h->buff, stpg_len, WRITE);
319 if (!rq)
320 return SCSI_DH_RES_TEMP_UNAVAIL;
321
322 /* Prepare the command. */
323 rq->cmd[0] = MAINTENANCE_OUT;
324 rq->cmd[1] = MO_SET_TARGET_PGS;
325 rq->cmd[6] = (stpg_len >> 24) & 0xff;
326 rq->cmd[7] = (stpg_len >> 16) & 0xff;
327 rq->cmd[8] = (stpg_len >> 8) & 0xff;
328 rq->cmd[9] = stpg_len & 0xff;
329 rq->cmd_len = COMMAND_SIZE(MAINTENANCE_OUT);
330
331 rq->sense = h->sense;
332 memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
333 rq->sense_len = h->senselen = 0;
334 rq->end_io_data = h;
335
336 blk_execute_rq_nowait(rq->q, NULL, rq, 1, stpg_endio);
337 return SCSI_DH_OK;
338}
339
340/*
341 * alua_std_inquiry - Evaluate standard INQUIRY command
342 * @sdev: device to be checked
343 *
344 * Just extract the TPGS setting to find out if ALUA
345 * is supported.
346 */
347static int alua_std_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
348{
349 int err;
350
351 err = submit_std_inquiry(sdev, h);
352
353 if (err != SCSI_DH_OK)
354 return err;
355
356 /* Check TPGS setting */
357 h->tpgs = (h->inq[5] >> 4) & 0x3;
358 switch (h->tpgs) {
359 case TPGS_MODE_EXPLICIT|TPGS_MODE_IMPLICIT:
360 sdev_printk(KERN_INFO, sdev,
361 "%s: supports implicit and explicit TPGS\n",
362 ALUA_DH_NAME);
363 break;
364 case TPGS_MODE_EXPLICIT:
365 sdev_printk(KERN_INFO, sdev, "%s: supports explicit TPGS\n",
366 ALUA_DH_NAME);
367 break;
368 case TPGS_MODE_IMPLICIT:
369 sdev_printk(KERN_INFO, sdev, "%s: supports implicit TPGS\n",
370 ALUA_DH_NAME);
371 break;
372 default:
373 h->tpgs = TPGS_MODE_NONE;
374 sdev_printk(KERN_INFO, sdev, "%s: not supported\n",
375 ALUA_DH_NAME);
376 err = SCSI_DH_DEV_UNSUPP;
377 break;
378 }
379
380 return err;
381}
382
383/*
384 * alua_vpd_inquiry - Evaluate INQUIRY vpd page 0x83
385 * @sdev: device to be checked
386 *
387 * Extract the relative target port and the target port group
388 * descriptor from the list of identificators.
389 */
390static int alua_vpd_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
391{
392 int len;
393 unsigned err;
394 unsigned char *d;
395
396 retry:
397 err = submit_vpd_inquiry(sdev, h);
398
399 if (err != SCSI_DH_OK)
400 return err;
401
402 /* Check if vpd page exceeds initial buffer */
403 len = (h->buff[2] << 8) + h->buff[3] + 4;
404 if (len > h->bufflen) {
405 /* Resubmit with the correct length */
406 if (realloc_buffer(h, len)) {
407 sdev_printk(KERN_WARNING, sdev,
408 "%s: kmalloc buffer failed\n",
409 ALUA_DH_NAME);
410 /* Temporary failure, bypass */
411 return SCSI_DH_DEV_TEMP_BUSY;
412 }
413 goto retry;
414 }
415
416 /*
417 * Now look for the correct descriptor.
418 */
419 d = h->buff + 4;
420 while (d < h->buff + len) {
421 switch (d[1] & 0xf) {
422 case 0x4:
423 /* Relative target port */
424 h->rel_port = (d[6] << 8) + d[7];
425 break;
426 case 0x5:
427 /* Target port group */
428 h->group_id = (d[6] << 8) + d[7];
429 break;
430 default:
431 break;
432 }
433 d += d[3] + 4;
434 }
435
436 if (h->group_id == -1) {
437 /*
438 * Internal error; TPGS supported but required
439 * VPD identification descriptors not present.
440 * Disable ALUA support
441 */
442 sdev_printk(KERN_INFO, sdev,
443 "%s: No target port descriptors found\n",
444 ALUA_DH_NAME);
445 h->state = TPGS_STATE_OPTIMIZED;
446 h->tpgs = TPGS_MODE_NONE;
447 err = SCSI_DH_DEV_UNSUPP;
448 } else {
449 sdev_printk(KERN_INFO, sdev,
450 "%s: port group %02x rel port %02x\n",
451 ALUA_DH_NAME, h->group_id, h->rel_port);
452 }
453
454 return err;
455}
456
457static char print_alua_state(int state)
458{
459 switch (state) {
460 case TPGS_STATE_OPTIMIZED:
461 return 'A';
462 case TPGS_STATE_NONOPTIMIZED:
463 return 'N';
464 case TPGS_STATE_STANDBY:
465 return 'S';
466 case TPGS_STATE_UNAVAILABLE:
467 return 'U';
468 case TPGS_STATE_LBA_DEPENDENT:
469 return 'L';
470 case TPGS_STATE_OFFLINE:
471 return 'O';
472 case TPGS_STATE_TRANSITIONING:
473 return 'T';
474 default:
475 return 'X';
476 }
477}
478
479static int alua_check_sense(struct scsi_device *sdev,
480 struct scsi_sense_hdr *sense_hdr)
481{
482 switch (sense_hdr->sense_key) {
483 case NOT_READY:
484 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a)
485 /*
486 * LUN Not Accessible - ALUA state transition
487 */
488 return ADD_TO_MLQUEUE;
489 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0b)
490 /*
491 * LUN Not Accessible -- Target port in standby state
492 */
493 return SUCCESS;
494 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0c)
495 /*
496 * LUN Not Accessible -- Target port in unavailable state
497 */
498 return SUCCESS;
499 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x12)
500 /*
501 * LUN Not Ready -- Offline
502 */
503 return SUCCESS;
504 break;
505 case UNIT_ATTENTION:
506 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x00)
507 /*
508 * Power On, Reset, or Bus Device Reset, just retry.
509 */
510 return ADD_TO_MLQUEUE;
511 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x06) {
512 /*
513 * ALUA state changed
514 */
515 return ADD_TO_MLQUEUE;
516 }
517 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x07) {
518 /*
519 * Implicit ALUA state transition failed
520 */
521 return ADD_TO_MLQUEUE;
522 }
523 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x0e) {
524 /*
525 * REPORTED_LUNS_DATA_HAS_CHANGED is reported
526 * when switching controllers on targets like
527 * Intel Multi-Flex. We can just retry.
528 */
529 return ADD_TO_MLQUEUE;
530 }
531
532 break;
533 }
534
535 return SCSI_RETURN_NOT_HANDLED;
536}
537
538/*
539 * alua_rtpg - Evaluate REPORT TARGET GROUP STATES
540 * @sdev: the device to be evaluated.
541 *
542 * Evaluate the Target Port Group State.
543 * Returns SCSI_DH_DEV_OFFLINED if the path is
544 * found to be unusable.
545 */
546static int alua_rtpg(struct scsi_device *sdev, struct alua_dh_data *h)
547{
548 struct scsi_sense_hdr sense_hdr;
549 int len, k, off, valid_states = 0;
550 char *ucp;
551 unsigned err;
552 unsigned long expiry, interval = 10;
553
554 expiry = round_jiffies_up(jiffies + ALUA_FAILOVER_TIMEOUT);
555 retry:
556 err = submit_rtpg(sdev, h);
557
558 if (err == SCSI_DH_IO && h->senselen > 0) {
559 err = scsi_normalize_sense(h->sense, SCSI_SENSE_BUFFERSIZE,
560 &sense_hdr);
561 if (!err)
562 return SCSI_DH_IO;
563
564 err = alua_check_sense(sdev, &sense_hdr);
565 if (err == ADD_TO_MLQUEUE && time_before(jiffies, expiry))
566 goto retry;
567 sdev_printk(KERN_INFO, sdev,
568 "%s: rtpg sense code %02x/%02x/%02x\n",
569 ALUA_DH_NAME, sense_hdr.sense_key,
570 sense_hdr.asc, sense_hdr.ascq);
571 err = SCSI_DH_IO;
572 }
573 if (err != SCSI_DH_OK)
574 return err;
575
576 len = (h->buff[0] << 24) + (h->buff[1] << 16) +
577 (h->buff[2] << 8) + h->buff[3] + 4;
578
579 if (len > h->bufflen) {
580 /* Resubmit with the correct length */
581 if (realloc_buffer(h, len)) {
582 sdev_printk(KERN_WARNING, sdev,
583 "%s: kmalloc buffer failed\n",__func__);
584 /* Temporary failure, bypass */
585 return SCSI_DH_DEV_TEMP_BUSY;
586 }
587 goto retry;
588 }
589
590 for (k = 4, ucp = h->buff + 4; k < len; k += off, ucp += off) {
591 if (h->group_id == (ucp[2] << 8) + ucp[3]) {
592 h->state = ucp[0] & 0x0f;
593 valid_states = ucp[1];
594 }
595 off = 8 + (ucp[7] * 4);
596 }
597
598 sdev_printk(KERN_INFO, sdev,
599 "%s: port group %02x state %c supports %c%c%c%c%c%c%c\n",
600 ALUA_DH_NAME, h->group_id, print_alua_state(h->state),
601 valid_states&TPGS_SUPPORT_TRANSITION?'T':'t',
602 valid_states&TPGS_SUPPORT_OFFLINE?'O':'o',
603 valid_states&TPGS_SUPPORT_LBA_DEPENDENT?'L':'l',
604 valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u',
605 valid_states&TPGS_SUPPORT_STANDBY?'S':'s',
606 valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n',
607 valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a');
608
609 switch (h->state) {
610 case TPGS_STATE_TRANSITIONING:
611 if (time_before(jiffies, expiry)) {
612 /* State transition, retry */
613 interval *= 10;
614 msleep(interval);
615 goto retry;
616 }
617 /* Transitioning time exceeded, set port to standby */
618 err = SCSI_DH_RETRY;
619 h->state = TPGS_STATE_STANDBY;
620 break;
621 case TPGS_STATE_OFFLINE:
622 case TPGS_STATE_UNAVAILABLE:
623 /* Path unusable for unavailable/offline */
624 err = SCSI_DH_DEV_OFFLINED;
625 break;
626 default:
627 /* Useable path if active */
628 err = SCSI_DH_OK;
629 break;
630 }
631 return err;
632}
633
634/*
635 * alua_initialize - Initialize ALUA state
636 * @sdev: the device to be initialized
637 *
638 * For the prep_fn to work correctly we have
639 * to initialize the ALUA state for the device.
640 */
641static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h)
642{
643 int err;
644
645 err = alua_std_inquiry(sdev, h);
646 if (err != SCSI_DH_OK)
647 goto out;
648
649 err = alua_vpd_inquiry(sdev, h);
650 if (err != SCSI_DH_OK)
651 goto out;
652
653 err = alua_rtpg(sdev, h);
654 if (err != SCSI_DH_OK)
655 goto out;
656
657out:
658 return err;
659}
660
661/*
662 * alua_activate - activate a path
663 * @sdev: device on the path to be activated
664 *
665 * We're currently switching the port group to be activated only and
666 * let the array figure out the rest.
667 * There may be other arrays which require us to switch all port groups
668 * based on a certain policy. But until we actually encounter them it
669 * should be okay.
670 */
671static int alua_activate(struct scsi_device *sdev,
672 activate_complete fn, void *data)
673{
674 struct alua_dh_data *h = get_alua_data(sdev);
675 int err = SCSI_DH_OK;
676
677 if (h->group_id != -1) {
678 err = alua_rtpg(sdev, h);
679 if (err != SCSI_DH_OK)
680 goto out;
681 }
682
683 if (h->tpgs & TPGS_MODE_EXPLICIT &&
684 h->state != TPGS_STATE_OPTIMIZED &&
685 h->state != TPGS_STATE_LBA_DEPENDENT) {
686 h->callback_fn = fn;
687 h->callback_data = data;
688 err = submit_stpg(h);
689 if (err == SCSI_DH_OK)
690 return 0;
691 h->callback_fn = h->callback_data = NULL;
692 }
693
694out:
695 if (fn)
696 fn(data, err);
697 return 0;
698}
699
700/*
701 * alua_prep_fn - request callback
702 *
703 * Fail I/O to all paths not in state
704 * active/optimized or active/non-optimized.
705 */
706static int alua_prep_fn(struct scsi_device *sdev, struct request *req)
707{
708 struct alua_dh_data *h = get_alua_data(sdev);
709 int ret = BLKPREP_OK;
710
711 if (h->state == TPGS_STATE_TRANSITIONING)
712 ret = BLKPREP_DEFER;
713 else if (h->state != TPGS_STATE_OPTIMIZED &&
714 h->state != TPGS_STATE_NONOPTIMIZED &&
715 h->state != TPGS_STATE_LBA_DEPENDENT) {
716 ret = BLKPREP_KILL;
717 req->cmd_flags |= REQ_QUIET;
718 }
719 return ret;
720
721}
722
723static const struct scsi_dh_devlist alua_dev_list[] = {
724 {"HP", "MSA VOLUME" },
725 {"HP", "HSV101" },
726 {"HP", "HSV111" },
727 {"HP", "HSV200" },
728 {"HP", "HSV210" },
729 {"HP", "HSV300" },
730 {"IBM", "2107900" },
731 {"IBM", "2145" },
732 {"Pillar", "Axiom" },
733 {"Intel", "Multi-Flex"},
734 {"NETAPP", "LUN"},
735 {"NETAPP", "LUN C-Mode"},
736 {"AIX", "NVDISK"},
737 {"Promise", "VTrak"},
738 {NULL, NULL}
739};
740
741static int alua_bus_attach(struct scsi_device *sdev);
742static void alua_bus_detach(struct scsi_device *sdev);
743
744static struct scsi_device_handler alua_dh = {
745 .name = ALUA_DH_NAME,
746 .module = THIS_MODULE,
747 .devlist = alua_dev_list,
748 .attach = alua_bus_attach,
749 .detach = alua_bus_detach,
750 .prep_fn = alua_prep_fn,
751 .check_sense = alua_check_sense,
752 .activate = alua_activate,
753};
754
755/*
756 * alua_bus_attach - Attach device handler
757 * @sdev: device to be attached to
758 */
759static int alua_bus_attach(struct scsi_device *sdev)
760{
761 struct scsi_dh_data *scsi_dh_data;
762 struct alua_dh_data *h;
763 unsigned long flags;
764 int err = SCSI_DH_OK;
765
766 scsi_dh_data = kzalloc(sizeof(*scsi_dh_data)
767 + sizeof(*h) , GFP_KERNEL);
768 if (!scsi_dh_data) {
769 sdev_printk(KERN_ERR, sdev, "%s: Attach failed\n",
770 ALUA_DH_NAME);
771 return -ENOMEM;
772 }
773
774 scsi_dh_data->scsi_dh = &alua_dh;
775 h = (struct alua_dh_data *) scsi_dh_data->buf;
776 h->tpgs = TPGS_MODE_UNINITIALIZED;
777 h->state = TPGS_STATE_OPTIMIZED;
778 h->group_id = -1;
779 h->rel_port = -1;
780 h->buff = h->inq;
781 h->bufflen = ALUA_INQUIRY_SIZE;
782 h->sdev = sdev;
783
784 err = alua_initialize(sdev, h);
785 if ((err != SCSI_DH_OK) && (err != SCSI_DH_DEV_OFFLINED))
786 goto failed;
787
788 if (!try_module_get(THIS_MODULE))
789 goto failed;
790
791 spin_lock_irqsave(sdev->request_queue->queue_lock, flags);
792 sdev->scsi_dh_data = scsi_dh_data;
793 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
794
795 return 0;
796
797failed:
798 kfree(scsi_dh_data);
799 sdev_printk(KERN_ERR, sdev, "%s: not attached\n", ALUA_DH_NAME);
800 return -EINVAL;
801}
802
803/*
804 * alua_bus_detach - Detach device handler
805 * @sdev: device to be detached from
806 */
807static void alua_bus_detach(struct scsi_device *sdev)
808{
809 struct scsi_dh_data *scsi_dh_data;
810 struct alua_dh_data *h;
811 unsigned long flags;
812
813 spin_lock_irqsave(sdev->request_queue->queue_lock, flags);
814 scsi_dh_data = sdev->scsi_dh_data;
815 sdev->scsi_dh_data = NULL;
816 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
817
818 h = (struct alua_dh_data *) scsi_dh_data->buf;
819 if (h->buff && h->inq != h->buff)
820 kfree(h->buff);
821 kfree(scsi_dh_data);
822 module_put(THIS_MODULE);
823 sdev_printk(KERN_NOTICE, sdev, "%s: Detached\n", ALUA_DH_NAME);
824}
825
826static int __init alua_init(void)
827{
828 int r;
829
830 r = scsi_register_device_handler(&alua_dh);
831 if (r != 0)
832 printk(KERN_ERR "%s: Failed to register scsi device handler",
833 ALUA_DH_NAME);
834 return r;
835}
836
837static void __exit alua_exit(void)
838{
839 scsi_unregister_device_handler(&alua_dh);
840}
841
842module_init(alua_init);
843module_exit(alua_exit);
844
845MODULE_DESCRIPTION("DM Multipath ALUA support");
846MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
847MODULE_LICENSE("GPL");
848MODULE_VERSION(ALUA_DH_VER);
1/*
2 * Generic SCSI-3 ALUA SCSI Device Handler
3 *
4 * Copyright (C) 2007-2010 Hannes Reinecke, SUSE Linux Products GmbH.
5 * All rights reserved.
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
20 *
21 */
22#include <linux/slab.h>
23#include <linux/delay.h>
24#include <linux/module.h>
25#include <scsi/scsi.h>
26#include <scsi/scsi_eh.h>
27#include <scsi/scsi_dh.h>
28
29#define ALUA_DH_NAME "alua"
30#define ALUA_DH_VER "1.3"
31
32#define TPGS_STATE_OPTIMIZED 0x0
33#define TPGS_STATE_NONOPTIMIZED 0x1
34#define TPGS_STATE_STANDBY 0x2
35#define TPGS_STATE_UNAVAILABLE 0x3
36#define TPGS_STATE_LBA_DEPENDENT 0x4
37#define TPGS_STATE_OFFLINE 0xe
38#define TPGS_STATE_TRANSITIONING 0xf
39
40#define TPGS_SUPPORT_NONE 0x00
41#define TPGS_SUPPORT_OPTIMIZED 0x01
42#define TPGS_SUPPORT_NONOPTIMIZED 0x02
43#define TPGS_SUPPORT_STANDBY 0x04
44#define TPGS_SUPPORT_UNAVAILABLE 0x08
45#define TPGS_SUPPORT_LBA_DEPENDENT 0x10
46#define TPGS_SUPPORT_OFFLINE 0x40
47#define TPGS_SUPPORT_TRANSITION 0x80
48
49#define TPGS_MODE_UNINITIALIZED -1
50#define TPGS_MODE_NONE 0x0
51#define TPGS_MODE_IMPLICIT 0x1
52#define TPGS_MODE_EXPLICIT 0x2
53
54#define ALUA_INQUIRY_SIZE 36
55#define ALUA_FAILOVER_TIMEOUT (60 * HZ)
56#define ALUA_FAILOVER_RETRIES 5
57
58/* flags passed from user level */
59#define ALUA_OPTIMIZE_STPG 1
60
61struct alua_dh_data {
62 int group_id;
63 int rel_port;
64 int tpgs;
65 int state;
66 int pref;
67 unsigned flags; /* used for optimizing STPG */
68 unsigned char inq[ALUA_INQUIRY_SIZE];
69 unsigned char *buff;
70 int bufflen;
71 unsigned char sense[SCSI_SENSE_BUFFERSIZE];
72 int senselen;
73 struct scsi_device *sdev;
74 activate_complete callback_fn;
75 void *callback_data;
76};
77
78#define ALUA_POLICY_SWITCH_CURRENT 0
79#define ALUA_POLICY_SWITCH_ALL 1
80
81static char print_alua_state(int);
82static int alua_check_sense(struct scsi_device *, struct scsi_sense_hdr *);
83
84static inline struct alua_dh_data *get_alua_data(struct scsi_device *sdev)
85{
86 struct scsi_dh_data *scsi_dh_data = sdev->scsi_dh_data;
87 BUG_ON(scsi_dh_data == NULL);
88 return ((struct alua_dh_data *) scsi_dh_data->buf);
89}
90
91static int realloc_buffer(struct alua_dh_data *h, unsigned len)
92{
93 if (h->buff && h->buff != h->inq)
94 kfree(h->buff);
95
96 h->buff = kmalloc(len, GFP_NOIO);
97 if (!h->buff) {
98 h->buff = h->inq;
99 h->bufflen = ALUA_INQUIRY_SIZE;
100 return 1;
101 }
102 h->bufflen = len;
103 return 0;
104}
105
106static struct request *get_alua_req(struct scsi_device *sdev,
107 void *buffer, unsigned buflen, int rw)
108{
109 struct request *rq;
110 struct request_queue *q = sdev->request_queue;
111
112 rq = blk_get_request(q, rw, GFP_NOIO);
113
114 if (!rq) {
115 sdev_printk(KERN_INFO, sdev,
116 "%s: blk_get_request failed\n", __func__);
117 return NULL;
118 }
119
120 if (buflen && blk_rq_map_kern(q, rq, buffer, buflen, GFP_NOIO)) {
121 blk_put_request(rq);
122 sdev_printk(KERN_INFO, sdev,
123 "%s: blk_rq_map_kern failed\n", __func__);
124 return NULL;
125 }
126
127 rq->cmd_type = REQ_TYPE_BLOCK_PC;
128 rq->cmd_flags |= REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
129 REQ_FAILFAST_DRIVER;
130 rq->retries = ALUA_FAILOVER_RETRIES;
131 rq->timeout = ALUA_FAILOVER_TIMEOUT;
132
133 return rq;
134}
135
136/*
137 * submit_vpd_inquiry - Issue an INQUIRY VPD page 0x83 command
138 * @sdev: sdev the command should be sent to
139 */
140static int submit_vpd_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
141{
142 struct request *rq;
143 int err = SCSI_DH_RES_TEMP_UNAVAIL;
144
145 rq = get_alua_req(sdev, h->buff, h->bufflen, READ);
146 if (!rq)
147 goto done;
148
149 /* Prepare the command. */
150 rq->cmd[0] = INQUIRY;
151 rq->cmd[1] = 1;
152 rq->cmd[2] = 0x83;
153 rq->cmd[4] = h->bufflen;
154 rq->cmd_len = COMMAND_SIZE(INQUIRY);
155
156 rq->sense = h->sense;
157 memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
158 rq->sense_len = h->senselen = 0;
159
160 err = blk_execute_rq(rq->q, NULL, rq, 1);
161 if (err == -EIO) {
162 sdev_printk(KERN_INFO, sdev,
163 "%s: evpd inquiry failed with %x\n",
164 ALUA_DH_NAME, rq->errors);
165 h->senselen = rq->sense_len;
166 err = SCSI_DH_IO;
167 }
168 blk_put_request(rq);
169done:
170 return err;
171}
172
173/*
174 * submit_rtpg - Issue a REPORT TARGET GROUP STATES command
175 * @sdev: sdev the command should be sent to
176 */
177static unsigned submit_rtpg(struct scsi_device *sdev, struct alua_dh_data *h)
178{
179 struct request *rq;
180 int err = SCSI_DH_RES_TEMP_UNAVAIL;
181
182 rq = get_alua_req(sdev, h->buff, h->bufflen, READ);
183 if (!rq)
184 goto done;
185
186 /* Prepare the command. */
187 rq->cmd[0] = MAINTENANCE_IN;
188 rq->cmd[1] = MI_REPORT_TARGET_PGS;
189 rq->cmd[6] = (h->bufflen >> 24) & 0xff;
190 rq->cmd[7] = (h->bufflen >> 16) & 0xff;
191 rq->cmd[8] = (h->bufflen >> 8) & 0xff;
192 rq->cmd[9] = h->bufflen & 0xff;
193 rq->cmd_len = COMMAND_SIZE(MAINTENANCE_IN);
194
195 rq->sense = h->sense;
196 memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
197 rq->sense_len = h->senselen = 0;
198
199 err = blk_execute_rq(rq->q, NULL, rq, 1);
200 if (err == -EIO) {
201 sdev_printk(KERN_INFO, sdev,
202 "%s: rtpg failed with %x\n",
203 ALUA_DH_NAME, rq->errors);
204 h->senselen = rq->sense_len;
205 err = SCSI_DH_IO;
206 }
207 blk_put_request(rq);
208done:
209 return err;
210}
211
212/*
213 * alua_stpg - Evaluate SET TARGET GROUP STATES
214 * @sdev: the device to be evaluated
215 * @state: the new target group state
216 *
217 * Send a SET TARGET GROUP STATES command to the device.
218 * We only have to test here if we should resubmit the command;
219 * any other error is assumed as a failure.
220 */
221static void stpg_endio(struct request *req, int error)
222{
223 struct alua_dh_data *h = req->end_io_data;
224 struct scsi_sense_hdr sense_hdr;
225 unsigned err = SCSI_DH_OK;
226
227 if (error || host_byte(req->errors) != DID_OK ||
228 msg_byte(req->errors) != COMMAND_COMPLETE) {
229 err = SCSI_DH_IO;
230 goto done;
231 }
232
233 if (h->senselen > 0) {
234 err = scsi_normalize_sense(h->sense, SCSI_SENSE_BUFFERSIZE,
235 &sense_hdr);
236 if (!err) {
237 err = SCSI_DH_IO;
238 goto done;
239 }
240 err = alua_check_sense(h->sdev, &sense_hdr);
241 if (err == ADD_TO_MLQUEUE) {
242 err = SCSI_DH_RETRY;
243 goto done;
244 }
245 sdev_printk(KERN_INFO, h->sdev,
246 "%s: stpg sense code: %02x/%02x/%02x\n",
247 ALUA_DH_NAME, sense_hdr.sense_key,
248 sense_hdr.asc, sense_hdr.ascq);
249 err = SCSI_DH_IO;
250 }
251 if (err == SCSI_DH_OK) {
252 h->state = TPGS_STATE_OPTIMIZED;
253 sdev_printk(KERN_INFO, h->sdev,
254 "%s: port group %02x switched to state %c\n",
255 ALUA_DH_NAME, h->group_id,
256 print_alua_state(h->state));
257 }
258done:
259 req->end_io_data = NULL;
260 __blk_put_request(req->q, req);
261 if (h->callback_fn) {
262 h->callback_fn(h->callback_data, err);
263 h->callback_fn = h->callback_data = NULL;
264 }
265 return;
266}
267
268/*
269 * submit_stpg - Issue a SET TARGET GROUP STATES command
270 *
271 * Currently we're only setting the current target port group state
272 * to 'active/optimized' and let the array firmware figure out
273 * the states of the remaining groups.
274 */
275static unsigned submit_stpg(struct alua_dh_data *h)
276{
277 struct request *rq;
278 int stpg_len = 8;
279 struct scsi_device *sdev = h->sdev;
280
281 /* Prepare the data buffer */
282 memset(h->buff, 0, stpg_len);
283 h->buff[4] = TPGS_STATE_OPTIMIZED & 0x0f;
284 h->buff[6] = (h->group_id >> 8) & 0xff;
285 h->buff[7] = h->group_id & 0xff;
286
287 rq = get_alua_req(sdev, h->buff, stpg_len, WRITE);
288 if (!rq)
289 return SCSI_DH_RES_TEMP_UNAVAIL;
290
291 /* Prepare the command. */
292 rq->cmd[0] = MAINTENANCE_OUT;
293 rq->cmd[1] = MO_SET_TARGET_PGS;
294 rq->cmd[6] = (stpg_len >> 24) & 0xff;
295 rq->cmd[7] = (stpg_len >> 16) & 0xff;
296 rq->cmd[8] = (stpg_len >> 8) & 0xff;
297 rq->cmd[9] = stpg_len & 0xff;
298 rq->cmd_len = COMMAND_SIZE(MAINTENANCE_OUT);
299
300 rq->sense = h->sense;
301 memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
302 rq->sense_len = h->senselen = 0;
303 rq->end_io_data = h;
304
305 blk_execute_rq_nowait(rq->q, NULL, rq, 1, stpg_endio);
306 return SCSI_DH_OK;
307}
308
309/*
310 * alua_check_tpgs - Evaluate TPGS setting
311 * @sdev: device to be checked
312 *
313 * Examine the TPGS setting of the sdev to find out if ALUA
314 * is supported.
315 */
316static int alua_check_tpgs(struct scsi_device *sdev, struct alua_dh_data *h)
317{
318 int err = SCSI_DH_OK;
319
320 h->tpgs = scsi_device_tpgs(sdev);
321 switch (h->tpgs) {
322 case TPGS_MODE_EXPLICIT|TPGS_MODE_IMPLICIT:
323 sdev_printk(KERN_INFO, sdev,
324 "%s: supports implicit and explicit TPGS\n",
325 ALUA_DH_NAME);
326 break;
327 case TPGS_MODE_EXPLICIT:
328 sdev_printk(KERN_INFO, sdev, "%s: supports explicit TPGS\n",
329 ALUA_DH_NAME);
330 break;
331 case TPGS_MODE_IMPLICIT:
332 sdev_printk(KERN_INFO, sdev, "%s: supports implicit TPGS\n",
333 ALUA_DH_NAME);
334 break;
335 default:
336 h->tpgs = TPGS_MODE_NONE;
337 sdev_printk(KERN_INFO, sdev, "%s: not supported\n",
338 ALUA_DH_NAME);
339 err = SCSI_DH_DEV_UNSUPP;
340 break;
341 }
342
343 return err;
344}
345
346/*
347 * alua_vpd_inquiry - Evaluate INQUIRY vpd page 0x83
348 * @sdev: device to be checked
349 *
350 * Extract the relative target port and the target port group
351 * descriptor from the list of identificators.
352 */
353static int alua_vpd_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
354{
355 int len;
356 unsigned err;
357 unsigned char *d;
358
359 retry:
360 err = submit_vpd_inquiry(sdev, h);
361
362 if (err != SCSI_DH_OK)
363 return err;
364
365 /* Check if vpd page exceeds initial buffer */
366 len = (h->buff[2] << 8) + h->buff[3] + 4;
367 if (len > h->bufflen) {
368 /* Resubmit with the correct length */
369 if (realloc_buffer(h, len)) {
370 sdev_printk(KERN_WARNING, sdev,
371 "%s: kmalloc buffer failed\n",
372 ALUA_DH_NAME);
373 /* Temporary failure, bypass */
374 return SCSI_DH_DEV_TEMP_BUSY;
375 }
376 goto retry;
377 }
378
379 /*
380 * Now look for the correct descriptor.
381 */
382 d = h->buff + 4;
383 while (d < h->buff + len) {
384 switch (d[1] & 0xf) {
385 case 0x4:
386 /* Relative target port */
387 h->rel_port = (d[6] << 8) + d[7];
388 break;
389 case 0x5:
390 /* Target port group */
391 h->group_id = (d[6] << 8) + d[7];
392 break;
393 default:
394 break;
395 }
396 d += d[3] + 4;
397 }
398
399 if (h->group_id == -1) {
400 /*
401 * Internal error; TPGS supported but required
402 * VPD identification descriptors not present.
403 * Disable ALUA support
404 */
405 sdev_printk(KERN_INFO, sdev,
406 "%s: No target port descriptors found\n",
407 ALUA_DH_NAME);
408 h->state = TPGS_STATE_OPTIMIZED;
409 h->tpgs = TPGS_MODE_NONE;
410 err = SCSI_DH_DEV_UNSUPP;
411 } else {
412 sdev_printk(KERN_INFO, sdev,
413 "%s: port group %02x rel port %02x\n",
414 ALUA_DH_NAME, h->group_id, h->rel_port);
415 }
416
417 return err;
418}
419
420static char print_alua_state(int state)
421{
422 switch (state) {
423 case TPGS_STATE_OPTIMIZED:
424 return 'A';
425 case TPGS_STATE_NONOPTIMIZED:
426 return 'N';
427 case TPGS_STATE_STANDBY:
428 return 'S';
429 case TPGS_STATE_UNAVAILABLE:
430 return 'U';
431 case TPGS_STATE_LBA_DEPENDENT:
432 return 'L';
433 case TPGS_STATE_OFFLINE:
434 return 'O';
435 case TPGS_STATE_TRANSITIONING:
436 return 'T';
437 default:
438 return 'X';
439 }
440}
441
442static int alua_check_sense(struct scsi_device *sdev,
443 struct scsi_sense_hdr *sense_hdr)
444{
445 switch (sense_hdr->sense_key) {
446 case NOT_READY:
447 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a)
448 /*
449 * LUN Not Accessible - ALUA state transition
450 */
451 return ADD_TO_MLQUEUE;
452 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0b)
453 /*
454 * LUN Not Accessible -- Target port in standby state
455 */
456 return SUCCESS;
457 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0c)
458 /*
459 * LUN Not Accessible -- Target port in unavailable state
460 */
461 return SUCCESS;
462 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x12)
463 /*
464 * LUN Not Ready -- Offline
465 */
466 return SUCCESS;
467 break;
468 case UNIT_ATTENTION:
469 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x00)
470 /*
471 * Power On, Reset, or Bus Device Reset, just retry.
472 */
473 return ADD_TO_MLQUEUE;
474 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x01)
475 /*
476 * Mode Parameters Changed
477 */
478 return ADD_TO_MLQUEUE;
479 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x06)
480 /*
481 * ALUA state changed
482 */
483 return ADD_TO_MLQUEUE;
484 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x07)
485 /*
486 * Implicit ALUA state transition failed
487 */
488 return ADD_TO_MLQUEUE;
489 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x03)
490 /*
491 * Inquiry data has changed
492 */
493 return ADD_TO_MLQUEUE;
494 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x0e)
495 /*
496 * REPORTED_LUNS_DATA_HAS_CHANGED is reported
497 * when switching controllers on targets like
498 * Intel Multi-Flex. We can just retry.
499 */
500 return ADD_TO_MLQUEUE;
501 break;
502 }
503
504 return SCSI_RETURN_NOT_HANDLED;
505}
506
507/*
508 * alua_rtpg - Evaluate REPORT TARGET GROUP STATES
509 * @sdev: the device to be evaluated.
510 *
511 * Evaluate the Target Port Group State.
512 * Returns SCSI_DH_DEV_OFFLINED if the path is
513 * found to be unusable.
514 */
515static int alua_rtpg(struct scsi_device *sdev, struct alua_dh_data *h)
516{
517 struct scsi_sense_hdr sense_hdr;
518 int len, k, off, valid_states = 0;
519 unsigned char *ucp;
520 unsigned err;
521 unsigned long expiry, interval = 1000;
522
523 expiry = round_jiffies_up(jiffies + ALUA_FAILOVER_TIMEOUT);
524 retry:
525 err = submit_rtpg(sdev, h);
526
527 if (err == SCSI_DH_IO && h->senselen > 0) {
528 err = scsi_normalize_sense(h->sense, SCSI_SENSE_BUFFERSIZE,
529 &sense_hdr);
530 if (!err)
531 return SCSI_DH_IO;
532
533 err = alua_check_sense(sdev, &sense_hdr);
534 if (err == ADD_TO_MLQUEUE && time_before(jiffies, expiry))
535 goto retry;
536 sdev_printk(KERN_INFO, sdev,
537 "%s: rtpg sense code %02x/%02x/%02x\n",
538 ALUA_DH_NAME, sense_hdr.sense_key,
539 sense_hdr.asc, sense_hdr.ascq);
540 err = SCSI_DH_IO;
541 }
542 if (err != SCSI_DH_OK)
543 return err;
544
545 len = (h->buff[0] << 24) + (h->buff[1] << 16) +
546 (h->buff[2] << 8) + h->buff[3] + 4;
547
548 if (len > h->bufflen) {
549 /* Resubmit with the correct length */
550 if (realloc_buffer(h, len)) {
551 sdev_printk(KERN_WARNING, sdev,
552 "%s: kmalloc buffer failed\n",__func__);
553 /* Temporary failure, bypass */
554 return SCSI_DH_DEV_TEMP_BUSY;
555 }
556 goto retry;
557 }
558
559 for (k = 4, ucp = h->buff + 4; k < len; k += off, ucp += off) {
560 if (h->group_id == (ucp[2] << 8) + ucp[3]) {
561 h->state = ucp[0] & 0x0f;
562 h->pref = ucp[0] >> 7;
563 valid_states = ucp[1];
564 }
565 off = 8 + (ucp[7] * 4);
566 }
567
568 sdev_printk(KERN_INFO, sdev,
569 "%s: port group %02x state %c %s supports %c%c%c%c%c%c%c\n",
570 ALUA_DH_NAME, h->group_id, print_alua_state(h->state),
571 h->pref ? "preferred" : "non-preferred",
572 valid_states&TPGS_SUPPORT_TRANSITION?'T':'t',
573 valid_states&TPGS_SUPPORT_OFFLINE?'O':'o',
574 valid_states&TPGS_SUPPORT_LBA_DEPENDENT?'L':'l',
575 valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u',
576 valid_states&TPGS_SUPPORT_STANDBY?'S':'s',
577 valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n',
578 valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a');
579
580 switch (h->state) {
581 case TPGS_STATE_TRANSITIONING:
582 if (time_before(jiffies, expiry)) {
583 /* State transition, retry */
584 interval *= 2;
585 msleep(interval);
586 goto retry;
587 }
588 /* Transitioning time exceeded, set port to standby */
589 err = SCSI_DH_RETRY;
590 h->state = TPGS_STATE_STANDBY;
591 break;
592 case TPGS_STATE_OFFLINE:
593 /* Path unusable */
594 err = SCSI_DH_DEV_OFFLINED;
595 break;
596 default:
597 /* Useable path if active */
598 err = SCSI_DH_OK;
599 break;
600 }
601 return err;
602}
603
604/*
605 * alua_initialize - Initialize ALUA state
606 * @sdev: the device to be initialized
607 *
608 * For the prep_fn to work correctly we have
609 * to initialize the ALUA state for the device.
610 */
611static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h)
612{
613 int err;
614
615 err = alua_check_tpgs(sdev, h);
616 if (err != SCSI_DH_OK)
617 goto out;
618
619 err = alua_vpd_inquiry(sdev, h);
620 if (err != SCSI_DH_OK)
621 goto out;
622
623 err = alua_rtpg(sdev, h);
624 if (err != SCSI_DH_OK)
625 goto out;
626
627out:
628 return err;
629}
630/*
631 * alua_set_params - set/unset the optimize flag
632 * @sdev: device on the path to be activated
633 * params - parameters in the following format
634 * "no_of_params\0param1\0param2\0param3\0...\0"
635 * For example, to set the flag pass the following parameters
636 * from multipath.conf
637 * hardware_handler "2 alua 1"
638 */
639static int alua_set_params(struct scsi_device *sdev, const char *params)
640{
641 struct alua_dh_data *h = get_alua_data(sdev);
642 unsigned int optimize = 0, argc;
643 const char *p = params;
644 int result = SCSI_DH_OK;
645
646 if ((sscanf(params, "%u", &argc) != 1) || (argc != 1))
647 return -EINVAL;
648
649 while (*p++)
650 ;
651 if ((sscanf(p, "%u", &optimize) != 1) || (optimize > 1))
652 return -EINVAL;
653
654 if (optimize)
655 h->flags |= ALUA_OPTIMIZE_STPG;
656 else
657 h->flags &= ~ALUA_OPTIMIZE_STPG;
658
659 return result;
660}
661
662/*
663 * alua_activate - activate a path
664 * @sdev: device on the path to be activated
665 *
666 * We're currently switching the port group to be activated only and
667 * let the array figure out the rest.
668 * There may be other arrays which require us to switch all port groups
669 * based on a certain policy. But until we actually encounter them it
670 * should be okay.
671 */
672static int alua_activate(struct scsi_device *sdev,
673 activate_complete fn, void *data)
674{
675 struct alua_dh_data *h = get_alua_data(sdev);
676 int err = SCSI_DH_OK;
677 int stpg = 0;
678
679 err = alua_rtpg(sdev, h);
680 if (err != SCSI_DH_OK)
681 goto out;
682
683 if (h->tpgs & TPGS_MODE_EXPLICIT) {
684 switch (h->state) {
685 case TPGS_STATE_NONOPTIMIZED:
686 stpg = 1;
687 if ((h->flags & ALUA_OPTIMIZE_STPG) &&
688 (!h->pref) &&
689 (h->tpgs & TPGS_MODE_IMPLICIT))
690 stpg = 0;
691 break;
692 case TPGS_STATE_STANDBY:
693 stpg = 1;
694 break;
695 case TPGS_STATE_UNAVAILABLE:
696 case TPGS_STATE_OFFLINE:
697 err = SCSI_DH_IO;
698 break;
699 case TPGS_STATE_TRANSITIONING:
700 err = SCSI_DH_RETRY;
701 break;
702 default:
703 break;
704 }
705 }
706
707 if (stpg) {
708 h->callback_fn = fn;
709 h->callback_data = data;
710 err = submit_stpg(h);
711 if (err == SCSI_DH_OK)
712 return 0;
713 h->callback_fn = h->callback_data = NULL;
714 }
715
716out:
717 if (fn)
718 fn(data, err);
719 return 0;
720}
721
722/*
723 * alua_prep_fn - request callback
724 *
725 * Fail I/O to all paths not in state
726 * active/optimized or active/non-optimized.
727 */
728static int alua_prep_fn(struct scsi_device *sdev, struct request *req)
729{
730 struct alua_dh_data *h = get_alua_data(sdev);
731 int ret = BLKPREP_OK;
732
733 if (h->state == TPGS_STATE_TRANSITIONING)
734 ret = BLKPREP_DEFER;
735 else if (h->state != TPGS_STATE_OPTIMIZED &&
736 h->state != TPGS_STATE_NONOPTIMIZED &&
737 h->state != TPGS_STATE_LBA_DEPENDENT) {
738 ret = BLKPREP_KILL;
739 req->cmd_flags |= REQ_QUIET;
740 }
741 return ret;
742
743}
744
745static bool alua_match(struct scsi_device *sdev)
746{
747 return (scsi_device_tpgs(sdev) != 0);
748}
749
750static int alua_bus_attach(struct scsi_device *sdev);
751static void alua_bus_detach(struct scsi_device *sdev);
752
753static struct scsi_device_handler alua_dh = {
754 .name = ALUA_DH_NAME,
755 .module = THIS_MODULE,
756 .attach = alua_bus_attach,
757 .detach = alua_bus_detach,
758 .prep_fn = alua_prep_fn,
759 .check_sense = alua_check_sense,
760 .activate = alua_activate,
761 .set_params = alua_set_params,
762 .match = alua_match,
763};
764
765/*
766 * alua_bus_attach - Attach device handler
767 * @sdev: device to be attached to
768 */
769static int alua_bus_attach(struct scsi_device *sdev)
770{
771 struct scsi_dh_data *scsi_dh_data;
772 struct alua_dh_data *h;
773 unsigned long flags;
774 int err = SCSI_DH_OK;
775
776 scsi_dh_data = kzalloc(sizeof(*scsi_dh_data)
777 + sizeof(*h) , GFP_KERNEL);
778 if (!scsi_dh_data) {
779 sdev_printk(KERN_ERR, sdev, "%s: Attach failed\n",
780 ALUA_DH_NAME);
781 return -ENOMEM;
782 }
783
784 scsi_dh_data->scsi_dh = &alua_dh;
785 h = (struct alua_dh_data *) scsi_dh_data->buf;
786 h->tpgs = TPGS_MODE_UNINITIALIZED;
787 h->state = TPGS_STATE_OPTIMIZED;
788 h->group_id = -1;
789 h->rel_port = -1;
790 h->buff = h->inq;
791 h->bufflen = ALUA_INQUIRY_SIZE;
792 h->sdev = sdev;
793
794 err = alua_initialize(sdev, h);
795 if ((err != SCSI_DH_OK) && (err != SCSI_DH_DEV_OFFLINED))
796 goto failed;
797
798 if (!try_module_get(THIS_MODULE))
799 goto failed;
800
801 spin_lock_irqsave(sdev->request_queue->queue_lock, flags);
802 sdev->scsi_dh_data = scsi_dh_data;
803 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
804 sdev_printk(KERN_NOTICE, sdev, "%s: Attached\n", ALUA_DH_NAME);
805
806 return 0;
807
808failed:
809 kfree(scsi_dh_data);
810 sdev_printk(KERN_ERR, sdev, "%s: not attached\n", ALUA_DH_NAME);
811 return -EINVAL;
812}
813
814/*
815 * alua_bus_detach - Detach device handler
816 * @sdev: device to be detached from
817 */
818static void alua_bus_detach(struct scsi_device *sdev)
819{
820 struct scsi_dh_data *scsi_dh_data;
821 struct alua_dh_data *h;
822 unsigned long flags;
823
824 spin_lock_irqsave(sdev->request_queue->queue_lock, flags);
825 scsi_dh_data = sdev->scsi_dh_data;
826 sdev->scsi_dh_data = NULL;
827 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
828
829 h = (struct alua_dh_data *) scsi_dh_data->buf;
830 if (h->buff && h->inq != h->buff)
831 kfree(h->buff);
832 kfree(scsi_dh_data);
833 module_put(THIS_MODULE);
834 sdev_printk(KERN_NOTICE, sdev, "%s: Detached\n", ALUA_DH_NAME);
835}
836
837static int __init alua_init(void)
838{
839 int r;
840
841 r = scsi_register_device_handler(&alua_dh);
842 if (r != 0)
843 printk(KERN_ERR "%s: Failed to register scsi device handler",
844 ALUA_DH_NAME);
845 return r;
846}
847
848static void __exit alua_exit(void)
849{
850 scsi_unregister_device_handler(&alua_dh);
851}
852
853module_init(alua_init);
854module_exit(alua_exit);
855
856MODULE_DESCRIPTION("DM Multipath ALUA support");
857MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
858MODULE_LICENSE("GPL");
859MODULE_VERSION(ALUA_DH_VER);