1 /*
2  * Generic SCSI-3 ALUA SCSI Device Handler
3  *
4  * Copyright (C) 2007, 2008 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 <scsi/scsi.h>
23 #include <scsi/scsi_eh.h>
24 #include <scsi/scsi_dh.h>
25 
26 #define ALUA_DH_NAME "alua"
27 #define ALUA_DH_VER "1.2"
28 
29 #define TPGS_STATE_OPTIMIZED		0x0
30 #define TPGS_STATE_NONOPTIMIZED		0x1
31 #define TPGS_STATE_STANDBY		0x2
32 #define TPGS_STATE_UNAVAILABLE		0x3
33 #define TPGS_STATE_OFFLINE		0xe
34 #define TPGS_STATE_TRANSITIONING	0xf
35 
36 #define TPGS_SUPPORT_NONE		0x00
37 #define TPGS_SUPPORT_OPTIMIZED		0x01
38 #define TPGS_SUPPORT_NONOPTIMIZED	0x02
39 #define TPGS_SUPPORT_STANDBY		0x04
40 #define TPGS_SUPPORT_UNAVAILABLE	0x08
41 #define TPGS_SUPPORT_OFFLINE		0x40
42 #define TPGS_SUPPORT_TRANSITION		0x80
43 
44 #define TPGS_MODE_UNINITIALIZED		 -1
45 #define TPGS_MODE_NONE			0x0
46 #define TPGS_MODE_IMPLICIT		0x1
47 #define TPGS_MODE_EXPLICIT		0x2
48 
49 #define ALUA_INQUIRY_SIZE		36
50 #define ALUA_FAILOVER_TIMEOUT		(60 * HZ)
51 #define ALUA_FAILOVER_RETRIES		5
52 
53 struct alua_dh_data {
54 	int			group_id;
55 	int			rel_port;
56 	int			tpgs;
57 	int			state;
58 	unsigned char		inq[ALUA_INQUIRY_SIZE];
59 	unsigned char		*buff;
60 	int			bufflen;
61 	unsigned char		sense[SCSI_SENSE_BUFFERSIZE];
62 	int			senselen;
63 	struct scsi_device	*sdev;
64 	activate_complete	callback_fn;
65 	void			*callback_data;
66 };
67 
68 #define ALUA_POLICY_SWITCH_CURRENT	0
69 #define ALUA_POLICY_SWITCH_ALL		1
70 
71 static char print_alua_state(int);
72 static int alua_check_sense(struct scsi_device *, struct scsi_sense_hdr *);
73 
74 static inline struct alua_dh_data *get_alua_data(struct scsi_device *sdev)
75 {
76 	struct scsi_dh_data *scsi_dh_data = sdev->scsi_dh_data;
77 	BUG_ON(scsi_dh_data == NULL);
78 	return ((struct alua_dh_data *) scsi_dh_data->buf);
79 }
80 
81 static int realloc_buffer(struct alua_dh_data *h, unsigned len)
82 {
83 	if (h->buff && h->buff != h->inq)
84 		kfree(h->buff);
85 
86 	h->buff = kmalloc(len, GFP_NOIO);
87 	if (!h->buff) {
88 		h->buff = h->inq;
89 		h->bufflen = ALUA_INQUIRY_SIZE;
90 		return 1;
91 	}
92 	h->bufflen = len;
93 	return 0;
94 }
95 
96 static struct request *get_alua_req(struct scsi_device *sdev,
97 				    void *buffer, unsigned buflen, int rw)
98 {
99 	struct request *rq;
100 	struct request_queue *q = sdev->request_queue;
101 
102 	rq = blk_get_request(q, rw, GFP_NOIO);
103 
104 	if (!rq) {
105 		sdev_printk(KERN_INFO, sdev,
106 			    "%s: blk_get_request failed\n", __func__);
107 		return NULL;
108 	}
109 
110 	if (buflen && blk_rq_map_kern(q, rq, buffer, buflen, GFP_NOIO)) {
111 		blk_put_request(rq);
112 		sdev_printk(KERN_INFO, sdev,
113 			    "%s: blk_rq_map_kern failed\n", __func__);
114 		return NULL;
115 	}
116 
117 	rq->cmd_type = REQ_TYPE_BLOCK_PC;
118 	rq->cmd_flags |= REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
119 			 REQ_FAILFAST_DRIVER;
120 	rq->retries = ALUA_FAILOVER_RETRIES;
121 	rq->timeout = ALUA_FAILOVER_TIMEOUT;
122 
123 	return rq;
124 }
125 
126 /*
127  * submit_std_inquiry - Issue a standard INQUIRY command
128  * @sdev: sdev the command should be send to
129  */
130 static int submit_std_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
131 {
132 	struct request *rq;
133 	int err = SCSI_DH_RES_TEMP_UNAVAIL;
134 
135 	rq = get_alua_req(sdev, h->inq, ALUA_INQUIRY_SIZE, READ);
136 	if (!rq)
137 		goto done;
138 
139 	/* Prepare the command. */
140 	rq->cmd[0] = INQUIRY;
141 	rq->cmd[1] = 0;
142 	rq->cmd[2] = 0;
143 	rq->cmd[4] = ALUA_INQUIRY_SIZE;
144 	rq->cmd_len = COMMAND_SIZE(INQUIRY);
145 
146 	rq->sense = h->sense;
147 	memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
148 	rq->sense_len = h->senselen = 0;
149 
150 	err = blk_execute_rq(rq->q, NULL, rq, 1);
151 	if (err == -EIO) {
152 		sdev_printk(KERN_INFO, sdev,
153 			    "%s: std inquiry failed with %x\n",
154 			    ALUA_DH_NAME, rq->errors);
155 		h->senselen = rq->sense_len;
156 		err = SCSI_DH_IO;
157 	}
158 	blk_put_request(rq);
159 done:
160 	return err;
161 }
162 
163 /*
164  * submit_vpd_inquiry - Issue an INQUIRY VPD page 0x83 command
165  * @sdev: sdev the command should be sent to
166  */
167 static int submit_vpd_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
168 {
169 	struct request *rq;
170 	int err = SCSI_DH_RES_TEMP_UNAVAIL;
171 
172 	rq = get_alua_req(sdev, h->buff, h->bufflen, READ);
173 	if (!rq)
174 		goto done;
175 
176 	/* Prepare the command. */
177 	rq->cmd[0] = INQUIRY;
178 	rq->cmd[1] = 1;
179 	rq->cmd[2] = 0x83;
180 	rq->cmd[4] = h->bufflen;
181 	rq->cmd_len = COMMAND_SIZE(INQUIRY);
182 
183 	rq->sense = h->sense;
184 	memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
185 	rq->sense_len = h->senselen = 0;
186 
187 	err = blk_execute_rq(rq->q, NULL, rq, 1);
188 	if (err == -EIO) {
189 		sdev_printk(KERN_INFO, sdev,
190 			    "%s: evpd inquiry failed with %x\n",
191 			    ALUA_DH_NAME, rq->errors);
192 		h->senselen = rq->sense_len;
193 		err = SCSI_DH_IO;
194 	}
195 	blk_put_request(rq);
196 done:
197 	return err;
198 }
199 
200 /*
201  * submit_rtpg - Issue a REPORT TARGET GROUP STATES command
202  * @sdev: sdev the command should be sent to
203  */
204 static unsigned submit_rtpg(struct scsi_device *sdev, struct alua_dh_data *h)
205 {
206 	struct request *rq;
207 	int err = SCSI_DH_RES_TEMP_UNAVAIL;
208 
209 	rq = get_alua_req(sdev, h->buff, h->bufflen, READ);
210 	if (!rq)
211 		goto done;
212 
213 	/* Prepare the command. */
214 	rq->cmd[0] = MAINTENANCE_IN;
215 	rq->cmd[1] = MI_REPORT_TARGET_PGS;
216 	rq->cmd[6] = (h->bufflen >> 24) & 0xff;
217 	rq->cmd[7] = (h->bufflen >> 16) & 0xff;
218 	rq->cmd[8] = (h->bufflen >>  8) & 0xff;
219 	rq->cmd[9] = h->bufflen & 0xff;
220 	rq->cmd_len = COMMAND_SIZE(MAINTENANCE_IN);
221 
222 	rq->sense = h->sense;
223 	memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
224 	rq->sense_len = h->senselen = 0;
225 
226 	err = blk_execute_rq(rq->q, NULL, rq, 1);
227 	if (err == -EIO) {
228 		sdev_printk(KERN_INFO, sdev,
229 			    "%s: rtpg failed with %x\n",
230 			    ALUA_DH_NAME, rq->errors);
231 		h->senselen = rq->sense_len;
232 		err = SCSI_DH_IO;
233 	}
234 	blk_put_request(rq);
235 done:
236 	return err;
237 }
238 
239 /*
240  * alua_stpg - Evaluate SET TARGET GROUP STATES
241  * @sdev: the device to be evaluated
242  * @state: the new target group state
243  *
244  * Send a SET TARGET GROUP STATES command to the device.
245  * We only have to test here if we should resubmit the command;
246  * any other error is assumed as a failure.
247  */
248 static void stpg_endio(struct request *req, int error)
249 {
250 	struct alua_dh_data *h = req->end_io_data;
251 	struct scsi_sense_hdr sense_hdr;
252 	unsigned err = SCSI_DH_IO;
253 
254 	if (error || host_byte(req->errors) != DID_OK ||
255 			msg_byte(req->errors) != COMMAND_COMPLETE)
256 		goto done;
257 
258 	if (err == SCSI_DH_IO && h->senselen > 0) {
259 		err = scsi_normalize_sense(h->sense, SCSI_SENSE_BUFFERSIZE,
260 					   &sense_hdr);
261 		if (!err) {
262 			err = SCSI_DH_IO;
263 			goto done;
264 		}
265 		err = alua_check_sense(h->sdev, &sense_hdr);
266 		if (err == ADD_TO_MLQUEUE) {
267 			err = SCSI_DH_RETRY;
268 			goto done;
269 		}
270 		sdev_printk(KERN_INFO, h->sdev,
271 			    "%s: stpg sense code: %02x/%02x/%02x\n",
272 			    ALUA_DH_NAME, sense_hdr.sense_key,
273 			    sense_hdr.asc, sense_hdr.ascq);
274 		err = SCSI_DH_IO;
275 	}
276 	if (err == SCSI_DH_OK) {
277 		h->state = TPGS_STATE_OPTIMIZED;
278 		sdev_printk(KERN_INFO, h->sdev,
279 			    "%s: port group %02x switched to state %c\n",
280 			    ALUA_DH_NAME, h->group_id,
281 			    print_alua_state(h->state));
282 	}
283 done:
284 	blk_put_request(req);
285 	if (h->callback_fn) {
286 		h->callback_fn(h->callback_data, err);
287 		h->callback_fn = h->callback_data = NULL;
288 	}
289 	return;
290 }
291 
292 /*
293  * submit_stpg - Issue a SET TARGET GROUP STATES command
294  *
295  * Currently we're only setting the current target port group state
296  * to 'active/optimized' and let the array firmware figure out
297  * the states of the remaining groups.
298  */
299 static unsigned submit_stpg(struct alua_dh_data *h)
300 {
301 	struct request *rq;
302 	int err = SCSI_DH_RES_TEMP_UNAVAIL;
303 	int stpg_len = 8;
304 	struct scsi_device *sdev = h->sdev;
305 
306 	/* Prepare the data buffer */
307 	memset(h->buff, 0, stpg_len);
308 	h->buff[4] = TPGS_STATE_OPTIMIZED & 0x0f;
309 	h->buff[6] = (h->group_id >> 8) & 0xff;
310 	h->buff[7] = h->group_id & 0xff;
311 
312 	rq = get_alua_req(sdev, h->buff, stpg_len, WRITE);
313 	if (!rq)
314 		return SCSI_DH_RES_TEMP_UNAVAIL;
315 
316 	/* Prepare the command. */
317 	rq->cmd[0] = MAINTENANCE_OUT;
318 	rq->cmd[1] = MO_SET_TARGET_PGS;
319 	rq->cmd[6] = (stpg_len >> 24) & 0xff;
320 	rq->cmd[7] = (stpg_len >> 16) & 0xff;
321 	rq->cmd[8] = (stpg_len >>  8) & 0xff;
322 	rq->cmd[9] = stpg_len & 0xff;
323 	rq->cmd_len = COMMAND_SIZE(MAINTENANCE_OUT);
324 
325 	rq->sense = h->sense;
326 	memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
327 	rq->sense_len = h->senselen = 0;
328 	rq->end_io_data = h;
329 
330 	blk_execute_rq_nowait(rq->q, NULL, rq, 1, stpg_endio);
331 	return err;
332 }
333 
334 /*
335  * alua_std_inquiry - Evaluate standard INQUIRY command
336  * @sdev: device to be checked
337  *
338  * Just extract the TPGS setting to find out if ALUA
339  * is supported.
340  */
341 static int alua_std_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
342 {
343 	int err;
344 
345 	err = submit_std_inquiry(sdev, h);
346 
347 	if (err != SCSI_DH_OK)
348 		return err;
349 
350 	/* Check TPGS setting */
351 	h->tpgs = (h->inq[5] >> 4) & 0x3;
352 	switch (h->tpgs) {
353 	case TPGS_MODE_EXPLICIT|TPGS_MODE_IMPLICIT:
354 		sdev_printk(KERN_INFO, sdev,
355 			    "%s: supports implicit and explicit TPGS\n",
356 			    ALUA_DH_NAME);
357 		break;
358 	case TPGS_MODE_EXPLICIT:
359 		sdev_printk(KERN_INFO, sdev, "%s: supports explicit TPGS\n",
360 			    ALUA_DH_NAME);
361 		break;
362 	case TPGS_MODE_IMPLICIT:
363 		sdev_printk(KERN_INFO, sdev, "%s: supports implicit TPGS\n",
364 			    ALUA_DH_NAME);
365 		break;
366 	default:
367 		h->tpgs = TPGS_MODE_NONE;
368 		sdev_printk(KERN_INFO, sdev, "%s: not supported\n",
369 			    ALUA_DH_NAME);
370 		err = SCSI_DH_DEV_UNSUPP;
371 		break;
372 	}
373 
374 	return err;
375 }
376 
377 /*
378  * alua_vpd_inquiry - Evaluate INQUIRY vpd page 0x83
379  * @sdev: device to be checked
380  *
381  * Extract the relative target port and the target port group
382  * descriptor from the list of identificators.
383  */
384 static int alua_vpd_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
385 {
386 	int len;
387 	unsigned err;
388 	unsigned char *d;
389 
390  retry:
391 	err = submit_vpd_inquiry(sdev, h);
392 
393 	if (err != SCSI_DH_OK)
394 		return err;
395 
396 	/* Check if vpd page exceeds initial buffer */
397 	len = (h->buff[2] << 8) + h->buff[3] + 4;
398 	if (len > h->bufflen) {
399 		/* Resubmit with the correct length */
400 		if (realloc_buffer(h, len)) {
401 			sdev_printk(KERN_WARNING, sdev,
402 				    "%s: kmalloc buffer failed\n",
403 				    ALUA_DH_NAME);
404 			/* Temporary failure, bypass */
405 			return SCSI_DH_DEV_TEMP_BUSY;
406 		}
407 		goto retry;
408 	}
409 
410 	/*
411 	 * Now look for the correct descriptor.
412 	 */
413 	d = h->buff + 4;
414 	while (d < h->buff + len) {
415 		switch (d[1] & 0xf) {
416 		case 0x4:
417 			/* Relative target port */
418 			h->rel_port = (d[6] << 8) + d[7];
419 			break;
420 		case 0x5:
421 			/* Target port group */
422 			h->group_id = (d[6] << 8) + d[7];
423 			break;
424 		default:
425 			break;
426 		}
427 		d += d[3] + 4;
428 	}
429 
430 	if (h->group_id == -1) {
431 		/*
432 		 * Internal error; TPGS supported but required
433 		 * VPD identification descriptors not present.
434 		 * Disable ALUA support
435 		 */
436 		sdev_printk(KERN_INFO, sdev,
437 			    "%s: No target port descriptors found\n",
438 			    ALUA_DH_NAME);
439 		h->state = TPGS_STATE_OPTIMIZED;
440 		h->tpgs = TPGS_MODE_NONE;
441 		err = SCSI_DH_DEV_UNSUPP;
442 	} else {
443 		sdev_printk(KERN_INFO, sdev,
444 			    "%s: port group %02x rel port %02x\n",
445 			    ALUA_DH_NAME, h->group_id, h->rel_port);
446 	}
447 
448 	return err;
449 }
450 
451 static char print_alua_state(int state)
452 {
453 	switch (state) {
454 	case TPGS_STATE_OPTIMIZED:
455 		return 'A';
456 	case TPGS_STATE_NONOPTIMIZED:
457 		return 'N';
458 	case TPGS_STATE_STANDBY:
459 		return 'S';
460 	case TPGS_STATE_UNAVAILABLE:
461 		return 'U';
462 	case TPGS_STATE_OFFLINE:
463 		return 'O';
464 	case TPGS_STATE_TRANSITIONING:
465 		return 'T';
466 	default:
467 		return 'X';
468 	}
469 }
470 
471 static int alua_check_sense(struct scsi_device *sdev,
472 			    struct scsi_sense_hdr *sense_hdr)
473 {
474 	switch (sense_hdr->sense_key) {
475 	case NOT_READY:
476 		if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a)
477 			/*
478 			 * LUN Not Accessible - ALUA state transition
479 			 */
480 			return ADD_TO_MLQUEUE;
481 		if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0b)
482 			/*
483 			 * LUN Not Accessible -- Target port in standby state
484 			 */
485 			return SUCCESS;
486 		if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0c)
487 			/*
488 			 * LUN Not Accessible -- Target port in unavailable state
489 			 */
490 			return SUCCESS;
491 		if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x12)
492 			/*
493 			 * LUN Not Ready -- Offline
494 			 */
495 			return SUCCESS;
496 		break;
497 	case UNIT_ATTENTION:
498 		if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x00)
499 			/*
500 			 * Power On, Reset, or Bus Device Reset, just retry.
501 			 */
502 			return ADD_TO_MLQUEUE;
503 		if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x06) {
504 			/*
505 			 * ALUA state changed
506 			 */
507 			return ADD_TO_MLQUEUE;
508 		}
509 		if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x07) {
510 			/*
511 			 * Implicit ALUA state transition failed
512 			 */
513 			return ADD_TO_MLQUEUE;
514 		}
515 		if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x0e) {
516 			/*
517 			 * REPORTED_LUNS_DATA_HAS_CHANGED is reported
518 			 * when switching controllers on targets like
519 			 * Intel Multi-Flex. We can just retry.
520 			 */
521 			return ADD_TO_MLQUEUE;
522 		}
523 
524 		break;
525 	}
526 
527 	return SCSI_RETURN_NOT_HANDLED;
528 }
529 
530 /*
531  * alua_rtpg - Evaluate REPORT TARGET GROUP STATES
532  * @sdev: the device to be evaluated.
533  *
534  * Evaluate the Target Port Group State.
535  * Returns SCSI_DH_DEV_OFFLINED if the path is
536  * found to be unuseable.
537  */
538 static int alua_rtpg(struct scsi_device *sdev, struct alua_dh_data *h)
539 {
540 	struct scsi_sense_hdr sense_hdr;
541 	int len, k, off, valid_states = 0;
542 	char *ucp;
543 	unsigned err;
544 
545  retry:
546 	err = submit_rtpg(sdev, h);
547 
548 	if (err == SCSI_DH_IO && h->senselen > 0) {
549 		err = scsi_normalize_sense(h->sense, SCSI_SENSE_BUFFERSIZE,
550 					   &sense_hdr);
551 		if (!err)
552 			return SCSI_DH_IO;
553 
554 		err = alua_check_sense(sdev, &sense_hdr);
555 		if (err == ADD_TO_MLQUEUE)
556 			goto retry;
557 		sdev_printk(KERN_INFO, sdev,
558 			    "%s: rtpg sense code %02x/%02x/%02x\n",
559 			    ALUA_DH_NAME, sense_hdr.sense_key,
560 			    sense_hdr.asc, sense_hdr.ascq);
561 		err = SCSI_DH_IO;
562 	}
563 	if (err != SCSI_DH_OK)
564 		return err;
565 
566 	len = (h->buff[0] << 24) + (h->buff[1] << 16) +
567 		(h->buff[2] << 8) + h->buff[3] + 4;
568 
569 	if (len > h->bufflen) {
570 		/* Resubmit with the correct length */
571 		if (realloc_buffer(h, len)) {
572 			sdev_printk(KERN_WARNING, sdev,
573 				    "%s: kmalloc buffer failed\n",__func__);
574 			/* Temporary failure, bypass */
575 			return SCSI_DH_DEV_TEMP_BUSY;
576 		}
577 		goto retry;
578 	}
579 
580 	for (k = 4, ucp = h->buff + 4; k < len; k += off, ucp += off) {
581 		if (h->group_id == (ucp[2] << 8) + ucp[3]) {
582 			h->state = ucp[0] & 0x0f;
583 			valid_states = ucp[1];
584 		}
585 		off = 8 + (ucp[7] * 4);
586 	}
587 
588 	sdev_printk(KERN_INFO, sdev,
589 		    "%s: port group %02x state %c supports %c%c%c%c%c%c\n",
590 		    ALUA_DH_NAME, h->group_id, print_alua_state(h->state),
591 		    valid_states&TPGS_SUPPORT_TRANSITION?'T':'t',
592 		    valid_states&TPGS_SUPPORT_OFFLINE?'O':'o',
593 		    valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u',
594 		    valid_states&TPGS_SUPPORT_STANDBY?'S':'s',
595 		    valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n',
596 		    valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a');
597 
598 	if (h->tpgs & TPGS_MODE_EXPLICIT) {
599 		switch (h->state) {
600 		case TPGS_STATE_TRANSITIONING:
601 			/* State transition, retry */
602 			goto retry;
603 			break;
604 		case TPGS_STATE_OFFLINE:
605 			/* Path is offline, fail */
606 			err = SCSI_DH_DEV_OFFLINED;
607 			break;
608 		default:
609 			break;
610 		}
611 	} else {
612 		/* Only Implicit ALUA support */
613 		if (h->state == TPGS_STATE_OPTIMIZED ||
614 		    h->state == TPGS_STATE_NONOPTIMIZED ||
615 		    h->state == TPGS_STATE_STANDBY)
616 			/* Useable path if active */
617 			err = SCSI_DH_OK;
618 		else
619 			/* Path unuseable for unavailable/offline */
620 			err = SCSI_DH_DEV_OFFLINED;
621 	}
622 	return err;
623 }
624 
625 /*
626  * alua_initialize - Initialize ALUA state
627  * @sdev: the device to be initialized
628  *
629  * For the prep_fn to work correctly we have
630  * to initialize the ALUA state for the device.
631  */
632 static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h)
633 {
634 	int err;
635 
636 	err = alua_std_inquiry(sdev, h);
637 	if (err != SCSI_DH_OK)
638 		goto out;
639 
640 	err = alua_vpd_inquiry(sdev, h);
641 	if (err != SCSI_DH_OK)
642 		goto out;
643 
644 	err = alua_rtpg(sdev, h);
645 	if (err != SCSI_DH_OK)
646 		goto out;
647 
648 out:
649 	return err;
650 }
651 
652 /*
653  * alua_activate - activate a path
654  * @sdev: device on the path to be activated
655  *
656  * We're currently switching the port group to be activated only and
657  * let the array figure out the rest.
658  * There may be other arrays which require us to switch all port groups
659  * based on a certain policy. But until we actually encounter them it
660  * should be okay.
661  */
662 static int alua_activate(struct scsi_device *sdev,
663 			activate_complete fn, void *data)
664 {
665 	struct alua_dh_data *h = get_alua_data(sdev);
666 	int err = SCSI_DH_OK;
667 
668 	if (h->group_id != -1) {
669 		err = alua_rtpg(sdev, h);
670 		if (err != SCSI_DH_OK)
671 			goto out;
672 	}
673 
674 	if (h->tpgs & TPGS_MODE_EXPLICIT && h->state != TPGS_STATE_OPTIMIZED) {
675 		h->callback_fn = fn;
676 		h->callback_data = data;
677 		err = submit_stpg(h);
678 		if (err == SCSI_DH_OK)
679 			return 0;
680 		h->callback_fn = h->callback_data = NULL;
681 	}
682 
683 out:
684 	if (fn)
685 		fn(data, err);
686 	return 0;
687 }
688 
689 /*
690  * alua_prep_fn - request callback
691  *
692  * Fail I/O to all paths not in state
693  * active/optimized or active/non-optimized.
694  */
695 static int alua_prep_fn(struct scsi_device *sdev, struct request *req)
696 {
697 	struct alua_dh_data *h = get_alua_data(sdev);
698 	int ret = BLKPREP_OK;
699 
700 	if (h->state != TPGS_STATE_OPTIMIZED &&
701 	    h->state != TPGS_STATE_NONOPTIMIZED) {
702 		ret = BLKPREP_KILL;
703 		req->cmd_flags |= REQ_QUIET;
704 	}
705 	return ret;
706 
707 }
708 
709 static const struct scsi_dh_devlist alua_dev_list[] = {
710 	{"HP", "MSA VOLUME" },
711 	{"HP", "HSV101" },
712 	{"HP", "HSV111" },
713 	{"HP", "HSV200" },
714 	{"HP", "HSV210" },
715 	{"HP", "HSV300" },
716 	{"IBM", "2107900" },
717 	{"IBM", "2145" },
718 	{"Pillar", "Axiom" },
719 	{"Intel", "Multi-Flex"},
720 	{NULL, NULL}
721 };
722 
723 static int alua_bus_attach(struct scsi_device *sdev);
724 static void alua_bus_detach(struct scsi_device *sdev);
725 
726 static struct scsi_device_handler alua_dh = {
727 	.name = ALUA_DH_NAME,
728 	.module = THIS_MODULE,
729 	.devlist = alua_dev_list,
730 	.attach = alua_bus_attach,
731 	.detach = alua_bus_detach,
732 	.prep_fn = alua_prep_fn,
733 	.check_sense = alua_check_sense,
734 	.activate = alua_activate,
735 };
736 
737 /*
738  * alua_bus_attach - Attach device handler
739  * @sdev: device to be attached to
740  */
741 static int alua_bus_attach(struct scsi_device *sdev)
742 {
743 	struct scsi_dh_data *scsi_dh_data;
744 	struct alua_dh_data *h;
745 	unsigned long flags;
746 	int err = SCSI_DH_OK;
747 
748 	scsi_dh_data = kzalloc(sizeof(struct scsi_device_handler *)
749 			       + sizeof(*h) , GFP_KERNEL);
750 	if (!scsi_dh_data) {
751 		sdev_printk(KERN_ERR, sdev, "%s: Attach failed\n",
752 			    ALUA_DH_NAME);
753 		return -ENOMEM;
754 	}
755 
756 	scsi_dh_data->scsi_dh = &alua_dh;
757 	h = (struct alua_dh_data *) scsi_dh_data->buf;
758 	h->tpgs = TPGS_MODE_UNINITIALIZED;
759 	h->state = TPGS_STATE_OPTIMIZED;
760 	h->group_id = -1;
761 	h->rel_port = -1;
762 	h->buff = h->inq;
763 	h->bufflen = ALUA_INQUIRY_SIZE;
764 	h->sdev = sdev;
765 
766 	err = alua_initialize(sdev, h);
767 	if (err != SCSI_DH_OK)
768 		goto failed;
769 
770 	if (!try_module_get(THIS_MODULE))
771 		goto failed;
772 
773 	spin_lock_irqsave(sdev->request_queue->queue_lock, flags);
774 	sdev->scsi_dh_data = scsi_dh_data;
775 	spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
776 
777 	return 0;
778 
779 failed:
780 	kfree(scsi_dh_data);
781 	sdev_printk(KERN_ERR, sdev, "%s: not attached\n", ALUA_DH_NAME);
782 	return -EINVAL;
783 }
784 
785 /*
786  * alua_bus_detach - Detach device handler
787  * @sdev: device to be detached from
788  */
789 static void alua_bus_detach(struct scsi_device *sdev)
790 {
791 	struct scsi_dh_data *scsi_dh_data;
792 	struct alua_dh_data *h;
793 	unsigned long flags;
794 
795 	spin_lock_irqsave(sdev->request_queue->queue_lock, flags);
796 	scsi_dh_data = sdev->scsi_dh_data;
797 	sdev->scsi_dh_data = NULL;
798 	spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
799 
800 	h = (struct alua_dh_data *) scsi_dh_data->buf;
801 	if (h->buff && h->inq != h->buff)
802 		kfree(h->buff);
803 	kfree(scsi_dh_data);
804 	module_put(THIS_MODULE);
805 	sdev_printk(KERN_NOTICE, sdev, "%s: Detached\n", ALUA_DH_NAME);
806 }
807 
808 static int __init alua_init(void)
809 {
810 	int r;
811 
812 	r = scsi_register_device_handler(&alua_dh);
813 	if (r != 0)
814 		printk(KERN_ERR "%s: Failed to register scsi device handler",
815 			ALUA_DH_NAME);
816 	return r;
817 }
818 
819 static void __exit alua_exit(void)
820 {
821 	scsi_unregister_device_handler(&alua_dh);
822 }
823 
824 module_init(alua_init);
825 module_exit(alua_exit);
826 
827 MODULE_DESCRIPTION("DM Multipath ALUA support");
828 MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
829 MODULE_LICENSE("GPL");
830 MODULE_VERSION(ALUA_DH_VER);
831