xref: /openbmc/linux/drivers/scsi/device_handler/scsi_dh_alua.c (revision baa7eb025ab14f3cba2e35c0a8648f9c9f01d24f)
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 
57 struct 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 
75 static char print_alua_state(int);
76 static int alua_check_sense(struct scsi_device *, struct scsi_sense_hdr *);
77 
78 static 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 
85 static 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 
100 static 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  */
134 static 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);
163 done:
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  */
171 static 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);
200 done:
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  */
208 static 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);
239 done:
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  */
252 static 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_IO;
257 
258 	if (error || host_byte(req->errors) != DID_OK ||
259 			msg_byte(req->errors) != COMMAND_COMPLETE)
260 		goto done;
261 
262 	if (err == SCSI_DH_IO && h->senselen > 0) {
263 		err = scsi_normalize_sense(h->sense, SCSI_SENSE_BUFFERSIZE,
264 					   &sense_hdr);
265 		if (!err) {
266 			err = SCSI_DH_IO;
267 			goto done;
268 		}
269 		err = alua_check_sense(h->sdev, &sense_hdr);
270 		if (err == ADD_TO_MLQUEUE) {
271 			err = SCSI_DH_RETRY;
272 			goto done;
273 		}
274 		sdev_printk(KERN_INFO, h->sdev,
275 			    "%s: stpg sense code: %02x/%02x/%02x\n",
276 			    ALUA_DH_NAME, sense_hdr.sense_key,
277 			    sense_hdr.asc, sense_hdr.ascq);
278 		err = SCSI_DH_IO;
279 	}
280 	if (err == SCSI_DH_OK) {
281 		h->state = TPGS_STATE_OPTIMIZED;
282 		sdev_printk(KERN_INFO, h->sdev,
283 			    "%s: port group %02x switched to state %c\n",
284 			    ALUA_DH_NAME, h->group_id,
285 			    print_alua_state(h->state));
286 	}
287 done:
288 	blk_put_request(req);
289 	if (h->callback_fn) {
290 		h->callback_fn(h->callback_data, err);
291 		h->callback_fn = h->callback_data = NULL;
292 	}
293 	return;
294 }
295 
296 /*
297  * submit_stpg - Issue a SET TARGET GROUP STATES command
298  *
299  * Currently we're only setting the current target port group state
300  * to 'active/optimized' and let the array firmware figure out
301  * the states of the remaining groups.
302  */
303 static unsigned submit_stpg(struct alua_dh_data *h)
304 {
305 	struct request *rq;
306 	int err = SCSI_DH_RES_TEMP_UNAVAIL;
307 	int stpg_len = 8;
308 	struct scsi_device *sdev = h->sdev;
309 
310 	/* Prepare the data buffer */
311 	memset(h->buff, 0, stpg_len);
312 	h->buff[4] = TPGS_STATE_OPTIMIZED & 0x0f;
313 	h->buff[6] = (h->group_id >> 8) & 0xff;
314 	h->buff[7] = h->group_id & 0xff;
315 
316 	rq = get_alua_req(sdev, h->buff, stpg_len, WRITE);
317 	if (!rq)
318 		return SCSI_DH_RES_TEMP_UNAVAIL;
319 
320 	/* Prepare the command. */
321 	rq->cmd[0] = MAINTENANCE_OUT;
322 	rq->cmd[1] = MO_SET_TARGET_PGS;
323 	rq->cmd[6] = (stpg_len >> 24) & 0xff;
324 	rq->cmd[7] = (stpg_len >> 16) & 0xff;
325 	rq->cmd[8] = (stpg_len >>  8) & 0xff;
326 	rq->cmd[9] = stpg_len & 0xff;
327 	rq->cmd_len = COMMAND_SIZE(MAINTENANCE_OUT);
328 
329 	rq->sense = h->sense;
330 	memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
331 	rq->sense_len = h->senselen = 0;
332 	rq->end_io_data = h;
333 
334 	blk_execute_rq_nowait(rq->q, NULL, rq, 1, stpg_endio);
335 	return err;
336 }
337 
338 /*
339  * alua_std_inquiry - Evaluate standard INQUIRY command
340  * @sdev: device to be checked
341  *
342  * Just extract the TPGS setting to find out if ALUA
343  * is supported.
344  */
345 static int alua_std_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
346 {
347 	int err;
348 
349 	err = submit_std_inquiry(sdev, h);
350 
351 	if (err != SCSI_DH_OK)
352 		return err;
353 
354 	/* Check TPGS setting */
355 	h->tpgs = (h->inq[5] >> 4) & 0x3;
356 	switch (h->tpgs) {
357 	case TPGS_MODE_EXPLICIT|TPGS_MODE_IMPLICIT:
358 		sdev_printk(KERN_INFO, sdev,
359 			    "%s: supports implicit and explicit TPGS\n",
360 			    ALUA_DH_NAME);
361 		break;
362 	case TPGS_MODE_EXPLICIT:
363 		sdev_printk(KERN_INFO, sdev, "%s: supports explicit TPGS\n",
364 			    ALUA_DH_NAME);
365 		break;
366 	case TPGS_MODE_IMPLICIT:
367 		sdev_printk(KERN_INFO, sdev, "%s: supports implicit TPGS\n",
368 			    ALUA_DH_NAME);
369 		break;
370 	default:
371 		h->tpgs = TPGS_MODE_NONE;
372 		sdev_printk(KERN_INFO, sdev, "%s: not supported\n",
373 			    ALUA_DH_NAME);
374 		err = SCSI_DH_DEV_UNSUPP;
375 		break;
376 	}
377 
378 	return err;
379 }
380 
381 /*
382  * alua_vpd_inquiry - Evaluate INQUIRY vpd page 0x83
383  * @sdev: device to be checked
384  *
385  * Extract the relative target port and the target port group
386  * descriptor from the list of identificators.
387  */
388 static int alua_vpd_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
389 {
390 	int len;
391 	unsigned err;
392 	unsigned char *d;
393 
394  retry:
395 	err = submit_vpd_inquiry(sdev, h);
396 
397 	if (err != SCSI_DH_OK)
398 		return err;
399 
400 	/* Check if vpd page exceeds initial buffer */
401 	len = (h->buff[2] << 8) + h->buff[3] + 4;
402 	if (len > h->bufflen) {
403 		/* Resubmit with the correct length */
404 		if (realloc_buffer(h, len)) {
405 			sdev_printk(KERN_WARNING, sdev,
406 				    "%s: kmalloc buffer failed\n",
407 				    ALUA_DH_NAME);
408 			/* Temporary failure, bypass */
409 			return SCSI_DH_DEV_TEMP_BUSY;
410 		}
411 		goto retry;
412 	}
413 
414 	/*
415 	 * Now look for the correct descriptor.
416 	 */
417 	d = h->buff + 4;
418 	while (d < h->buff + len) {
419 		switch (d[1] & 0xf) {
420 		case 0x4:
421 			/* Relative target port */
422 			h->rel_port = (d[6] << 8) + d[7];
423 			break;
424 		case 0x5:
425 			/* Target port group */
426 			h->group_id = (d[6] << 8) + d[7];
427 			break;
428 		default:
429 			break;
430 		}
431 		d += d[3] + 4;
432 	}
433 
434 	if (h->group_id == -1) {
435 		/*
436 		 * Internal error; TPGS supported but required
437 		 * VPD identification descriptors not present.
438 		 * Disable ALUA support
439 		 */
440 		sdev_printk(KERN_INFO, sdev,
441 			    "%s: No target port descriptors found\n",
442 			    ALUA_DH_NAME);
443 		h->state = TPGS_STATE_OPTIMIZED;
444 		h->tpgs = TPGS_MODE_NONE;
445 		err = SCSI_DH_DEV_UNSUPP;
446 	} else {
447 		sdev_printk(KERN_INFO, sdev,
448 			    "%s: port group %02x rel port %02x\n",
449 			    ALUA_DH_NAME, h->group_id, h->rel_port);
450 	}
451 
452 	return err;
453 }
454 
455 static char print_alua_state(int state)
456 {
457 	switch (state) {
458 	case TPGS_STATE_OPTIMIZED:
459 		return 'A';
460 	case TPGS_STATE_NONOPTIMIZED:
461 		return 'N';
462 	case TPGS_STATE_STANDBY:
463 		return 'S';
464 	case TPGS_STATE_UNAVAILABLE:
465 		return 'U';
466 	case TPGS_STATE_LBA_DEPENDENT:
467 		return 'L';
468 	case TPGS_STATE_OFFLINE:
469 		return 'O';
470 	case TPGS_STATE_TRANSITIONING:
471 		return 'T';
472 	default:
473 		return 'X';
474 	}
475 }
476 
477 static int alua_check_sense(struct scsi_device *sdev,
478 			    struct scsi_sense_hdr *sense_hdr)
479 {
480 	switch (sense_hdr->sense_key) {
481 	case NOT_READY:
482 		if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a)
483 			/*
484 			 * LUN Not Accessible - ALUA state transition
485 			 */
486 			return ADD_TO_MLQUEUE;
487 		if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0b)
488 			/*
489 			 * LUN Not Accessible -- Target port in standby state
490 			 */
491 			return SUCCESS;
492 		if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0c)
493 			/*
494 			 * LUN Not Accessible -- Target port in unavailable state
495 			 */
496 			return SUCCESS;
497 		if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x12)
498 			/*
499 			 * LUN Not Ready -- Offline
500 			 */
501 			return SUCCESS;
502 		break;
503 	case UNIT_ATTENTION:
504 		if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x00)
505 			/*
506 			 * Power On, Reset, or Bus Device Reset, just retry.
507 			 */
508 			return ADD_TO_MLQUEUE;
509 		if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x06) {
510 			/*
511 			 * ALUA state changed
512 			 */
513 			return ADD_TO_MLQUEUE;
514 		}
515 		if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x07) {
516 			/*
517 			 * Implicit ALUA state transition failed
518 			 */
519 			return ADD_TO_MLQUEUE;
520 		}
521 		if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x0e) {
522 			/*
523 			 * REPORTED_LUNS_DATA_HAS_CHANGED is reported
524 			 * when switching controllers on targets like
525 			 * Intel Multi-Flex. We can just retry.
526 			 */
527 			return ADD_TO_MLQUEUE;
528 		}
529 
530 		break;
531 	}
532 
533 	return SCSI_RETURN_NOT_HANDLED;
534 }
535 
536 /*
537  * alua_rtpg - Evaluate REPORT TARGET GROUP STATES
538  * @sdev: the device to be evaluated.
539  *
540  * Evaluate the Target Port Group State.
541  * Returns SCSI_DH_DEV_OFFLINED if the path is
542  * found to be unuseable.
543  */
544 static int alua_rtpg(struct scsi_device *sdev, struct alua_dh_data *h)
545 {
546 	struct scsi_sense_hdr sense_hdr;
547 	int len, k, off, valid_states = 0;
548 	char *ucp;
549 	unsigned err;
550 	unsigned long expiry, interval = 10;
551 
552 	expiry = round_jiffies_up(jiffies + ALUA_FAILOVER_TIMEOUT);
553  retry:
554 	err = submit_rtpg(sdev, h);
555 
556 	if (err == SCSI_DH_IO && h->senselen > 0) {
557 		err = scsi_normalize_sense(h->sense, SCSI_SENSE_BUFFERSIZE,
558 					   &sense_hdr);
559 		if (!err)
560 			return SCSI_DH_IO;
561 
562 		err = alua_check_sense(sdev, &sense_hdr);
563 		if (err == ADD_TO_MLQUEUE && time_before(jiffies, expiry))
564 			goto retry;
565 		sdev_printk(KERN_INFO, sdev,
566 			    "%s: rtpg sense code %02x/%02x/%02x\n",
567 			    ALUA_DH_NAME, sense_hdr.sense_key,
568 			    sense_hdr.asc, sense_hdr.ascq);
569 		err = SCSI_DH_IO;
570 	}
571 	if (err != SCSI_DH_OK)
572 		return err;
573 
574 	len = (h->buff[0] << 24) + (h->buff[1] << 16) +
575 		(h->buff[2] << 8) + h->buff[3] + 4;
576 
577 	if (len > h->bufflen) {
578 		/* Resubmit with the correct length */
579 		if (realloc_buffer(h, len)) {
580 			sdev_printk(KERN_WARNING, sdev,
581 				    "%s: kmalloc buffer failed\n",__func__);
582 			/* Temporary failure, bypass */
583 			return SCSI_DH_DEV_TEMP_BUSY;
584 		}
585 		goto retry;
586 	}
587 
588 	for (k = 4, ucp = h->buff + 4; k < len; k += off, ucp += off) {
589 		if (h->group_id == (ucp[2] << 8) + ucp[3]) {
590 			h->state = ucp[0] & 0x0f;
591 			valid_states = ucp[1];
592 		}
593 		off = 8 + (ucp[7] * 4);
594 	}
595 
596 	sdev_printk(KERN_INFO, sdev,
597 		    "%s: port group %02x state %c supports %c%c%c%c%c%c%c\n",
598 		    ALUA_DH_NAME, h->group_id, print_alua_state(h->state),
599 		    valid_states&TPGS_SUPPORT_TRANSITION?'T':'t',
600 		    valid_states&TPGS_SUPPORT_OFFLINE?'O':'o',
601 		    valid_states&TPGS_SUPPORT_LBA_DEPENDENT?'L':'l',
602 		    valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u',
603 		    valid_states&TPGS_SUPPORT_STANDBY?'S':'s',
604 		    valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n',
605 		    valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a');
606 
607 	switch (h->state) {
608 	case TPGS_STATE_TRANSITIONING:
609 		if (time_before(jiffies, expiry)) {
610 			/* State transition, retry */
611 			interval *= 10;
612 			msleep(interval);
613 			goto retry;
614 		}
615 		/* Transitioning time exceeded, set port to standby */
616 		err = SCSI_DH_RETRY;
617 		h->state = TPGS_STATE_STANDBY;
618 		break;
619 	case TPGS_STATE_OFFLINE:
620 	case TPGS_STATE_UNAVAILABLE:
621 		/* Path unuseable for unavailable/offline */
622 		err = SCSI_DH_DEV_OFFLINED;
623 		break;
624 	default:
625 		/* Useable path if active */
626 		err = SCSI_DH_OK;
627 		break;
628 	}
629 	return err;
630 }
631 
632 /*
633  * alua_initialize - Initialize ALUA state
634  * @sdev: the device to be initialized
635  *
636  * For the prep_fn to work correctly we have
637  * to initialize the ALUA state for the device.
638  */
639 static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h)
640 {
641 	int err;
642 
643 	err = alua_std_inquiry(sdev, h);
644 	if (err != SCSI_DH_OK)
645 		goto out;
646 
647 	err = alua_vpd_inquiry(sdev, h);
648 	if (err != SCSI_DH_OK)
649 		goto out;
650 
651 	err = alua_rtpg(sdev, h);
652 	if (err != SCSI_DH_OK)
653 		goto out;
654 
655 out:
656 	return err;
657 }
658 
659 /*
660  * alua_activate - activate a path
661  * @sdev: device on the path to be activated
662  *
663  * We're currently switching the port group to be activated only and
664  * let the array figure out the rest.
665  * There may be other arrays which require us to switch all port groups
666  * based on a certain policy. But until we actually encounter them it
667  * should be okay.
668  */
669 static int alua_activate(struct scsi_device *sdev,
670 			activate_complete fn, void *data)
671 {
672 	struct alua_dh_data *h = get_alua_data(sdev);
673 	int err = SCSI_DH_OK;
674 
675 	if (h->group_id != -1) {
676 		err = alua_rtpg(sdev, h);
677 		if (err != SCSI_DH_OK)
678 			goto out;
679 	}
680 
681 	if (h->tpgs & TPGS_MODE_EXPLICIT &&
682 	    h->state != TPGS_STATE_OPTIMIZED &&
683 	    h->state != TPGS_STATE_LBA_DEPENDENT) {
684 		h->callback_fn = fn;
685 		h->callback_data = data;
686 		err = submit_stpg(h);
687 		if (err == SCSI_DH_OK)
688 			return 0;
689 		h->callback_fn = h->callback_data = NULL;
690 	}
691 
692 out:
693 	if (fn)
694 		fn(data, err);
695 	return 0;
696 }
697 
698 /*
699  * alua_prep_fn - request callback
700  *
701  * Fail I/O to all paths not in state
702  * active/optimized or active/non-optimized.
703  */
704 static int alua_prep_fn(struct scsi_device *sdev, struct request *req)
705 {
706 	struct alua_dh_data *h = get_alua_data(sdev);
707 	int ret = BLKPREP_OK;
708 
709 	if (h->state == TPGS_STATE_TRANSITIONING)
710 		ret = BLKPREP_DEFER;
711 	else if (h->state != TPGS_STATE_OPTIMIZED &&
712 		 h->state != TPGS_STATE_NONOPTIMIZED &&
713 		 h->state != TPGS_STATE_LBA_DEPENDENT) {
714 		ret = BLKPREP_KILL;
715 		req->cmd_flags |= REQ_QUIET;
716 	}
717 	return ret;
718 
719 }
720 
721 static const struct scsi_dh_devlist alua_dev_list[] = {
722 	{"HP", "MSA VOLUME" },
723 	{"HP", "HSV101" },
724 	{"HP", "HSV111" },
725 	{"HP", "HSV200" },
726 	{"HP", "HSV210" },
727 	{"HP", "HSV300" },
728 	{"IBM", "2107900" },
729 	{"IBM", "2145" },
730 	{"Pillar", "Axiom" },
731 	{"Intel", "Multi-Flex"},
732 	{"NETAPP", "LUN"},
733 	{"AIX", "NVDISK"},
734 	{NULL, NULL}
735 };
736 
737 static int alua_bus_attach(struct scsi_device *sdev);
738 static void alua_bus_detach(struct scsi_device *sdev);
739 
740 static struct scsi_device_handler alua_dh = {
741 	.name = ALUA_DH_NAME,
742 	.module = THIS_MODULE,
743 	.devlist = alua_dev_list,
744 	.attach = alua_bus_attach,
745 	.detach = alua_bus_detach,
746 	.prep_fn = alua_prep_fn,
747 	.check_sense = alua_check_sense,
748 	.activate = alua_activate,
749 };
750 
751 /*
752  * alua_bus_attach - Attach device handler
753  * @sdev: device to be attached to
754  */
755 static int alua_bus_attach(struct scsi_device *sdev)
756 {
757 	struct scsi_dh_data *scsi_dh_data;
758 	struct alua_dh_data *h;
759 	unsigned long flags;
760 	int err = SCSI_DH_OK;
761 
762 	scsi_dh_data = kzalloc(sizeof(struct scsi_device_handler *)
763 			       + sizeof(*h) , GFP_KERNEL);
764 	if (!scsi_dh_data) {
765 		sdev_printk(KERN_ERR, sdev, "%s: Attach failed\n",
766 			    ALUA_DH_NAME);
767 		return -ENOMEM;
768 	}
769 
770 	scsi_dh_data->scsi_dh = &alua_dh;
771 	h = (struct alua_dh_data *) scsi_dh_data->buf;
772 	h->tpgs = TPGS_MODE_UNINITIALIZED;
773 	h->state = TPGS_STATE_OPTIMIZED;
774 	h->group_id = -1;
775 	h->rel_port = -1;
776 	h->buff = h->inq;
777 	h->bufflen = ALUA_INQUIRY_SIZE;
778 	h->sdev = sdev;
779 
780 	err = alua_initialize(sdev, h);
781 	if (err != SCSI_DH_OK)
782 		goto failed;
783 
784 	if (!try_module_get(THIS_MODULE))
785 		goto failed;
786 
787 	spin_lock_irqsave(sdev->request_queue->queue_lock, flags);
788 	sdev->scsi_dh_data = scsi_dh_data;
789 	spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
790 
791 	return 0;
792 
793 failed:
794 	kfree(scsi_dh_data);
795 	sdev_printk(KERN_ERR, sdev, "%s: not attached\n", ALUA_DH_NAME);
796 	return -EINVAL;
797 }
798 
799 /*
800  * alua_bus_detach - Detach device handler
801  * @sdev: device to be detached from
802  */
803 static void alua_bus_detach(struct scsi_device *sdev)
804 {
805 	struct scsi_dh_data *scsi_dh_data;
806 	struct alua_dh_data *h;
807 	unsigned long flags;
808 
809 	spin_lock_irqsave(sdev->request_queue->queue_lock, flags);
810 	scsi_dh_data = sdev->scsi_dh_data;
811 	sdev->scsi_dh_data = NULL;
812 	spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
813 
814 	h = (struct alua_dh_data *) scsi_dh_data->buf;
815 	if (h->buff && h->inq != h->buff)
816 		kfree(h->buff);
817 	kfree(scsi_dh_data);
818 	module_put(THIS_MODULE);
819 	sdev_printk(KERN_NOTICE, sdev, "%s: Detached\n", ALUA_DH_NAME);
820 }
821 
822 static int __init alua_init(void)
823 {
824 	int r;
825 
826 	r = scsi_register_device_handler(&alua_dh);
827 	if (r != 0)
828 		printk(KERN_ERR "%s: Failed to register scsi device handler",
829 			ALUA_DH_NAME);
830 	return r;
831 }
832 
833 static void __exit alua_exit(void)
834 {
835 	scsi_unregister_device_handler(&alua_dh);
836 }
837 
838 module_init(alua_init);
839 module_exit(alua_exit);
840 
841 MODULE_DESCRIPTION("DM Multipath ALUA support");
842 MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
843 MODULE_LICENSE("GPL");
844 MODULE_VERSION(ALUA_DH_VER);
845