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 <asm/unaligned.h>
26 #include <scsi/scsi.h>
27 #include <scsi/scsi_proto.h>
28 #include <scsi/scsi_dbg.h>
29 #include <scsi/scsi_eh.h>
30 #include <scsi/scsi_dh.h>
31 
32 #define ALUA_DH_NAME "alua"
33 #define ALUA_DH_VER "2.0"
34 
35 #define TPGS_SUPPORT_NONE		0x00
36 #define TPGS_SUPPORT_OPTIMIZED		0x01
37 #define TPGS_SUPPORT_NONOPTIMIZED	0x02
38 #define TPGS_SUPPORT_STANDBY		0x04
39 #define TPGS_SUPPORT_UNAVAILABLE	0x08
40 #define TPGS_SUPPORT_LBA_DEPENDENT	0x10
41 #define TPGS_SUPPORT_OFFLINE		0x40
42 #define TPGS_SUPPORT_TRANSITION		0x80
43 
44 #define RTPG_FMT_MASK			0x70
45 #define RTPG_FMT_EXT_HDR		0x10
46 
47 #define TPGS_MODE_UNINITIALIZED		 -1
48 #define TPGS_MODE_NONE			0x0
49 #define TPGS_MODE_IMPLICIT		0x1
50 #define TPGS_MODE_EXPLICIT		0x2
51 
52 #define ALUA_RTPG_SIZE			128
53 #define ALUA_FAILOVER_TIMEOUT		60
54 #define ALUA_FAILOVER_RETRIES		5
55 #define ALUA_RTPG_DELAY_MSECS		5
56 
57 /* device handler flags */
58 #define ALUA_OPTIMIZE_STPG		0x01
59 #define ALUA_RTPG_EXT_HDR_UNSUPP	0x02
60 #define ALUA_SYNC_STPG			0x04
61 /* State machine flags */
62 #define ALUA_PG_RUN_RTPG		0x10
63 #define ALUA_PG_RUN_STPG		0x20
64 #define ALUA_PG_RUNNING			0x40
65 
66 static uint optimize_stpg;
67 module_param(optimize_stpg, uint, S_IRUGO|S_IWUSR);
68 MODULE_PARM_DESC(optimize_stpg, "Allow use of a non-optimized path, rather than sending a STPG, when implicit TPGS is supported (0=No,1=Yes). Default is 0.");
69 
70 static LIST_HEAD(port_group_list);
71 static DEFINE_SPINLOCK(port_group_lock);
72 static struct workqueue_struct *kaluad_wq;
73 static struct workqueue_struct *kaluad_sync_wq;
74 
75 struct alua_port_group {
76 	struct kref		kref;
77 	struct rcu_head		rcu;
78 	struct list_head	node;
79 	struct list_head	dh_list;
80 	unsigned char		device_id_str[256];
81 	int			device_id_len;
82 	int			group_id;
83 	int			tpgs;
84 	int			state;
85 	int			pref;
86 	unsigned		flags; /* used for optimizing STPG */
87 	unsigned char		transition_tmo;
88 	unsigned long		expiry;
89 	unsigned long		interval;
90 	struct delayed_work	rtpg_work;
91 	spinlock_t		lock;
92 	struct list_head	rtpg_list;
93 	struct scsi_device	*rtpg_sdev;
94 };
95 
96 struct alua_dh_data {
97 	struct list_head	node;
98 	struct alua_port_group	*pg;
99 	int			group_id;
100 	spinlock_t		pg_lock;
101 	struct scsi_device	*sdev;
102 	int			init_error;
103 	struct mutex		init_mutex;
104 };
105 
106 struct alua_queue_data {
107 	struct list_head	entry;
108 	activate_complete	callback_fn;
109 	void			*callback_data;
110 };
111 
112 #define ALUA_POLICY_SWITCH_CURRENT	0
113 #define ALUA_POLICY_SWITCH_ALL		1
114 
115 static void alua_rtpg_work(struct work_struct *work);
116 static void alua_rtpg_queue(struct alua_port_group *pg,
117 			    struct scsi_device *sdev,
118 			    struct alua_queue_data *qdata, bool force);
119 static void alua_check(struct scsi_device *sdev, bool force);
120 
121 static void release_port_group(struct kref *kref)
122 {
123 	struct alua_port_group *pg;
124 
125 	pg = container_of(kref, struct alua_port_group, kref);
126 	if (pg->rtpg_sdev)
127 		flush_delayed_work(&pg->rtpg_work);
128 	spin_lock(&port_group_lock);
129 	list_del(&pg->node);
130 	spin_unlock(&port_group_lock);
131 	kfree_rcu(pg, rcu);
132 }
133 
134 /*
135  * submit_rtpg - Issue a REPORT TARGET GROUP STATES command
136  * @sdev: sdev the command should be sent to
137  */
138 static int submit_rtpg(struct scsi_device *sdev, unsigned char *buff,
139 		       int bufflen, struct scsi_sense_hdr *sshdr, int flags)
140 {
141 	u8 cdb[COMMAND_SIZE(MAINTENANCE_IN)];
142 	int req_flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
143 		REQ_FAILFAST_DRIVER;
144 
145 	/* Prepare the command. */
146 	memset(cdb, 0x0, COMMAND_SIZE(MAINTENANCE_IN));
147 	cdb[0] = MAINTENANCE_IN;
148 	if (!(flags & ALUA_RTPG_EXT_HDR_UNSUPP))
149 		cdb[1] = MI_REPORT_TARGET_PGS | MI_EXT_HDR_PARAM_FMT;
150 	else
151 		cdb[1] = MI_REPORT_TARGET_PGS;
152 	put_unaligned_be32(bufflen, &cdb[6]);
153 
154 	return scsi_execute_req_flags(sdev, cdb, DMA_FROM_DEVICE,
155 				      buff, bufflen, sshdr,
156 				      ALUA_FAILOVER_TIMEOUT * HZ,
157 				      ALUA_FAILOVER_RETRIES, NULL, req_flags);
158 }
159 
160 /*
161  * submit_stpg - Issue a SET TARGET PORT GROUP command
162  *
163  * Currently we're only setting the current target port group state
164  * to 'active/optimized' and let the array firmware figure out
165  * the states of the remaining groups.
166  */
167 static int submit_stpg(struct scsi_device *sdev, int group_id,
168 		       struct scsi_sense_hdr *sshdr)
169 {
170 	u8 cdb[COMMAND_SIZE(MAINTENANCE_OUT)];
171 	unsigned char stpg_data[8];
172 	int stpg_len = 8;
173 	int req_flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
174 		REQ_FAILFAST_DRIVER;
175 
176 	/* Prepare the data buffer */
177 	memset(stpg_data, 0, stpg_len);
178 	stpg_data[4] = SCSI_ACCESS_STATE_OPTIMAL;
179 	put_unaligned_be16(group_id, &stpg_data[6]);
180 
181 	/* Prepare the command. */
182 	memset(cdb, 0x0, COMMAND_SIZE(MAINTENANCE_OUT));
183 	cdb[0] = MAINTENANCE_OUT;
184 	cdb[1] = MO_SET_TARGET_PGS;
185 	put_unaligned_be32(stpg_len, &cdb[6]);
186 
187 	return scsi_execute_req_flags(sdev, cdb, DMA_TO_DEVICE,
188 				      stpg_data, stpg_len,
189 				      sshdr, ALUA_FAILOVER_TIMEOUT * HZ,
190 				      ALUA_FAILOVER_RETRIES, NULL, req_flags);
191 }
192 
193 static struct alua_port_group *alua_find_get_pg(char *id_str, size_t id_size,
194 						int group_id)
195 {
196 	struct alua_port_group *pg;
197 
198 	if (!id_str || !id_size || !strlen(id_str))
199 		return NULL;
200 
201 	list_for_each_entry(pg, &port_group_list, node) {
202 		if (pg->group_id != group_id)
203 			continue;
204 		if (!pg->device_id_len || pg->device_id_len != id_size)
205 			continue;
206 		if (strncmp(pg->device_id_str, id_str, id_size))
207 			continue;
208 		if (!kref_get_unless_zero(&pg->kref))
209 			continue;
210 		return pg;
211 	}
212 
213 	return NULL;
214 }
215 
216 /*
217  * alua_alloc_pg - Allocate a new port_group structure
218  * @sdev: scsi device
219  * @h: alua device_handler data
220  * @group_id: port group id
221  *
222  * Allocate a new port_group structure for a given
223  * device.
224  */
225 static struct alua_port_group *alua_alloc_pg(struct scsi_device *sdev,
226 					     int group_id, int tpgs)
227 {
228 	struct alua_port_group *pg, *tmp_pg;
229 
230 	pg = kzalloc(sizeof(struct alua_port_group), GFP_KERNEL);
231 	if (!pg)
232 		return ERR_PTR(-ENOMEM);
233 
234 	pg->device_id_len = scsi_vpd_lun_id(sdev, pg->device_id_str,
235 					    sizeof(pg->device_id_str));
236 	if (pg->device_id_len <= 0) {
237 		/*
238 		 * TPGS supported but no device identification found.
239 		 * Generate private device identification.
240 		 */
241 		sdev_printk(KERN_INFO, sdev,
242 			    "%s: No device descriptors found\n",
243 			    ALUA_DH_NAME);
244 		pg->device_id_str[0] = '\0';
245 		pg->device_id_len = 0;
246 	}
247 	pg->group_id = group_id;
248 	pg->tpgs = tpgs;
249 	pg->state = SCSI_ACCESS_STATE_OPTIMAL;
250 	if (optimize_stpg)
251 		pg->flags |= ALUA_OPTIMIZE_STPG;
252 	kref_init(&pg->kref);
253 	INIT_DELAYED_WORK(&pg->rtpg_work, alua_rtpg_work);
254 	INIT_LIST_HEAD(&pg->rtpg_list);
255 	INIT_LIST_HEAD(&pg->node);
256 	INIT_LIST_HEAD(&pg->dh_list);
257 	spin_lock_init(&pg->lock);
258 
259 	spin_lock(&port_group_lock);
260 	tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len,
261 				  group_id);
262 	if (tmp_pg) {
263 		spin_unlock(&port_group_lock);
264 		kfree(pg);
265 		return tmp_pg;
266 	}
267 
268 	list_add(&pg->node, &port_group_list);
269 	spin_unlock(&port_group_lock);
270 
271 	return pg;
272 }
273 
274 /*
275  * alua_check_tpgs - Evaluate TPGS setting
276  * @sdev: device to be checked
277  *
278  * Examine the TPGS setting of the sdev to find out if ALUA
279  * is supported.
280  */
281 static int alua_check_tpgs(struct scsi_device *sdev)
282 {
283 	int tpgs = TPGS_MODE_NONE;
284 
285 	/*
286 	 * ALUA support for non-disk devices is fraught with
287 	 * difficulties, so disable it for now.
288 	 */
289 	if (sdev->type != TYPE_DISK) {
290 		sdev_printk(KERN_INFO, sdev,
291 			    "%s: disable for non-disk devices\n",
292 			    ALUA_DH_NAME);
293 		return tpgs;
294 	}
295 
296 	tpgs = scsi_device_tpgs(sdev);
297 	switch (tpgs) {
298 	case TPGS_MODE_EXPLICIT|TPGS_MODE_IMPLICIT:
299 		sdev_printk(KERN_INFO, sdev,
300 			    "%s: supports implicit and explicit TPGS\n",
301 			    ALUA_DH_NAME);
302 		break;
303 	case TPGS_MODE_EXPLICIT:
304 		sdev_printk(KERN_INFO, sdev, "%s: supports explicit TPGS\n",
305 			    ALUA_DH_NAME);
306 		break;
307 	case TPGS_MODE_IMPLICIT:
308 		sdev_printk(KERN_INFO, sdev, "%s: supports implicit TPGS\n",
309 			    ALUA_DH_NAME);
310 		break;
311 	case TPGS_MODE_NONE:
312 		sdev_printk(KERN_INFO, sdev, "%s: not supported\n",
313 			    ALUA_DH_NAME);
314 		break;
315 	default:
316 		sdev_printk(KERN_INFO, sdev,
317 			    "%s: unsupported TPGS setting %d\n",
318 			    ALUA_DH_NAME, tpgs);
319 		tpgs = TPGS_MODE_NONE;
320 		break;
321 	}
322 
323 	return tpgs;
324 }
325 
326 /*
327  * alua_check_vpd - Evaluate INQUIRY vpd page 0x83
328  * @sdev: device to be checked
329  *
330  * Extract the relative target port and the target port group
331  * descriptor from the list of identificators.
332  */
333 static int alua_check_vpd(struct scsi_device *sdev, struct alua_dh_data *h,
334 			  int tpgs)
335 {
336 	int rel_port = -1, group_id;
337 	struct alua_port_group *pg, *old_pg = NULL;
338 	bool pg_updated = false;
339 	unsigned long flags;
340 
341 	group_id = scsi_vpd_tpg_id(sdev, &rel_port);
342 	if (group_id < 0) {
343 		/*
344 		 * Internal error; TPGS supported but required
345 		 * VPD identification descriptors not present.
346 		 * Disable ALUA support
347 		 */
348 		sdev_printk(KERN_INFO, sdev,
349 			    "%s: No target port descriptors found\n",
350 			    ALUA_DH_NAME);
351 		return SCSI_DH_DEV_UNSUPP;
352 	}
353 
354 	pg = alua_alloc_pg(sdev, group_id, tpgs);
355 	if (IS_ERR(pg)) {
356 		if (PTR_ERR(pg) == -ENOMEM)
357 			return SCSI_DH_NOMEM;
358 		return SCSI_DH_DEV_UNSUPP;
359 	}
360 	if (pg->device_id_len)
361 		sdev_printk(KERN_INFO, sdev,
362 			    "%s: device %s port group %x rel port %x\n",
363 			    ALUA_DH_NAME, pg->device_id_str,
364 			    group_id, rel_port);
365 	else
366 		sdev_printk(KERN_INFO, sdev,
367 			    "%s: port group %x rel port %x\n",
368 			    ALUA_DH_NAME, group_id, rel_port);
369 
370 	/* Check for existing port group references */
371 	spin_lock(&h->pg_lock);
372 	old_pg = h->pg;
373 	if (old_pg != pg) {
374 		/* port group has changed. Update to new port group */
375 		if (h->pg) {
376 			spin_lock_irqsave(&old_pg->lock, flags);
377 			list_del_rcu(&h->node);
378 			spin_unlock_irqrestore(&old_pg->lock, flags);
379 		}
380 		rcu_assign_pointer(h->pg, pg);
381 		pg_updated = true;
382 	}
383 
384 	spin_lock_irqsave(&pg->lock, flags);
385 	if (sdev->synchronous_alua)
386 		pg->flags |= ALUA_SYNC_STPG;
387 	if (pg_updated)
388 		list_add_rcu(&h->node, &pg->dh_list);
389 	spin_unlock_irqrestore(&pg->lock, flags);
390 
391 	alua_rtpg_queue(h->pg, sdev, NULL, true);
392 	spin_unlock(&h->pg_lock);
393 
394 	if (old_pg)
395 		kref_put(&old_pg->kref, release_port_group);
396 
397 	return SCSI_DH_OK;
398 }
399 
400 static char print_alua_state(unsigned char state)
401 {
402 	switch (state) {
403 	case SCSI_ACCESS_STATE_OPTIMAL:
404 		return 'A';
405 	case SCSI_ACCESS_STATE_ACTIVE:
406 		return 'N';
407 	case SCSI_ACCESS_STATE_STANDBY:
408 		return 'S';
409 	case SCSI_ACCESS_STATE_UNAVAILABLE:
410 		return 'U';
411 	case SCSI_ACCESS_STATE_LBA:
412 		return 'L';
413 	case SCSI_ACCESS_STATE_OFFLINE:
414 		return 'O';
415 	case SCSI_ACCESS_STATE_TRANSITIONING:
416 		return 'T';
417 	default:
418 		return 'X';
419 	}
420 }
421 
422 static int alua_check_sense(struct scsi_device *sdev,
423 			    struct scsi_sense_hdr *sense_hdr)
424 {
425 	switch (sense_hdr->sense_key) {
426 	case NOT_READY:
427 		if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a) {
428 			/*
429 			 * LUN Not Accessible - ALUA state transition
430 			 */
431 			alua_check(sdev, false);
432 			return NEEDS_RETRY;
433 		}
434 		break;
435 	case UNIT_ATTENTION:
436 		if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x00) {
437 			/*
438 			 * Power On, Reset, or Bus Device Reset.
439 			 * Might have obscured a state transition,
440 			 * so schedule a recheck.
441 			 */
442 			alua_check(sdev, true);
443 			return ADD_TO_MLQUEUE;
444 		}
445 		if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x04)
446 			/*
447 			 * Device internal reset
448 			 */
449 			return ADD_TO_MLQUEUE;
450 		if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x01)
451 			/*
452 			 * Mode Parameters Changed
453 			 */
454 			return ADD_TO_MLQUEUE;
455 		if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x06) {
456 			/*
457 			 * ALUA state changed
458 			 */
459 			alua_check(sdev, true);
460 			return ADD_TO_MLQUEUE;
461 		}
462 		if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x07) {
463 			/*
464 			 * Implicit ALUA state transition failed
465 			 */
466 			alua_check(sdev, true);
467 			return ADD_TO_MLQUEUE;
468 		}
469 		if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x03)
470 			/*
471 			 * Inquiry data has changed
472 			 */
473 			return ADD_TO_MLQUEUE;
474 		if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x0e)
475 			/*
476 			 * REPORTED_LUNS_DATA_HAS_CHANGED is reported
477 			 * when switching controllers on targets like
478 			 * Intel Multi-Flex. We can just retry.
479 			 */
480 			return ADD_TO_MLQUEUE;
481 		break;
482 	}
483 
484 	return SCSI_RETURN_NOT_HANDLED;
485 }
486 
487 /*
488  * alua_tur - Send a TEST UNIT READY
489  * @sdev: device to which the TEST UNIT READY command should be send
490  *
491  * Send a TEST UNIT READY to @sdev to figure out the device state
492  * Returns SCSI_DH_RETRY if the sense code is NOT READY/ALUA TRANSITIONING,
493  * SCSI_DH_OK if no error occurred, and SCSI_DH_IO otherwise.
494  */
495 static int alua_tur(struct scsi_device *sdev)
496 {
497 	struct scsi_sense_hdr sense_hdr;
498 	int retval;
499 
500 	retval = scsi_test_unit_ready(sdev, ALUA_FAILOVER_TIMEOUT * HZ,
501 				      ALUA_FAILOVER_RETRIES, &sense_hdr);
502 	if (sense_hdr.sense_key == NOT_READY &&
503 	    sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a)
504 		return SCSI_DH_RETRY;
505 	else if (retval)
506 		return SCSI_DH_IO;
507 	else
508 		return SCSI_DH_OK;
509 }
510 
511 /*
512  * alua_rtpg - Evaluate REPORT TARGET GROUP STATES
513  * @sdev: the device to be evaluated.
514  *
515  * Evaluate the Target Port Group State.
516  * Returns SCSI_DH_DEV_OFFLINED if the path is
517  * found to be unusable.
518  */
519 static int alua_rtpg(struct scsi_device *sdev, struct alua_port_group *pg)
520 {
521 	struct scsi_sense_hdr sense_hdr;
522 	struct alua_port_group *tmp_pg;
523 	int len, k, off, valid_states = 0, bufflen = ALUA_RTPG_SIZE;
524 	unsigned char *desc, *buff;
525 	unsigned err, retval;
526 	unsigned int tpg_desc_tbl_off;
527 	unsigned char orig_transition_tmo;
528 	unsigned long flags;
529 
530 	if (!pg->expiry) {
531 		unsigned long transition_tmo = ALUA_FAILOVER_TIMEOUT * HZ;
532 
533 		if (pg->transition_tmo)
534 			transition_tmo = pg->transition_tmo * HZ;
535 
536 		pg->expiry = round_jiffies_up(jiffies + transition_tmo);
537 	}
538 
539 	buff = kzalloc(bufflen, GFP_KERNEL);
540 	if (!buff)
541 		return SCSI_DH_DEV_TEMP_BUSY;
542 
543  retry:
544 	err = 0;
545 	retval = submit_rtpg(sdev, buff, bufflen, &sense_hdr, pg->flags);
546 
547 	if (retval) {
548 		if (!scsi_sense_valid(&sense_hdr)) {
549 			sdev_printk(KERN_INFO, sdev,
550 				    "%s: rtpg failed, result %d\n",
551 				    ALUA_DH_NAME, retval);
552 			kfree(buff);
553 			if (driver_byte(retval) == DRIVER_ERROR)
554 				return SCSI_DH_DEV_TEMP_BUSY;
555 			return SCSI_DH_IO;
556 		}
557 
558 		/*
559 		 * submit_rtpg() has failed on existing arrays
560 		 * when requesting extended header info, and
561 		 * the array doesn't support extended headers,
562 		 * even though it shouldn't according to T10.
563 		 * The retry without rtpg_ext_hdr_req set
564 		 * handles this.
565 		 */
566 		if (!(pg->flags & ALUA_RTPG_EXT_HDR_UNSUPP) &&
567 		    sense_hdr.sense_key == ILLEGAL_REQUEST &&
568 		    sense_hdr.asc == 0x24 && sense_hdr.ascq == 0) {
569 			pg->flags |= ALUA_RTPG_EXT_HDR_UNSUPP;
570 			goto retry;
571 		}
572 		/*
573 		 * Retry on ALUA state transition or if any
574 		 * UNIT ATTENTION occurred.
575 		 */
576 		if (sense_hdr.sense_key == NOT_READY &&
577 		    sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a)
578 			err = SCSI_DH_RETRY;
579 		else if (sense_hdr.sense_key == UNIT_ATTENTION)
580 			err = SCSI_DH_RETRY;
581 		if (err == SCSI_DH_RETRY &&
582 		    pg->expiry != 0 && time_before(jiffies, pg->expiry)) {
583 			sdev_printk(KERN_ERR, sdev, "%s: rtpg retry\n",
584 				    ALUA_DH_NAME);
585 			scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
586 			return err;
587 		}
588 		sdev_printk(KERN_ERR, sdev, "%s: rtpg failed\n",
589 			    ALUA_DH_NAME);
590 		scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
591 		kfree(buff);
592 		pg->expiry = 0;
593 		return SCSI_DH_IO;
594 	}
595 
596 	len = get_unaligned_be32(&buff[0]) + 4;
597 
598 	if (len > bufflen) {
599 		/* Resubmit with the correct length */
600 		kfree(buff);
601 		bufflen = len;
602 		buff = kmalloc(bufflen, GFP_KERNEL);
603 		if (!buff) {
604 			sdev_printk(KERN_WARNING, sdev,
605 				    "%s: kmalloc buffer failed\n",__func__);
606 			/* Temporary failure, bypass */
607 			pg->expiry = 0;
608 			return SCSI_DH_DEV_TEMP_BUSY;
609 		}
610 		goto retry;
611 	}
612 
613 	orig_transition_tmo = pg->transition_tmo;
614 	if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR && buff[5] != 0)
615 		pg->transition_tmo = buff[5];
616 	else
617 		pg->transition_tmo = ALUA_FAILOVER_TIMEOUT;
618 
619 	if (orig_transition_tmo != pg->transition_tmo) {
620 		sdev_printk(KERN_INFO, sdev,
621 			    "%s: transition timeout set to %d seconds\n",
622 			    ALUA_DH_NAME, pg->transition_tmo);
623 		pg->expiry = jiffies + pg->transition_tmo * HZ;
624 	}
625 
626 	if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR)
627 		tpg_desc_tbl_off = 8;
628 	else
629 		tpg_desc_tbl_off = 4;
630 
631 	for (k = tpg_desc_tbl_off, desc = buff + tpg_desc_tbl_off;
632 	     k < len;
633 	     k += off, desc += off) {
634 		u16 group_id = get_unaligned_be16(&desc[2]);
635 
636 		spin_lock_irqsave(&port_group_lock, flags);
637 		tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len,
638 					  group_id);
639 		spin_unlock_irqrestore(&port_group_lock, flags);
640 		if (tmp_pg) {
641 			if (spin_trylock_irqsave(&tmp_pg->lock, flags)) {
642 				if ((tmp_pg == pg) ||
643 				    !(tmp_pg->flags & ALUA_PG_RUNNING)) {
644 					struct alua_dh_data *h;
645 
646 					tmp_pg->state = desc[0] & 0x0f;
647 					tmp_pg->pref = desc[0] >> 7;
648 					rcu_read_lock();
649 					list_for_each_entry_rcu(h,
650 						&tmp_pg->dh_list, node) {
651 						/* h->sdev should always be valid */
652 						BUG_ON(!h->sdev);
653 						h->sdev->access_state = desc[0];
654 					}
655 					rcu_read_unlock();
656 				}
657 				if (tmp_pg == pg)
658 					valid_states = desc[1];
659 				spin_unlock_irqrestore(&tmp_pg->lock, flags);
660 			}
661 			kref_put(&tmp_pg->kref, release_port_group);
662 		}
663 		off = 8 + (desc[7] * 4);
664 	}
665 
666 	spin_lock_irqsave(&pg->lock, flags);
667 	sdev_printk(KERN_INFO, sdev,
668 		    "%s: port group %02x state %c %s supports %c%c%c%c%c%c%c\n",
669 		    ALUA_DH_NAME, pg->group_id, print_alua_state(pg->state),
670 		    pg->pref ? "preferred" : "non-preferred",
671 		    valid_states&TPGS_SUPPORT_TRANSITION?'T':'t',
672 		    valid_states&TPGS_SUPPORT_OFFLINE?'O':'o',
673 		    valid_states&TPGS_SUPPORT_LBA_DEPENDENT?'L':'l',
674 		    valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u',
675 		    valid_states&TPGS_SUPPORT_STANDBY?'S':'s',
676 		    valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n',
677 		    valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a');
678 
679 	switch (pg->state) {
680 	case SCSI_ACCESS_STATE_TRANSITIONING:
681 		if (time_before(jiffies, pg->expiry)) {
682 			/* State transition, retry */
683 			pg->interval = 2;
684 			err = SCSI_DH_RETRY;
685 		} else {
686 			struct alua_dh_data *h;
687 
688 			/* Transitioning time exceeded, set port to standby */
689 			err = SCSI_DH_IO;
690 			pg->state = SCSI_ACCESS_STATE_STANDBY;
691 			pg->expiry = 0;
692 			rcu_read_lock();
693 			list_for_each_entry_rcu(h, &pg->dh_list, node) {
694 				BUG_ON(!h->sdev);
695 				h->sdev->access_state =
696 					(pg->state & SCSI_ACCESS_STATE_MASK);
697 				if (pg->pref)
698 					h->sdev->access_state |=
699 						SCSI_ACCESS_STATE_PREFERRED;
700 			}
701 			rcu_read_unlock();
702 		}
703 		break;
704 	case SCSI_ACCESS_STATE_OFFLINE:
705 		/* Path unusable */
706 		err = SCSI_DH_DEV_OFFLINED;
707 		pg->expiry = 0;
708 		break;
709 	default:
710 		/* Useable path if active */
711 		err = SCSI_DH_OK;
712 		pg->expiry = 0;
713 		break;
714 	}
715 	spin_unlock_irqrestore(&pg->lock, flags);
716 	kfree(buff);
717 	return err;
718 }
719 
720 /*
721  * alua_stpg - Issue a SET TARGET PORT GROUP command
722  *
723  * Issue a SET TARGET PORT GROUP command and evaluate the
724  * response. Returns SCSI_DH_RETRY per default to trigger
725  * a re-evaluation of the target group state or SCSI_DH_OK
726  * if no further action needs to be taken.
727  */
728 static unsigned alua_stpg(struct scsi_device *sdev, struct alua_port_group *pg)
729 {
730 	int retval;
731 	struct scsi_sense_hdr sense_hdr;
732 
733 	if (!(pg->tpgs & TPGS_MODE_EXPLICIT)) {
734 		/* Only implicit ALUA supported, retry */
735 		return SCSI_DH_RETRY;
736 	}
737 	switch (pg->state) {
738 	case SCSI_ACCESS_STATE_OPTIMAL:
739 		return SCSI_DH_OK;
740 	case SCSI_ACCESS_STATE_ACTIVE:
741 		if ((pg->flags & ALUA_OPTIMIZE_STPG) &&
742 		    !pg->pref &&
743 		    (pg->tpgs & TPGS_MODE_IMPLICIT))
744 			return SCSI_DH_OK;
745 		break;
746 	case SCSI_ACCESS_STATE_STANDBY:
747 	case SCSI_ACCESS_STATE_UNAVAILABLE:
748 		break;
749 	case SCSI_ACCESS_STATE_OFFLINE:
750 		return SCSI_DH_IO;
751 	case SCSI_ACCESS_STATE_TRANSITIONING:
752 		break;
753 	default:
754 		sdev_printk(KERN_INFO, sdev,
755 			    "%s: stpg failed, unhandled TPGS state %d",
756 			    ALUA_DH_NAME, pg->state);
757 		return SCSI_DH_NOSYS;
758 	}
759 	retval = submit_stpg(sdev, pg->group_id, &sense_hdr);
760 
761 	if (retval) {
762 		if (!scsi_sense_valid(&sense_hdr)) {
763 			sdev_printk(KERN_INFO, sdev,
764 				    "%s: stpg failed, result %d",
765 				    ALUA_DH_NAME, retval);
766 			if (driver_byte(retval) == DRIVER_ERROR)
767 				return SCSI_DH_DEV_TEMP_BUSY;
768 		} else {
769 			sdev_printk(KERN_INFO, sdev, "%s: stpg failed\n",
770 				    ALUA_DH_NAME);
771 			scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
772 		}
773 	}
774 	/* Retry RTPG */
775 	return SCSI_DH_RETRY;
776 }
777 
778 static void alua_rtpg_work(struct work_struct *work)
779 {
780 	struct alua_port_group *pg =
781 		container_of(work, struct alua_port_group, rtpg_work.work);
782 	struct scsi_device *sdev;
783 	LIST_HEAD(qdata_list);
784 	int err = SCSI_DH_OK;
785 	struct alua_queue_data *qdata, *tmp;
786 	unsigned long flags;
787 	struct workqueue_struct *alua_wq = kaluad_wq;
788 
789 	spin_lock_irqsave(&pg->lock, flags);
790 	sdev = pg->rtpg_sdev;
791 	if (!sdev) {
792 		WARN_ON(pg->flags & ALUA_PG_RUN_RTPG);
793 		WARN_ON(pg->flags & ALUA_PG_RUN_STPG);
794 		spin_unlock_irqrestore(&pg->lock, flags);
795 		return;
796 	}
797 	if (pg->flags & ALUA_SYNC_STPG)
798 		alua_wq = kaluad_sync_wq;
799 	pg->flags |= ALUA_PG_RUNNING;
800 	if (pg->flags & ALUA_PG_RUN_RTPG) {
801 		int state = pg->state;
802 
803 		pg->flags &= ~ALUA_PG_RUN_RTPG;
804 		spin_unlock_irqrestore(&pg->lock, flags);
805 		if (state == SCSI_ACCESS_STATE_TRANSITIONING) {
806 			if (alua_tur(sdev) == SCSI_DH_RETRY) {
807 				spin_lock_irqsave(&pg->lock, flags);
808 				pg->flags &= ~ALUA_PG_RUNNING;
809 				pg->flags |= ALUA_PG_RUN_RTPG;
810 				spin_unlock_irqrestore(&pg->lock, flags);
811 				queue_delayed_work(alua_wq, &pg->rtpg_work,
812 						   pg->interval * HZ);
813 				return;
814 			}
815 			/* Send RTPG on failure or if TUR indicates SUCCESS */
816 		}
817 		err = alua_rtpg(sdev, pg);
818 		spin_lock_irqsave(&pg->lock, flags);
819 		if (err == SCSI_DH_RETRY || pg->flags & ALUA_PG_RUN_RTPG) {
820 			pg->flags &= ~ALUA_PG_RUNNING;
821 			pg->flags |= ALUA_PG_RUN_RTPG;
822 			spin_unlock_irqrestore(&pg->lock, flags);
823 			queue_delayed_work(alua_wq, &pg->rtpg_work,
824 					   pg->interval * HZ);
825 			return;
826 		}
827 		if (err != SCSI_DH_OK)
828 			pg->flags &= ~ALUA_PG_RUN_STPG;
829 	}
830 	if (pg->flags & ALUA_PG_RUN_STPG) {
831 		pg->flags &= ~ALUA_PG_RUN_STPG;
832 		spin_unlock_irqrestore(&pg->lock, flags);
833 		err = alua_stpg(sdev, pg);
834 		spin_lock_irqsave(&pg->lock, flags);
835 		if (err == SCSI_DH_RETRY || pg->flags & ALUA_PG_RUN_RTPG) {
836 			pg->flags |= ALUA_PG_RUN_RTPG;
837 			pg->interval = 0;
838 			pg->flags &= ~ALUA_PG_RUNNING;
839 			spin_unlock_irqrestore(&pg->lock, flags);
840 			queue_delayed_work(alua_wq, &pg->rtpg_work,
841 					   pg->interval * HZ);
842 			return;
843 		}
844 	}
845 
846 	list_splice_init(&pg->rtpg_list, &qdata_list);
847 	pg->rtpg_sdev = NULL;
848 	spin_unlock_irqrestore(&pg->lock, flags);
849 
850 	list_for_each_entry_safe(qdata, tmp, &qdata_list, entry) {
851 		list_del(&qdata->entry);
852 		if (qdata->callback_fn)
853 			qdata->callback_fn(qdata->callback_data, err);
854 		kfree(qdata);
855 	}
856 	spin_lock_irqsave(&pg->lock, flags);
857 	pg->flags &= ~ALUA_PG_RUNNING;
858 	spin_unlock_irqrestore(&pg->lock, flags);
859 	scsi_device_put(sdev);
860 	kref_put(&pg->kref, release_port_group);
861 }
862 
863 static void alua_rtpg_queue(struct alua_port_group *pg,
864 			    struct scsi_device *sdev,
865 			    struct alua_queue_data *qdata, bool force)
866 {
867 	int start_queue = 0;
868 	unsigned long flags;
869 	struct workqueue_struct *alua_wq = kaluad_wq;
870 
871 	if (!pg)
872 		return;
873 
874 	spin_lock_irqsave(&pg->lock, flags);
875 	if (qdata) {
876 		list_add_tail(&qdata->entry, &pg->rtpg_list);
877 		pg->flags |= ALUA_PG_RUN_STPG;
878 		force = true;
879 	}
880 	if (pg->rtpg_sdev == NULL) {
881 		pg->interval = 0;
882 		pg->flags |= ALUA_PG_RUN_RTPG;
883 		kref_get(&pg->kref);
884 		pg->rtpg_sdev = sdev;
885 		scsi_device_get(sdev);
886 		start_queue = 1;
887 	} else if (!(pg->flags & ALUA_PG_RUN_RTPG) && force) {
888 		pg->flags |= ALUA_PG_RUN_RTPG;
889 		/* Do not queue if the worker is already running */
890 		if (!(pg->flags & ALUA_PG_RUNNING)) {
891 			kref_get(&pg->kref);
892 			start_queue = 1;
893 		}
894 	}
895 
896 	if (pg->flags & ALUA_SYNC_STPG)
897 		alua_wq = kaluad_sync_wq;
898 	spin_unlock_irqrestore(&pg->lock, flags);
899 
900 	if (start_queue &&
901 	    !queue_delayed_work(alua_wq, &pg->rtpg_work,
902 				msecs_to_jiffies(ALUA_RTPG_DELAY_MSECS))) {
903 		scsi_device_put(sdev);
904 		kref_put(&pg->kref, release_port_group);
905 	}
906 }
907 
908 /*
909  * alua_initialize - Initialize ALUA state
910  * @sdev: the device to be initialized
911  *
912  * For the prep_fn to work correctly we have
913  * to initialize the ALUA state for the device.
914  */
915 static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h)
916 {
917 	int err = SCSI_DH_DEV_UNSUPP, tpgs;
918 
919 	mutex_lock(&h->init_mutex);
920 	tpgs = alua_check_tpgs(sdev);
921 	if (tpgs != TPGS_MODE_NONE)
922 		err = alua_check_vpd(sdev, h, tpgs);
923 	h->init_error = err;
924 	mutex_unlock(&h->init_mutex);
925 	return err;
926 }
927 /*
928  * alua_set_params - set/unset the optimize flag
929  * @sdev: device on the path to be activated
930  * params - parameters in the following format
931  *      "no_of_params\0param1\0param2\0param3\0...\0"
932  * For example, to set the flag pass the following parameters
933  * from multipath.conf
934  *     hardware_handler        "2 alua 1"
935  */
936 static int alua_set_params(struct scsi_device *sdev, const char *params)
937 {
938 	struct alua_dh_data *h = sdev->handler_data;
939 	struct alua_port_group __rcu *pg = NULL;
940 	unsigned int optimize = 0, argc;
941 	const char *p = params;
942 	int result = SCSI_DH_OK;
943 	unsigned long flags;
944 
945 	if ((sscanf(params, "%u", &argc) != 1) || (argc != 1))
946 		return -EINVAL;
947 
948 	while (*p++)
949 		;
950 	if ((sscanf(p, "%u", &optimize) != 1) || (optimize > 1))
951 		return -EINVAL;
952 
953 	rcu_read_lock();
954 	pg = rcu_dereference(h->pg);
955 	if (!pg) {
956 		rcu_read_unlock();
957 		return -ENXIO;
958 	}
959 	spin_lock_irqsave(&pg->lock, flags);
960 	if (optimize)
961 		pg->flags |= ALUA_OPTIMIZE_STPG;
962 	else
963 		pg->flags &= ~ALUA_OPTIMIZE_STPG;
964 	spin_unlock_irqrestore(&pg->lock, flags);
965 	rcu_read_unlock();
966 
967 	return result;
968 }
969 
970 /*
971  * alua_activate - activate a path
972  * @sdev: device on the path to be activated
973  *
974  * We're currently switching the port group to be activated only and
975  * let the array figure out the rest.
976  * There may be other arrays which require us to switch all port groups
977  * based on a certain policy. But until we actually encounter them it
978  * should be okay.
979  */
980 static int alua_activate(struct scsi_device *sdev,
981 			activate_complete fn, void *data)
982 {
983 	struct alua_dh_data *h = sdev->handler_data;
984 	int err = SCSI_DH_OK;
985 	struct alua_queue_data *qdata;
986 	struct alua_port_group __rcu *pg;
987 
988 	qdata = kzalloc(sizeof(*qdata), GFP_KERNEL);
989 	if (!qdata) {
990 		err = SCSI_DH_RES_TEMP_UNAVAIL;
991 		goto out;
992 	}
993 	qdata->callback_fn = fn;
994 	qdata->callback_data = data;
995 
996 	mutex_lock(&h->init_mutex);
997 	rcu_read_lock();
998 	pg = rcu_dereference(h->pg);
999 	if (!pg || !kref_get_unless_zero(&pg->kref)) {
1000 		rcu_read_unlock();
1001 		kfree(qdata);
1002 		err = h->init_error;
1003 		mutex_unlock(&h->init_mutex);
1004 		goto out;
1005 	}
1006 	fn = NULL;
1007 	rcu_read_unlock();
1008 	mutex_unlock(&h->init_mutex);
1009 
1010 	alua_rtpg_queue(pg, sdev, qdata, true);
1011 	kref_put(&pg->kref, release_port_group);
1012 out:
1013 	if (fn)
1014 		fn(data, err);
1015 	return 0;
1016 }
1017 
1018 /*
1019  * alua_check - check path status
1020  * @sdev: device on the path to be checked
1021  *
1022  * Check the device status
1023  */
1024 static void alua_check(struct scsi_device *sdev, bool force)
1025 {
1026 	struct alua_dh_data *h = sdev->handler_data;
1027 	struct alua_port_group *pg;
1028 
1029 	rcu_read_lock();
1030 	pg = rcu_dereference(h->pg);
1031 	if (!pg || !kref_get_unless_zero(&pg->kref)) {
1032 		rcu_read_unlock();
1033 		return;
1034 	}
1035 	rcu_read_unlock();
1036 
1037 	alua_rtpg_queue(pg, sdev, NULL, force);
1038 	kref_put(&pg->kref, release_port_group);
1039 }
1040 
1041 /*
1042  * alua_prep_fn - request callback
1043  *
1044  * Fail I/O to all paths not in state
1045  * active/optimized or active/non-optimized.
1046  */
1047 static int alua_prep_fn(struct scsi_device *sdev, struct request *req)
1048 {
1049 	struct alua_dh_data *h = sdev->handler_data;
1050 	struct alua_port_group __rcu *pg;
1051 	unsigned char state = SCSI_ACCESS_STATE_OPTIMAL;
1052 	int ret = BLKPREP_OK;
1053 
1054 	rcu_read_lock();
1055 	pg = rcu_dereference(h->pg);
1056 	if (pg)
1057 		state = pg->state;
1058 	rcu_read_unlock();
1059 	if (state == SCSI_ACCESS_STATE_TRANSITIONING)
1060 		ret = BLKPREP_DEFER;
1061 	else if (state != SCSI_ACCESS_STATE_OPTIMAL &&
1062 		 state != SCSI_ACCESS_STATE_ACTIVE &&
1063 		 state != SCSI_ACCESS_STATE_LBA) {
1064 		ret = BLKPREP_KILL;
1065 		req->cmd_flags |= REQ_QUIET;
1066 	}
1067 	return ret;
1068 
1069 }
1070 
1071 static void alua_rescan(struct scsi_device *sdev)
1072 {
1073 	struct alua_dh_data *h = sdev->handler_data;
1074 
1075 	alua_initialize(sdev, h);
1076 }
1077 
1078 /*
1079  * alua_bus_attach - Attach device handler
1080  * @sdev: device to be attached to
1081  */
1082 static int alua_bus_attach(struct scsi_device *sdev)
1083 {
1084 	struct alua_dh_data *h;
1085 	int err, ret = -EINVAL;
1086 
1087 	h = kzalloc(sizeof(*h) , GFP_KERNEL);
1088 	if (!h)
1089 		return -ENOMEM;
1090 	spin_lock_init(&h->pg_lock);
1091 	rcu_assign_pointer(h->pg, NULL);
1092 	h->init_error = SCSI_DH_OK;
1093 	h->sdev = sdev;
1094 	INIT_LIST_HEAD(&h->node);
1095 
1096 	mutex_init(&h->init_mutex);
1097 	err = alua_initialize(sdev, h);
1098 	if (err == SCSI_DH_NOMEM)
1099 		ret = -ENOMEM;
1100 	if (err != SCSI_DH_OK && err != SCSI_DH_DEV_OFFLINED)
1101 		goto failed;
1102 
1103 	sdev->handler_data = h;
1104 	return 0;
1105 failed:
1106 	kfree(h);
1107 	return ret;
1108 }
1109 
1110 /*
1111  * alua_bus_detach - Detach device handler
1112  * @sdev: device to be detached from
1113  */
1114 static void alua_bus_detach(struct scsi_device *sdev)
1115 {
1116 	struct alua_dh_data *h = sdev->handler_data;
1117 	struct alua_port_group *pg;
1118 
1119 	spin_lock(&h->pg_lock);
1120 	pg = h->pg;
1121 	rcu_assign_pointer(h->pg, NULL);
1122 	h->sdev = NULL;
1123 	spin_unlock(&h->pg_lock);
1124 	if (pg) {
1125 		spin_lock_irq(&pg->lock);
1126 		list_del_rcu(&h->node);
1127 		spin_unlock_irq(&pg->lock);
1128 		kref_put(&pg->kref, release_port_group);
1129 	}
1130 	sdev->handler_data = NULL;
1131 	kfree(h);
1132 }
1133 
1134 static struct scsi_device_handler alua_dh = {
1135 	.name = ALUA_DH_NAME,
1136 	.module = THIS_MODULE,
1137 	.attach = alua_bus_attach,
1138 	.detach = alua_bus_detach,
1139 	.prep_fn = alua_prep_fn,
1140 	.check_sense = alua_check_sense,
1141 	.activate = alua_activate,
1142 	.rescan = alua_rescan,
1143 	.set_params = alua_set_params,
1144 };
1145 
1146 static int __init alua_init(void)
1147 {
1148 	int r;
1149 
1150 	kaluad_wq = alloc_workqueue("kaluad", WQ_MEM_RECLAIM, 0);
1151 	if (!kaluad_wq) {
1152 		/* Temporary failure, bypass */
1153 		return SCSI_DH_DEV_TEMP_BUSY;
1154 	}
1155 	kaluad_sync_wq = create_workqueue("kaluad_sync");
1156 	if (!kaluad_sync_wq) {
1157 		destroy_workqueue(kaluad_wq);
1158 		return SCSI_DH_DEV_TEMP_BUSY;
1159 	}
1160 	r = scsi_register_device_handler(&alua_dh);
1161 	if (r != 0) {
1162 		printk(KERN_ERR "%s: Failed to register scsi device handler",
1163 			ALUA_DH_NAME);
1164 		destroy_workqueue(kaluad_sync_wq);
1165 		destroy_workqueue(kaluad_wq);
1166 	}
1167 	return r;
1168 }
1169 
1170 static void __exit alua_exit(void)
1171 {
1172 	scsi_unregister_device_handler(&alua_dh);
1173 	destroy_workqueue(kaluad_sync_wq);
1174 	destroy_workqueue(kaluad_wq);
1175 }
1176 
1177 module_init(alua_init);
1178 module_exit(alua_exit);
1179 
1180 MODULE_DESCRIPTION("DM Multipath ALUA support");
1181 MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
1182 MODULE_LICENSE("GPL");
1183 MODULE_VERSION(ALUA_DH_VER);
1184