1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Generic SCSI-3 ALUA SCSI Device Handler
4  *
5  * Copyright (C) 2007-2010 Hannes Reinecke, SUSE Linux Products GmbH.
6  * All rights reserved.
7  */
8 #include <linux/slab.h>
9 #include <linux/delay.h>
10 #include <linux/module.h>
11 #include <asm/unaligned.h>
12 #include <scsi/scsi.h>
13 #include <scsi/scsi_proto.h>
14 #include <scsi/scsi_dbg.h>
15 #include <scsi/scsi_eh.h>
16 #include <scsi/scsi_dh.h>
17 
18 #define ALUA_DH_NAME "alua"
19 #define ALUA_DH_VER "2.0"
20 
21 #define TPGS_SUPPORT_NONE		0x00
22 #define TPGS_SUPPORT_OPTIMIZED		0x01
23 #define TPGS_SUPPORT_NONOPTIMIZED	0x02
24 #define TPGS_SUPPORT_STANDBY		0x04
25 #define TPGS_SUPPORT_UNAVAILABLE	0x08
26 #define TPGS_SUPPORT_LBA_DEPENDENT	0x10
27 #define TPGS_SUPPORT_OFFLINE		0x40
28 #define TPGS_SUPPORT_TRANSITION		0x80
29 #define TPGS_SUPPORT_ALL		0xdf
30 
31 #define RTPG_FMT_MASK			0x70
32 #define RTPG_FMT_EXT_HDR		0x10
33 
34 #define TPGS_MODE_UNINITIALIZED		 -1
35 #define TPGS_MODE_NONE			0x0
36 #define TPGS_MODE_IMPLICIT		0x1
37 #define TPGS_MODE_EXPLICIT		0x2
38 
39 #define ALUA_RTPG_SIZE			128
40 #define ALUA_FAILOVER_TIMEOUT		60
41 #define ALUA_FAILOVER_RETRIES		5
42 #define ALUA_RTPG_DELAY_MSECS		5
43 #define ALUA_RTPG_RETRY_DELAY		2
44 
45 /* device handler flags */
46 #define ALUA_OPTIMIZE_STPG		0x01
47 #define ALUA_RTPG_EXT_HDR_UNSUPP	0x02
48 /* State machine flags */
49 #define ALUA_PG_RUN_RTPG		0x10
50 #define ALUA_PG_RUN_STPG		0x20
51 #define ALUA_PG_RUNNING			0x40
52 
53 static uint optimize_stpg;
54 module_param(optimize_stpg, uint, S_IRUGO|S_IWUSR);
55 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.");
56 
57 static LIST_HEAD(port_group_list);
58 static DEFINE_SPINLOCK(port_group_lock);
59 static struct workqueue_struct *kaluad_wq;
60 
61 struct alua_port_group {
62 	struct kref		kref;
63 	struct rcu_head		rcu;
64 	struct list_head	node;
65 	struct list_head	dh_list;
66 	unsigned char		device_id_str[256];
67 	int			device_id_len;
68 	int			group_id;
69 	int			tpgs;
70 	int			state;
71 	int			pref;
72 	int			valid_states;
73 	unsigned		flags; /* used for optimizing STPG */
74 	unsigned char		transition_tmo;
75 	unsigned long		expiry;
76 	unsigned long		interval;
77 	struct delayed_work	rtpg_work;
78 	spinlock_t		lock;
79 	struct list_head	rtpg_list;
80 	struct scsi_device	*rtpg_sdev;
81 };
82 
83 struct alua_dh_data {
84 	struct list_head	node;
85 	struct alua_port_group __rcu *pg;
86 	int			group_id;
87 	spinlock_t		pg_lock;
88 	struct scsi_device	*sdev;
89 	int			init_error;
90 	struct mutex		init_mutex;
91 };
92 
93 struct alua_queue_data {
94 	struct list_head	entry;
95 	activate_complete	callback_fn;
96 	void			*callback_data;
97 };
98 
99 #define ALUA_POLICY_SWITCH_CURRENT	0
100 #define ALUA_POLICY_SWITCH_ALL		1
101 
102 static void alua_rtpg_work(struct work_struct *work);
103 static bool alua_rtpg_queue(struct alua_port_group *pg,
104 			    struct scsi_device *sdev,
105 			    struct alua_queue_data *qdata, bool force);
106 static void alua_check(struct scsi_device *sdev, bool force);
107 
108 static void release_port_group(struct kref *kref)
109 {
110 	struct alua_port_group *pg;
111 
112 	pg = container_of(kref, struct alua_port_group, kref);
113 	if (pg->rtpg_sdev)
114 		flush_delayed_work(&pg->rtpg_work);
115 	spin_lock(&port_group_lock);
116 	list_del(&pg->node);
117 	spin_unlock(&port_group_lock);
118 	kfree_rcu(pg, rcu);
119 }
120 
121 /*
122  * submit_rtpg - Issue a REPORT TARGET GROUP STATES command
123  * @sdev: sdev the command should be sent to
124  */
125 static int submit_rtpg(struct scsi_device *sdev, unsigned char *buff,
126 		       int bufflen, struct scsi_sense_hdr *sshdr, int flags)
127 {
128 	u8 cdb[MAX_COMMAND_SIZE];
129 	int req_flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
130 		REQ_FAILFAST_DRIVER;
131 
132 	/* Prepare the command. */
133 	memset(cdb, 0x0, MAX_COMMAND_SIZE);
134 	cdb[0] = MAINTENANCE_IN;
135 	if (!(flags & ALUA_RTPG_EXT_HDR_UNSUPP))
136 		cdb[1] = MI_REPORT_TARGET_PGS | MI_EXT_HDR_PARAM_FMT;
137 	else
138 		cdb[1] = MI_REPORT_TARGET_PGS;
139 	put_unaligned_be32(bufflen, &cdb[6]);
140 
141 	return scsi_execute(sdev, cdb, DMA_FROM_DEVICE, buff, bufflen, NULL,
142 			sshdr, ALUA_FAILOVER_TIMEOUT * HZ,
143 			ALUA_FAILOVER_RETRIES, req_flags, 0, NULL);
144 }
145 
146 /*
147  * submit_stpg - Issue a SET TARGET PORT GROUP command
148  *
149  * Currently we're only setting the current target port group state
150  * to 'active/optimized' and let the array firmware figure out
151  * the states of the remaining groups.
152  */
153 static int submit_stpg(struct scsi_device *sdev, int group_id,
154 		       struct scsi_sense_hdr *sshdr)
155 {
156 	u8 cdb[MAX_COMMAND_SIZE];
157 	unsigned char stpg_data[8];
158 	int stpg_len = 8;
159 	int req_flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
160 		REQ_FAILFAST_DRIVER;
161 
162 	/* Prepare the data buffer */
163 	memset(stpg_data, 0, stpg_len);
164 	stpg_data[4] = SCSI_ACCESS_STATE_OPTIMAL;
165 	put_unaligned_be16(group_id, &stpg_data[6]);
166 
167 	/* Prepare the command. */
168 	memset(cdb, 0x0, MAX_COMMAND_SIZE);
169 	cdb[0] = MAINTENANCE_OUT;
170 	cdb[1] = MO_SET_TARGET_PGS;
171 	put_unaligned_be32(stpg_len, &cdb[6]);
172 
173 	return scsi_execute(sdev, cdb, DMA_TO_DEVICE, stpg_data, stpg_len, NULL,
174 			sshdr, ALUA_FAILOVER_TIMEOUT * HZ,
175 			ALUA_FAILOVER_RETRIES, req_flags, 0, NULL);
176 }
177 
178 static struct alua_port_group *alua_find_get_pg(char *id_str, size_t id_size,
179 						int group_id)
180 {
181 	struct alua_port_group *pg;
182 
183 	if (!id_str || !id_size || !strlen(id_str))
184 		return NULL;
185 
186 	list_for_each_entry(pg, &port_group_list, node) {
187 		if (pg->group_id != group_id)
188 			continue;
189 		if (!pg->device_id_len || pg->device_id_len != id_size)
190 			continue;
191 		if (strncmp(pg->device_id_str, id_str, id_size))
192 			continue;
193 		if (!kref_get_unless_zero(&pg->kref))
194 			continue;
195 		return pg;
196 	}
197 
198 	return NULL;
199 }
200 
201 /*
202  * alua_alloc_pg - Allocate a new port_group structure
203  * @sdev: scsi device
204  * @group_id: port group id
205  * @tpgs: target port group settings
206  *
207  * Allocate a new port_group structure for a given
208  * device.
209  */
210 static struct alua_port_group *alua_alloc_pg(struct scsi_device *sdev,
211 					     int group_id, int tpgs)
212 {
213 	struct alua_port_group *pg, *tmp_pg;
214 
215 	pg = kzalloc(sizeof(struct alua_port_group), GFP_KERNEL);
216 	if (!pg)
217 		return ERR_PTR(-ENOMEM);
218 
219 	pg->device_id_len = scsi_vpd_lun_id(sdev, pg->device_id_str,
220 					    sizeof(pg->device_id_str));
221 	if (pg->device_id_len <= 0) {
222 		/*
223 		 * TPGS supported but no device identification found.
224 		 * Generate private device identification.
225 		 */
226 		sdev_printk(KERN_INFO, sdev,
227 			    "%s: No device descriptors found\n",
228 			    ALUA_DH_NAME);
229 		pg->device_id_str[0] = '\0';
230 		pg->device_id_len = 0;
231 	}
232 	pg->group_id = group_id;
233 	pg->tpgs = tpgs;
234 	pg->state = SCSI_ACCESS_STATE_OPTIMAL;
235 	pg->valid_states = TPGS_SUPPORT_ALL;
236 	if (optimize_stpg)
237 		pg->flags |= ALUA_OPTIMIZE_STPG;
238 	kref_init(&pg->kref);
239 	INIT_DELAYED_WORK(&pg->rtpg_work, alua_rtpg_work);
240 	INIT_LIST_HEAD(&pg->rtpg_list);
241 	INIT_LIST_HEAD(&pg->node);
242 	INIT_LIST_HEAD(&pg->dh_list);
243 	spin_lock_init(&pg->lock);
244 
245 	spin_lock(&port_group_lock);
246 	tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len,
247 				  group_id);
248 	if (tmp_pg) {
249 		spin_unlock(&port_group_lock);
250 		kfree(pg);
251 		return tmp_pg;
252 	}
253 
254 	list_add(&pg->node, &port_group_list);
255 	spin_unlock(&port_group_lock);
256 
257 	return pg;
258 }
259 
260 /*
261  * alua_check_tpgs - Evaluate TPGS setting
262  * @sdev: device to be checked
263  *
264  * Examine the TPGS setting of the sdev to find out if ALUA
265  * is supported.
266  */
267 static int alua_check_tpgs(struct scsi_device *sdev)
268 {
269 	int tpgs = TPGS_MODE_NONE;
270 
271 	/*
272 	 * ALUA support for non-disk devices is fraught with
273 	 * difficulties, so disable it for now.
274 	 */
275 	if (sdev->type != TYPE_DISK) {
276 		sdev_printk(KERN_INFO, sdev,
277 			    "%s: disable for non-disk devices\n",
278 			    ALUA_DH_NAME);
279 		return tpgs;
280 	}
281 
282 	tpgs = scsi_device_tpgs(sdev);
283 	switch (tpgs) {
284 	case TPGS_MODE_EXPLICIT|TPGS_MODE_IMPLICIT:
285 		sdev_printk(KERN_INFO, sdev,
286 			    "%s: supports implicit and explicit TPGS\n",
287 			    ALUA_DH_NAME);
288 		break;
289 	case TPGS_MODE_EXPLICIT:
290 		sdev_printk(KERN_INFO, sdev, "%s: supports explicit TPGS\n",
291 			    ALUA_DH_NAME);
292 		break;
293 	case TPGS_MODE_IMPLICIT:
294 		sdev_printk(KERN_INFO, sdev, "%s: supports implicit TPGS\n",
295 			    ALUA_DH_NAME);
296 		break;
297 	case TPGS_MODE_NONE:
298 		sdev_printk(KERN_INFO, sdev, "%s: not supported\n",
299 			    ALUA_DH_NAME);
300 		break;
301 	default:
302 		sdev_printk(KERN_INFO, sdev,
303 			    "%s: unsupported TPGS setting %d\n",
304 			    ALUA_DH_NAME, tpgs);
305 		tpgs = TPGS_MODE_NONE;
306 		break;
307 	}
308 
309 	return tpgs;
310 }
311 
312 /*
313  * alua_check_vpd - Evaluate INQUIRY vpd page 0x83
314  * @sdev: device to be checked
315  *
316  * Extract the relative target port and the target port group
317  * descriptor from the list of identificators.
318  */
319 static int alua_check_vpd(struct scsi_device *sdev, struct alua_dh_data *h,
320 			  int tpgs)
321 {
322 	int rel_port = -1, group_id;
323 	struct alua_port_group *pg, *old_pg = NULL;
324 	bool pg_updated = false;
325 	unsigned long flags;
326 
327 	group_id = scsi_vpd_tpg_id(sdev, &rel_port);
328 	if (group_id < 0) {
329 		/*
330 		 * Internal error; TPGS supported but required
331 		 * VPD identification descriptors not present.
332 		 * Disable ALUA support
333 		 */
334 		sdev_printk(KERN_INFO, sdev,
335 			    "%s: No target port descriptors found\n",
336 			    ALUA_DH_NAME);
337 		return SCSI_DH_DEV_UNSUPP;
338 	}
339 
340 	pg = alua_alloc_pg(sdev, group_id, tpgs);
341 	if (IS_ERR(pg)) {
342 		if (PTR_ERR(pg) == -ENOMEM)
343 			return SCSI_DH_NOMEM;
344 		return SCSI_DH_DEV_UNSUPP;
345 	}
346 	if (pg->device_id_len)
347 		sdev_printk(KERN_INFO, sdev,
348 			    "%s: device %s port group %x rel port %x\n",
349 			    ALUA_DH_NAME, pg->device_id_str,
350 			    group_id, rel_port);
351 	else
352 		sdev_printk(KERN_INFO, sdev,
353 			    "%s: port group %x rel port %x\n",
354 			    ALUA_DH_NAME, group_id, rel_port);
355 
356 	/* Check for existing port group references */
357 	spin_lock(&h->pg_lock);
358 	old_pg = rcu_dereference_protected(h->pg, lockdep_is_held(&h->pg_lock));
359 	if (old_pg != pg) {
360 		/* port group has changed. Update to new port group */
361 		if (h->pg) {
362 			spin_lock_irqsave(&old_pg->lock, flags);
363 			list_del_rcu(&h->node);
364 			spin_unlock_irqrestore(&old_pg->lock, flags);
365 		}
366 		rcu_assign_pointer(h->pg, pg);
367 		pg_updated = true;
368 	}
369 
370 	spin_lock_irqsave(&pg->lock, flags);
371 	if (pg_updated)
372 		list_add_rcu(&h->node, &pg->dh_list);
373 	spin_unlock_irqrestore(&pg->lock, flags);
374 
375 	alua_rtpg_queue(rcu_dereference_protected(h->pg,
376 						  lockdep_is_held(&h->pg_lock)),
377 			sdev, NULL, true);
378 	spin_unlock(&h->pg_lock);
379 
380 	if (old_pg)
381 		kref_put(&old_pg->kref, release_port_group);
382 
383 	return SCSI_DH_OK;
384 }
385 
386 static char print_alua_state(unsigned char state)
387 {
388 	switch (state) {
389 	case SCSI_ACCESS_STATE_OPTIMAL:
390 		return 'A';
391 	case SCSI_ACCESS_STATE_ACTIVE:
392 		return 'N';
393 	case SCSI_ACCESS_STATE_STANDBY:
394 		return 'S';
395 	case SCSI_ACCESS_STATE_UNAVAILABLE:
396 		return 'U';
397 	case SCSI_ACCESS_STATE_LBA:
398 		return 'L';
399 	case SCSI_ACCESS_STATE_OFFLINE:
400 		return 'O';
401 	case SCSI_ACCESS_STATE_TRANSITIONING:
402 		return 'T';
403 	default:
404 		return 'X';
405 	}
406 }
407 
408 static int alua_check_sense(struct scsi_device *sdev,
409 			    struct scsi_sense_hdr *sense_hdr)
410 {
411 	switch (sense_hdr->sense_key) {
412 	case NOT_READY:
413 		if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a) {
414 			/*
415 			 * LUN Not Accessible - ALUA state transition
416 			 */
417 			alua_check(sdev, false);
418 			return NEEDS_RETRY;
419 		}
420 		break;
421 	case UNIT_ATTENTION:
422 		if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x00) {
423 			/*
424 			 * Power On, Reset, or Bus Device Reset.
425 			 * Might have obscured a state transition,
426 			 * so schedule a recheck.
427 			 */
428 			alua_check(sdev, true);
429 			return ADD_TO_MLQUEUE;
430 		}
431 		if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x04)
432 			/*
433 			 * Device internal reset
434 			 */
435 			return ADD_TO_MLQUEUE;
436 		if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x01)
437 			/*
438 			 * Mode Parameters Changed
439 			 */
440 			return ADD_TO_MLQUEUE;
441 		if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x06) {
442 			/*
443 			 * ALUA state changed
444 			 */
445 			alua_check(sdev, true);
446 			return ADD_TO_MLQUEUE;
447 		}
448 		if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x07) {
449 			/*
450 			 * Implicit ALUA state transition failed
451 			 */
452 			alua_check(sdev, true);
453 			return ADD_TO_MLQUEUE;
454 		}
455 		if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x03)
456 			/*
457 			 * Inquiry data has changed
458 			 */
459 			return ADD_TO_MLQUEUE;
460 		if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x0e)
461 			/*
462 			 * REPORTED_LUNS_DATA_HAS_CHANGED is reported
463 			 * when switching controllers on targets like
464 			 * Intel Multi-Flex. We can just retry.
465 			 */
466 			return ADD_TO_MLQUEUE;
467 		break;
468 	}
469 
470 	return SCSI_RETURN_NOT_HANDLED;
471 }
472 
473 /*
474  * alua_tur - Send a TEST UNIT READY
475  * @sdev: device to which the TEST UNIT READY command should be send
476  *
477  * Send a TEST UNIT READY to @sdev to figure out the device state
478  * Returns SCSI_DH_RETRY if the sense code is NOT READY/ALUA TRANSITIONING,
479  * SCSI_DH_OK if no error occurred, and SCSI_DH_IO otherwise.
480  */
481 static int alua_tur(struct scsi_device *sdev)
482 {
483 	struct scsi_sense_hdr sense_hdr;
484 	int retval;
485 
486 	retval = scsi_test_unit_ready(sdev, ALUA_FAILOVER_TIMEOUT * HZ,
487 				      ALUA_FAILOVER_RETRIES, &sense_hdr);
488 	if (sense_hdr.sense_key == NOT_READY &&
489 	    sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a)
490 		return SCSI_DH_RETRY;
491 	else if (retval)
492 		return SCSI_DH_IO;
493 	else
494 		return SCSI_DH_OK;
495 }
496 
497 /*
498  * alua_rtpg - Evaluate REPORT TARGET GROUP STATES
499  * @sdev: the device to be evaluated.
500  *
501  * Evaluate the Target Port Group State.
502  * Returns SCSI_DH_DEV_OFFLINED if the path is
503  * found to be unusable.
504  */
505 static int alua_rtpg(struct scsi_device *sdev, struct alua_port_group *pg)
506 {
507 	struct scsi_sense_hdr sense_hdr;
508 	struct alua_port_group *tmp_pg;
509 	int len, k, off, bufflen = ALUA_RTPG_SIZE;
510 	unsigned char *desc, *buff;
511 	unsigned err, retval;
512 	unsigned int tpg_desc_tbl_off;
513 	unsigned char orig_transition_tmo;
514 	unsigned long flags;
515 
516 	if (!pg->expiry) {
517 		unsigned long transition_tmo = ALUA_FAILOVER_TIMEOUT * HZ;
518 
519 		if (pg->transition_tmo)
520 			transition_tmo = pg->transition_tmo * HZ;
521 
522 		pg->expiry = round_jiffies_up(jiffies + transition_tmo);
523 	}
524 
525 	buff = kzalloc(bufflen, GFP_KERNEL);
526 	if (!buff)
527 		return SCSI_DH_DEV_TEMP_BUSY;
528 
529  retry:
530 	err = 0;
531 	retval = submit_rtpg(sdev, buff, bufflen, &sense_hdr, pg->flags);
532 
533 	if (retval) {
534 		/*
535 		 * Some (broken) implementations have a habit of returning
536 		 * an error during things like firmware update etc.
537 		 * But if the target only supports active/optimized there's
538 		 * not much we can do; it's not that we can switch paths
539 		 * or anything.
540 		 * So ignore any errors to avoid spurious failures during
541 		 * path failover.
542 		 */
543 		if ((pg->valid_states & ~TPGS_SUPPORT_OPTIMIZED) == 0) {
544 			sdev_printk(KERN_INFO, sdev,
545 				    "%s: ignoring rtpg result %d\n",
546 				    ALUA_DH_NAME, retval);
547 			kfree(buff);
548 			return SCSI_DH_OK;
549 		}
550 		if (!scsi_sense_valid(&sense_hdr)) {
551 			sdev_printk(KERN_INFO, sdev,
552 				    "%s: rtpg failed, result %d\n",
553 				    ALUA_DH_NAME, retval);
554 			kfree(buff);
555 			if (driver_byte(retval) == DRIVER_ERROR)
556 				return SCSI_DH_DEV_TEMP_BUSY;
557 			return SCSI_DH_IO;
558 		}
559 
560 		/*
561 		 * submit_rtpg() has failed on existing arrays
562 		 * when requesting extended header info, and
563 		 * the array doesn't support extended headers,
564 		 * even though it shouldn't according to T10.
565 		 * The retry without rtpg_ext_hdr_req set
566 		 * handles this.
567 		 */
568 		if (!(pg->flags & ALUA_RTPG_EXT_HDR_UNSUPP) &&
569 		    sense_hdr.sense_key == ILLEGAL_REQUEST &&
570 		    sense_hdr.asc == 0x24 && sense_hdr.ascq == 0) {
571 			pg->flags |= ALUA_RTPG_EXT_HDR_UNSUPP;
572 			goto retry;
573 		}
574 		/*
575 		 * Retry on ALUA state transition or if any
576 		 * UNIT ATTENTION occurred.
577 		 */
578 		if (sense_hdr.sense_key == NOT_READY &&
579 		    sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a)
580 			err = SCSI_DH_RETRY;
581 		else if (sense_hdr.sense_key == UNIT_ATTENTION)
582 			err = SCSI_DH_RETRY;
583 		if (err == SCSI_DH_RETRY &&
584 		    pg->expiry != 0 && time_before(jiffies, pg->expiry)) {
585 			sdev_printk(KERN_ERR, sdev, "%s: rtpg retry\n",
586 				    ALUA_DH_NAME);
587 			scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
588 			kfree(buff);
589 			return err;
590 		}
591 		sdev_printk(KERN_ERR, sdev, "%s: rtpg failed\n",
592 			    ALUA_DH_NAME);
593 		scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
594 		kfree(buff);
595 		pg->expiry = 0;
596 		return SCSI_DH_IO;
597 	}
598 
599 	len = get_unaligned_be32(&buff[0]) + 4;
600 
601 	if (len > bufflen) {
602 		/* Resubmit with the correct length */
603 		kfree(buff);
604 		bufflen = len;
605 		buff = kmalloc(bufflen, GFP_KERNEL);
606 		if (!buff) {
607 			sdev_printk(KERN_WARNING, sdev,
608 				    "%s: kmalloc buffer failed\n",__func__);
609 			/* Temporary failure, bypass */
610 			pg->expiry = 0;
611 			return SCSI_DH_DEV_TEMP_BUSY;
612 		}
613 		goto retry;
614 	}
615 
616 	orig_transition_tmo = pg->transition_tmo;
617 	if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR && buff[5] != 0)
618 		pg->transition_tmo = buff[5];
619 	else
620 		pg->transition_tmo = ALUA_FAILOVER_TIMEOUT;
621 
622 	if (orig_transition_tmo != pg->transition_tmo) {
623 		sdev_printk(KERN_INFO, sdev,
624 			    "%s: transition timeout set to %d seconds\n",
625 			    ALUA_DH_NAME, pg->transition_tmo);
626 		pg->expiry = jiffies + pg->transition_tmo * HZ;
627 	}
628 
629 	if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR)
630 		tpg_desc_tbl_off = 8;
631 	else
632 		tpg_desc_tbl_off = 4;
633 
634 	for (k = tpg_desc_tbl_off, desc = buff + tpg_desc_tbl_off;
635 	     k < len;
636 	     k += off, desc += off) {
637 		u16 group_id = get_unaligned_be16(&desc[2]);
638 
639 		spin_lock_irqsave(&port_group_lock, flags);
640 		tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len,
641 					  group_id);
642 		spin_unlock_irqrestore(&port_group_lock, flags);
643 		if (tmp_pg) {
644 			if (spin_trylock_irqsave(&tmp_pg->lock, flags)) {
645 				if ((tmp_pg == pg) ||
646 				    !(tmp_pg->flags & ALUA_PG_RUNNING)) {
647 					struct alua_dh_data *h;
648 
649 					tmp_pg->state = desc[0] & 0x0f;
650 					tmp_pg->pref = desc[0] >> 7;
651 					rcu_read_lock();
652 					list_for_each_entry_rcu(h,
653 						&tmp_pg->dh_list, node) {
654 						/* h->sdev should always be valid */
655 						BUG_ON(!h->sdev);
656 						h->sdev->access_state = desc[0];
657 					}
658 					rcu_read_unlock();
659 				}
660 				if (tmp_pg == pg)
661 					tmp_pg->valid_states = desc[1];
662 				spin_unlock_irqrestore(&tmp_pg->lock, flags);
663 			}
664 			kref_put(&tmp_pg->kref, release_port_group);
665 		}
666 		off = 8 + (desc[7] * 4);
667 	}
668 
669 	spin_lock_irqsave(&pg->lock, flags);
670 	sdev_printk(KERN_INFO, sdev,
671 		    "%s: port group %02x state %c %s supports %c%c%c%c%c%c%c\n",
672 		    ALUA_DH_NAME, pg->group_id, print_alua_state(pg->state),
673 		    pg->pref ? "preferred" : "non-preferred",
674 		    pg->valid_states&TPGS_SUPPORT_TRANSITION?'T':'t',
675 		    pg->valid_states&TPGS_SUPPORT_OFFLINE?'O':'o',
676 		    pg->valid_states&TPGS_SUPPORT_LBA_DEPENDENT?'L':'l',
677 		    pg->valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u',
678 		    pg->valid_states&TPGS_SUPPORT_STANDBY?'S':'s',
679 		    pg->valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n',
680 		    pg->valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a');
681 
682 	switch (pg->state) {
683 	case SCSI_ACCESS_STATE_TRANSITIONING:
684 		if (time_before(jiffies, pg->expiry)) {
685 			/* State transition, retry */
686 			pg->interval = ALUA_RTPG_RETRY_DELAY;
687 			err = SCSI_DH_RETRY;
688 		} else {
689 			struct alua_dh_data *h;
690 
691 			/* Transitioning time exceeded, set port to standby */
692 			err = SCSI_DH_IO;
693 			pg->state = SCSI_ACCESS_STATE_STANDBY;
694 			pg->expiry = 0;
695 			rcu_read_lock();
696 			list_for_each_entry_rcu(h, &pg->dh_list, node) {
697 				BUG_ON(!h->sdev);
698 				h->sdev->access_state =
699 					(pg->state & SCSI_ACCESS_STATE_MASK);
700 				if (pg->pref)
701 					h->sdev->access_state |=
702 						SCSI_ACCESS_STATE_PREFERRED;
703 			}
704 			rcu_read_unlock();
705 		}
706 		break;
707 	case SCSI_ACCESS_STATE_OFFLINE:
708 		/* Path unusable */
709 		err = SCSI_DH_DEV_OFFLINED;
710 		pg->expiry = 0;
711 		break;
712 	default:
713 		/* Useable path if active */
714 		err = SCSI_DH_OK;
715 		pg->expiry = 0;
716 		break;
717 	}
718 	spin_unlock_irqrestore(&pg->lock, flags);
719 	kfree(buff);
720 	return err;
721 }
722 
723 /*
724  * alua_stpg - Issue a SET TARGET PORT GROUP command
725  *
726  * Issue a SET TARGET PORT GROUP command and evaluate the
727  * response. Returns SCSI_DH_RETRY per default to trigger
728  * a re-evaluation of the target group state or SCSI_DH_OK
729  * if no further action needs to be taken.
730  */
731 static unsigned alua_stpg(struct scsi_device *sdev, struct alua_port_group *pg)
732 {
733 	int retval;
734 	struct scsi_sense_hdr sense_hdr;
735 
736 	if (!(pg->tpgs & TPGS_MODE_EXPLICIT)) {
737 		/* Only implicit ALUA supported, retry */
738 		return SCSI_DH_RETRY;
739 	}
740 	switch (pg->state) {
741 	case SCSI_ACCESS_STATE_OPTIMAL:
742 		return SCSI_DH_OK;
743 	case SCSI_ACCESS_STATE_ACTIVE:
744 		if ((pg->flags & ALUA_OPTIMIZE_STPG) &&
745 		    !pg->pref &&
746 		    (pg->tpgs & TPGS_MODE_IMPLICIT))
747 			return SCSI_DH_OK;
748 		break;
749 	case SCSI_ACCESS_STATE_STANDBY:
750 	case SCSI_ACCESS_STATE_UNAVAILABLE:
751 		break;
752 	case SCSI_ACCESS_STATE_OFFLINE:
753 		return SCSI_DH_IO;
754 	case SCSI_ACCESS_STATE_TRANSITIONING:
755 		break;
756 	default:
757 		sdev_printk(KERN_INFO, sdev,
758 			    "%s: stpg failed, unhandled TPGS state %d",
759 			    ALUA_DH_NAME, pg->state);
760 		return SCSI_DH_NOSYS;
761 	}
762 	retval = submit_stpg(sdev, pg->group_id, &sense_hdr);
763 
764 	if (retval) {
765 		if (!scsi_sense_valid(&sense_hdr)) {
766 			sdev_printk(KERN_INFO, sdev,
767 				    "%s: stpg failed, result %d",
768 				    ALUA_DH_NAME, retval);
769 			if (driver_byte(retval) == DRIVER_ERROR)
770 				return SCSI_DH_DEV_TEMP_BUSY;
771 		} else {
772 			sdev_printk(KERN_INFO, sdev, "%s: stpg failed\n",
773 				    ALUA_DH_NAME);
774 			scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
775 		}
776 	}
777 	/* Retry RTPG */
778 	return SCSI_DH_RETRY;
779 }
780 
781 static void alua_rtpg_work(struct work_struct *work)
782 {
783 	struct alua_port_group *pg =
784 		container_of(work, struct alua_port_group, rtpg_work.work);
785 	struct scsi_device *sdev;
786 	LIST_HEAD(qdata_list);
787 	int err = SCSI_DH_OK;
788 	struct alua_queue_data *qdata, *tmp;
789 	unsigned long flags;
790 
791 	spin_lock_irqsave(&pg->lock, flags);
792 	sdev = pg->rtpg_sdev;
793 	if (!sdev) {
794 		WARN_ON(pg->flags & ALUA_PG_RUN_RTPG);
795 		WARN_ON(pg->flags & ALUA_PG_RUN_STPG);
796 		spin_unlock_irqrestore(&pg->lock, flags);
797 		kref_put(&pg->kref, release_port_group);
798 		return;
799 	}
800 	pg->flags |= ALUA_PG_RUNNING;
801 	if (pg->flags & ALUA_PG_RUN_RTPG) {
802 		int state = pg->state;
803 
804 		pg->flags &= ~ALUA_PG_RUN_RTPG;
805 		spin_unlock_irqrestore(&pg->lock, flags);
806 		if (state == SCSI_ACCESS_STATE_TRANSITIONING) {
807 			if (alua_tur(sdev) == SCSI_DH_RETRY) {
808 				spin_lock_irqsave(&pg->lock, flags);
809 				pg->flags &= ~ALUA_PG_RUNNING;
810 				pg->flags |= ALUA_PG_RUN_RTPG;
811 				if (!pg->interval)
812 					pg->interval = ALUA_RTPG_RETRY_DELAY;
813 				spin_unlock_irqrestore(&pg->lock, flags);
814 				queue_delayed_work(kaluad_wq, &pg->rtpg_work,
815 						   pg->interval * HZ);
816 				return;
817 			}
818 			/* Send RTPG on failure or if TUR indicates SUCCESS */
819 		}
820 		err = alua_rtpg(sdev, pg);
821 		spin_lock_irqsave(&pg->lock, flags);
822 		if (err == SCSI_DH_RETRY || pg->flags & ALUA_PG_RUN_RTPG) {
823 			pg->flags &= ~ALUA_PG_RUNNING;
824 			if (!pg->interval && !(pg->flags & ALUA_PG_RUN_RTPG))
825 				pg->interval = ALUA_RTPG_RETRY_DELAY;
826 			pg->flags |= ALUA_PG_RUN_RTPG;
827 			spin_unlock_irqrestore(&pg->lock, flags);
828 			queue_delayed_work(kaluad_wq, &pg->rtpg_work,
829 					   pg->interval * HZ);
830 			return;
831 		}
832 		if (err != SCSI_DH_OK)
833 			pg->flags &= ~ALUA_PG_RUN_STPG;
834 	}
835 	if (pg->flags & ALUA_PG_RUN_STPG) {
836 		pg->flags &= ~ALUA_PG_RUN_STPG;
837 		spin_unlock_irqrestore(&pg->lock, flags);
838 		err = alua_stpg(sdev, pg);
839 		spin_lock_irqsave(&pg->lock, flags);
840 		if (err == SCSI_DH_RETRY || pg->flags & ALUA_PG_RUN_RTPG) {
841 			pg->flags |= ALUA_PG_RUN_RTPG;
842 			pg->interval = 0;
843 			pg->flags &= ~ALUA_PG_RUNNING;
844 			spin_unlock_irqrestore(&pg->lock, flags);
845 			queue_delayed_work(kaluad_wq, &pg->rtpg_work,
846 					   pg->interval * HZ);
847 			return;
848 		}
849 	}
850 
851 	list_splice_init(&pg->rtpg_list, &qdata_list);
852 	pg->rtpg_sdev = NULL;
853 	spin_unlock_irqrestore(&pg->lock, flags);
854 
855 	list_for_each_entry_safe(qdata, tmp, &qdata_list, entry) {
856 		list_del(&qdata->entry);
857 		if (qdata->callback_fn)
858 			qdata->callback_fn(qdata->callback_data, err);
859 		kfree(qdata);
860 	}
861 	spin_lock_irqsave(&pg->lock, flags);
862 	pg->flags &= ~ALUA_PG_RUNNING;
863 	spin_unlock_irqrestore(&pg->lock, flags);
864 	scsi_device_put(sdev);
865 	kref_put(&pg->kref, release_port_group);
866 }
867 
868 /**
869  * alua_rtpg_queue() - cause RTPG to be submitted asynchronously
870  * @pg: ALUA port group associated with @sdev.
871  * @sdev: SCSI device for which to submit an RTPG.
872  * @qdata: Information about the callback to invoke after the RTPG.
873  * @force: Whether or not to submit an RTPG if a work item that will submit an
874  *         RTPG already has been scheduled.
875  *
876  * Returns true if and only if alua_rtpg_work() will be called asynchronously.
877  * That function is responsible for calling @qdata->fn().
878  */
879 static bool alua_rtpg_queue(struct alua_port_group *pg,
880 			    struct scsi_device *sdev,
881 			    struct alua_queue_data *qdata, bool force)
882 {
883 	int start_queue = 0;
884 	unsigned long flags;
885 	if (WARN_ON_ONCE(!pg) || scsi_device_get(sdev))
886 		return false;
887 
888 	spin_lock_irqsave(&pg->lock, flags);
889 	if (qdata) {
890 		list_add_tail(&qdata->entry, &pg->rtpg_list);
891 		pg->flags |= ALUA_PG_RUN_STPG;
892 		force = true;
893 	}
894 	if (pg->rtpg_sdev == NULL) {
895 		pg->interval = 0;
896 		pg->flags |= ALUA_PG_RUN_RTPG;
897 		kref_get(&pg->kref);
898 		pg->rtpg_sdev = sdev;
899 		start_queue = 1;
900 	} else if (!(pg->flags & ALUA_PG_RUN_RTPG) && force) {
901 		pg->flags |= ALUA_PG_RUN_RTPG;
902 		/* Do not queue if the worker is already running */
903 		if (!(pg->flags & ALUA_PG_RUNNING)) {
904 			kref_get(&pg->kref);
905 			start_queue = 1;
906 		}
907 	}
908 
909 	spin_unlock_irqrestore(&pg->lock, flags);
910 
911 	if (start_queue) {
912 		if (queue_delayed_work(kaluad_wq, &pg->rtpg_work,
913 				msecs_to_jiffies(ALUA_RTPG_DELAY_MSECS)))
914 			sdev = NULL;
915 		else
916 			kref_put(&pg->kref, release_port_group);
917 	}
918 	if (sdev)
919 		scsi_device_put(sdev);
920 
921 	return true;
922 }
923 
924 /*
925  * alua_initialize - Initialize ALUA state
926  * @sdev: the device to be initialized
927  *
928  * For the prep_fn to work correctly we have
929  * to initialize the ALUA state for the device.
930  */
931 static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h)
932 {
933 	int err = SCSI_DH_DEV_UNSUPP, tpgs;
934 
935 	mutex_lock(&h->init_mutex);
936 	tpgs = alua_check_tpgs(sdev);
937 	if (tpgs != TPGS_MODE_NONE)
938 		err = alua_check_vpd(sdev, h, tpgs);
939 	h->init_error = err;
940 	mutex_unlock(&h->init_mutex);
941 	return err;
942 }
943 /*
944  * alua_set_params - set/unset the optimize flag
945  * @sdev: device on the path to be activated
946  * params - parameters in the following format
947  *      "no_of_params\0param1\0param2\0param3\0...\0"
948  * For example, to set the flag pass the following parameters
949  * from multipath.conf
950  *     hardware_handler        "2 alua 1"
951  */
952 static int alua_set_params(struct scsi_device *sdev, const char *params)
953 {
954 	struct alua_dh_data *h = sdev->handler_data;
955 	struct alua_port_group *pg = NULL;
956 	unsigned int optimize = 0, argc;
957 	const char *p = params;
958 	int result = SCSI_DH_OK;
959 	unsigned long flags;
960 
961 	if ((sscanf(params, "%u", &argc) != 1) || (argc != 1))
962 		return -EINVAL;
963 
964 	while (*p++)
965 		;
966 	if ((sscanf(p, "%u", &optimize) != 1) || (optimize > 1))
967 		return -EINVAL;
968 
969 	rcu_read_lock();
970 	pg = rcu_dereference(h->pg);
971 	if (!pg) {
972 		rcu_read_unlock();
973 		return -ENXIO;
974 	}
975 	spin_lock_irqsave(&pg->lock, flags);
976 	if (optimize)
977 		pg->flags |= ALUA_OPTIMIZE_STPG;
978 	else
979 		pg->flags &= ~ALUA_OPTIMIZE_STPG;
980 	spin_unlock_irqrestore(&pg->lock, flags);
981 	rcu_read_unlock();
982 
983 	return result;
984 }
985 
986 /*
987  * alua_activate - activate a path
988  * @sdev: device on the path to be activated
989  *
990  * We're currently switching the port group to be activated only and
991  * let the array figure out the rest.
992  * There may be other arrays which require us to switch all port groups
993  * based on a certain policy. But until we actually encounter them it
994  * should be okay.
995  */
996 static int alua_activate(struct scsi_device *sdev,
997 			activate_complete fn, void *data)
998 {
999 	struct alua_dh_data *h = sdev->handler_data;
1000 	int err = SCSI_DH_OK;
1001 	struct alua_queue_data *qdata;
1002 	struct alua_port_group *pg;
1003 
1004 	qdata = kzalloc(sizeof(*qdata), GFP_KERNEL);
1005 	if (!qdata) {
1006 		err = SCSI_DH_RES_TEMP_UNAVAIL;
1007 		goto out;
1008 	}
1009 	qdata->callback_fn = fn;
1010 	qdata->callback_data = data;
1011 
1012 	mutex_lock(&h->init_mutex);
1013 	rcu_read_lock();
1014 	pg = rcu_dereference(h->pg);
1015 	if (!pg || !kref_get_unless_zero(&pg->kref)) {
1016 		rcu_read_unlock();
1017 		kfree(qdata);
1018 		err = h->init_error;
1019 		mutex_unlock(&h->init_mutex);
1020 		goto out;
1021 	}
1022 	rcu_read_unlock();
1023 	mutex_unlock(&h->init_mutex);
1024 
1025 	if (alua_rtpg_queue(pg, sdev, qdata, true))
1026 		fn = NULL;
1027 	else
1028 		err = SCSI_DH_DEV_OFFLINED;
1029 	kref_put(&pg->kref, release_port_group);
1030 out:
1031 	if (fn)
1032 		fn(data, err);
1033 	return 0;
1034 }
1035 
1036 /*
1037  * alua_check - check path status
1038  * @sdev: device on the path to be checked
1039  *
1040  * Check the device status
1041  */
1042 static void alua_check(struct scsi_device *sdev, bool force)
1043 {
1044 	struct alua_dh_data *h = sdev->handler_data;
1045 	struct alua_port_group *pg;
1046 
1047 	rcu_read_lock();
1048 	pg = rcu_dereference(h->pg);
1049 	if (!pg || !kref_get_unless_zero(&pg->kref)) {
1050 		rcu_read_unlock();
1051 		return;
1052 	}
1053 	rcu_read_unlock();
1054 
1055 	alua_rtpg_queue(pg, sdev, NULL, force);
1056 	kref_put(&pg->kref, release_port_group);
1057 }
1058 
1059 /*
1060  * alua_prep_fn - request callback
1061  *
1062  * Fail I/O to all paths not in state
1063  * active/optimized or active/non-optimized.
1064  */
1065 static blk_status_t alua_prep_fn(struct scsi_device *sdev, struct request *req)
1066 {
1067 	struct alua_dh_data *h = sdev->handler_data;
1068 	struct alua_port_group *pg;
1069 	unsigned char state = SCSI_ACCESS_STATE_OPTIMAL;
1070 
1071 	rcu_read_lock();
1072 	pg = rcu_dereference(h->pg);
1073 	if (pg)
1074 		state = pg->state;
1075 	rcu_read_unlock();
1076 
1077 	switch (state) {
1078 	case SCSI_ACCESS_STATE_OPTIMAL:
1079 	case SCSI_ACCESS_STATE_ACTIVE:
1080 	case SCSI_ACCESS_STATE_LBA:
1081 		return BLK_STS_OK;
1082 	case SCSI_ACCESS_STATE_TRANSITIONING:
1083 		return BLK_STS_RESOURCE;
1084 	default:
1085 		req->rq_flags |= RQF_QUIET;
1086 		return BLK_STS_IOERR;
1087 	}
1088 }
1089 
1090 static void alua_rescan(struct scsi_device *sdev)
1091 {
1092 	struct alua_dh_data *h = sdev->handler_data;
1093 
1094 	alua_initialize(sdev, h);
1095 }
1096 
1097 /*
1098  * alua_bus_attach - Attach device handler
1099  * @sdev: device to be attached to
1100  */
1101 static int alua_bus_attach(struct scsi_device *sdev)
1102 {
1103 	struct alua_dh_data *h;
1104 	int err;
1105 
1106 	h = kzalloc(sizeof(*h) , GFP_KERNEL);
1107 	if (!h)
1108 		return SCSI_DH_NOMEM;
1109 	spin_lock_init(&h->pg_lock);
1110 	rcu_assign_pointer(h->pg, NULL);
1111 	h->init_error = SCSI_DH_OK;
1112 	h->sdev = sdev;
1113 	INIT_LIST_HEAD(&h->node);
1114 
1115 	mutex_init(&h->init_mutex);
1116 	err = alua_initialize(sdev, h);
1117 	if (err != SCSI_DH_OK && err != SCSI_DH_DEV_OFFLINED)
1118 		goto failed;
1119 
1120 	sdev->handler_data = h;
1121 	return SCSI_DH_OK;
1122 failed:
1123 	kfree(h);
1124 	return err;
1125 }
1126 
1127 /*
1128  * alua_bus_detach - Detach device handler
1129  * @sdev: device to be detached from
1130  */
1131 static void alua_bus_detach(struct scsi_device *sdev)
1132 {
1133 	struct alua_dh_data *h = sdev->handler_data;
1134 	struct alua_port_group *pg;
1135 
1136 	spin_lock(&h->pg_lock);
1137 	pg = rcu_dereference_protected(h->pg, lockdep_is_held(&h->pg_lock));
1138 	rcu_assign_pointer(h->pg, NULL);
1139 	h->sdev = NULL;
1140 	spin_unlock(&h->pg_lock);
1141 	if (pg) {
1142 		spin_lock_irq(&pg->lock);
1143 		list_del_rcu(&h->node);
1144 		spin_unlock_irq(&pg->lock);
1145 		kref_put(&pg->kref, release_port_group);
1146 	}
1147 	sdev->handler_data = NULL;
1148 	kfree(h);
1149 }
1150 
1151 static struct scsi_device_handler alua_dh = {
1152 	.name = ALUA_DH_NAME,
1153 	.module = THIS_MODULE,
1154 	.attach = alua_bus_attach,
1155 	.detach = alua_bus_detach,
1156 	.prep_fn = alua_prep_fn,
1157 	.check_sense = alua_check_sense,
1158 	.activate = alua_activate,
1159 	.rescan = alua_rescan,
1160 	.set_params = alua_set_params,
1161 };
1162 
1163 static int __init alua_init(void)
1164 {
1165 	int r;
1166 
1167 	kaluad_wq = alloc_workqueue("kaluad", WQ_MEM_RECLAIM, 0);
1168 	if (!kaluad_wq)
1169 		return -ENOMEM;
1170 
1171 	r = scsi_register_device_handler(&alua_dh);
1172 	if (r != 0) {
1173 		printk(KERN_ERR "%s: Failed to register scsi device handler",
1174 			ALUA_DH_NAME);
1175 		destroy_workqueue(kaluad_wq);
1176 	}
1177 	return r;
1178 }
1179 
1180 static void __exit alua_exit(void)
1181 {
1182 	scsi_unregister_device_handler(&alua_dh);
1183 	destroy_workqueue(kaluad_wq);
1184 }
1185 
1186 module_init(alua_init);
1187 module_exit(alua_exit);
1188 
1189 MODULE_DESCRIPTION("DM Multipath ALUA support");
1190 MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
1191 MODULE_LICENSE("GPL");
1192 MODULE_VERSION(ALUA_DH_VER);
1193