1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Generic SCSI-3 ALUA SCSI Device Handler
5 * Copyright (C) 2007-2010 Hannes Reinecke, SUSE Linux Products GmbH.
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>
18 #define ALUA_DH_NAME "alua"
19 #define ALUA_DH_VER "2.0"
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
31 #define RTPG_FMT_MASK 0x70
32 #define RTPG_FMT_EXT_HDR 0x10
34 #define TPGS_MODE_UNINITIALIZED -1
35 #define TPGS_MODE_NONE 0x0
36 #define TPGS_MODE_IMPLICIT 0x1
37 #define TPGS_MODE_EXPLICIT 0x2
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
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
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.");
57 static LIST_HEAD(port_group_list);
58 static DEFINE_SPINLOCK(port_group_lock);
59 static struct workqueue_struct *kaluad_wq;
61 struct alua_port_group {
64 struct list_head node;
65 struct list_head dh_list;
66 unsigned char device_id_str[256];
73 unsigned flags; /* used for optimizing STPG */
74 unsigned char transition_tmo;
76 unsigned long interval;
77 struct delayed_work rtpg_work;
79 struct list_head rtpg_list;
80 struct scsi_device *rtpg_sdev;
84 struct list_head node;
85 struct alua_port_group __rcu *pg;
88 struct scsi_device *sdev;
90 struct mutex init_mutex;
94 struct alua_queue_data {
95 struct list_head entry;
96 activate_complete callback_fn;
100 #define ALUA_POLICY_SWITCH_CURRENT 0
101 #define ALUA_POLICY_SWITCH_ALL 1
103 static void alua_rtpg_work(struct work_struct *work);
104 static bool alua_rtpg_queue(struct alua_port_group *pg,
105 struct scsi_device *sdev,
106 struct alua_queue_data *qdata, bool force);
107 static void alua_check(struct scsi_device *sdev, bool force);
109 static void release_port_group(struct kref *kref)
111 struct alua_port_group *pg;
113 pg = container_of(kref, struct alua_port_group, kref);
115 flush_delayed_work(&pg->rtpg_work);
116 spin_lock(&port_group_lock);
118 spin_unlock(&port_group_lock);
123 * submit_rtpg - Issue a REPORT TARGET GROUP STATES command
124 * @sdev: sdev the command should be sent to
126 static int submit_rtpg(struct scsi_device *sdev, unsigned char *buff,
127 int bufflen, struct scsi_sense_hdr *sshdr, int flags)
129 u8 cdb[MAX_COMMAND_SIZE];
130 int req_flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
133 /* Prepare the command. */
134 memset(cdb, 0x0, MAX_COMMAND_SIZE);
135 cdb[0] = MAINTENANCE_IN;
136 if (!(flags & ALUA_RTPG_EXT_HDR_UNSUPP))
137 cdb[1] = MI_REPORT_TARGET_PGS | MI_EXT_HDR_PARAM_FMT;
139 cdb[1] = MI_REPORT_TARGET_PGS;
140 put_unaligned_be32(bufflen, &cdb[6]);
142 return scsi_execute(sdev, cdb, DMA_FROM_DEVICE, buff, bufflen, NULL,
143 sshdr, ALUA_FAILOVER_TIMEOUT * HZ,
144 ALUA_FAILOVER_RETRIES, req_flags, 0, NULL);
148 * submit_stpg - Issue a SET TARGET PORT GROUP command
150 * Currently we're only setting the current target port group state
151 * to 'active/optimized' and let the array firmware figure out
152 * the states of the remaining groups.
154 static int submit_stpg(struct scsi_device *sdev, int group_id,
155 struct scsi_sense_hdr *sshdr)
157 u8 cdb[MAX_COMMAND_SIZE];
158 unsigned char stpg_data[8];
160 int req_flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
163 /* Prepare the data buffer */
164 memset(stpg_data, 0, stpg_len);
165 stpg_data[4] = SCSI_ACCESS_STATE_OPTIMAL;
166 put_unaligned_be16(group_id, &stpg_data[6]);
168 /* Prepare the command. */
169 memset(cdb, 0x0, MAX_COMMAND_SIZE);
170 cdb[0] = MAINTENANCE_OUT;
171 cdb[1] = MO_SET_TARGET_PGS;
172 put_unaligned_be32(stpg_len, &cdb[6]);
174 return scsi_execute(sdev, cdb, DMA_TO_DEVICE, stpg_data, stpg_len, NULL,
175 sshdr, ALUA_FAILOVER_TIMEOUT * HZ,
176 ALUA_FAILOVER_RETRIES, req_flags, 0, NULL);
179 static struct alua_port_group *alua_find_get_pg(char *id_str, size_t id_size,
182 struct alua_port_group *pg;
184 if (!id_str || !id_size || !strlen(id_str))
187 list_for_each_entry(pg, &port_group_list, node) {
188 if (pg->group_id != group_id)
190 if (!pg->device_id_len || pg->device_id_len != id_size)
192 if (strncmp(pg->device_id_str, id_str, id_size))
194 if (!kref_get_unless_zero(&pg->kref))
203 * alua_alloc_pg - Allocate a new port_group structure
205 * @group_id: port group id
206 * @tpgs: target port group settings
208 * Allocate a new port_group structure for a given
211 static struct alua_port_group *alua_alloc_pg(struct scsi_device *sdev,
212 int group_id, int tpgs)
214 struct alua_port_group *pg, *tmp_pg;
216 pg = kzalloc(sizeof(struct alua_port_group), GFP_KERNEL);
218 return ERR_PTR(-ENOMEM);
220 pg->device_id_len = scsi_vpd_lun_id(sdev, pg->device_id_str,
221 sizeof(pg->device_id_str));
222 if (pg->device_id_len <= 0) {
224 * TPGS supported but no device identification found.
225 * Generate private device identification.
227 sdev_printk(KERN_INFO, sdev,
228 "%s: No device descriptors found\n",
230 pg->device_id_str[0] = '\0';
231 pg->device_id_len = 0;
233 pg->group_id = group_id;
235 pg->state = SCSI_ACCESS_STATE_OPTIMAL;
236 pg->valid_states = TPGS_SUPPORT_ALL;
238 pg->flags |= ALUA_OPTIMIZE_STPG;
239 kref_init(&pg->kref);
240 INIT_DELAYED_WORK(&pg->rtpg_work, alua_rtpg_work);
241 INIT_LIST_HEAD(&pg->rtpg_list);
242 INIT_LIST_HEAD(&pg->node);
243 INIT_LIST_HEAD(&pg->dh_list);
244 spin_lock_init(&pg->lock);
246 spin_lock(&port_group_lock);
247 tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len,
250 spin_unlock(&port_group_lock);
255 list_add(&pg->node, &port_group_list);
256 spin_unlock(&port_group_lock);
262 * alua_check_tpgs - Evaluate TPGS setting
263 * @sdev: device to be checked
265 * Examine the TPGS setting of the sdev to find out if ALUA
268 static int alua_check_tpgs(struct scsi_device *sdev)
270 int tpgs = TPGS_MODE_NONE;
273 * ALUA support for non-disk devices is fraught with
274 * difficulties, so disable it for now.
276 if (sdev->type != TYPE_DISK) {
277 sdev_printk(KERN_INFO, sdev,
278 "%s: disable for non-disk devices\n",
283 tpgs = scsi_device_tpgs(sdev);
285 case TPGS_MODE_EXPLICIT|TPGS_MODE_IMPLICIT:
286 sdev_printk(KERN_INFO, sdev,
287 "%s: supports implicit and explicit TPGS\n",
290 case TPGS_MODE_EXPLICIT:
291 sdev_printk(KERN_INFO, sdev, "%s: supports explicit TPGS\n",
294 case TPGS_MODE_IMPLICIT:
295 sdev_printk(KERN_INFO, sdev, "%s: supports implicit TPGS\n",
299 sdev_printk(KERN_INFO, sdev, "%s: not supported\n",
303 sdev_printk(KERN_INFO, sdev,
304 "%s: unsupported TPGS setting %d\n",
306 tpgs = TPGS_MODE_NONE;
314 * alua_check_vpd - Evaluate INQUIRY vpd page 0x83
315 * @sdev: device to be checked
317 * Extract the relative target port and the target port group
318 * descriptor from the list of identificators.
320 static int alua_check_vpd(struct scsi_device *sdev, struct alua_dh_data *h,
323 int rel_port = -1, group_id;
324 struct alua_port_group *pg, *old_pg = NULL;
325 bool pg_updated = false;
328 group_id = scsi_vpd_tpg_id(sdev, &rel_port);
331 * Internal error; TPGS supported but required
332 * VPD identification descriptors not present.
333 * Disable ALUA support
335 sdev_printk(KERN_INFO, sdev,
336 "%s: No target port descriptors found\n",
338 return SCSI_DH_DEV_UNSUPP;
341 pg = alua_alloc_pg(sdev, group_id, tpgs);
343 if (PTR_ERR(pg) == -ENOMEM)
344 return SCSI_DH_NOMEM;
345 return SCSI_DH_DEV_UNSUPP;
347 if (pg->device_id_len)
348 sdev_printk(KERN_INFO, sdev,
349 "%s: device %s port group %x rel port %x\n",
350 ALUA_DH_NAME, pg->device_id_str,
353 sdev_printk(KERN_INFO, sdev,
354 "%s: port group %x rel port %x\n",
355 ALUA_DH_NAME, group_id, rel_port);
357 /* Check for existing port group references */
358 spin_lock(&h->pg_lock);
359 old_pg = rcu_dereference_protected(h->pg, lockdep_is_held(&h->pg_lock));
361 /* port group has changed. Update to new port group */
363 spin_lock_irqsave(&old_pg->lock, flags);
364 list_del_rcu(&h->node);
365 spin_unlock_irqrestore(&old_pg->lock, flags);
367 rcu_assign_pointer(h->pg, pg);
371 spin_lock_irqsave(&pg->lock, flags);
373 list_add_rcu(&h->node, &pg->dh_list);
374 spin_unlock_irqrestore(&pg->lock, flags);
376 alua_rtpg_queue(rcu_dereference_protected(h->pg,
377 lockdep_is_held(&h->pg_lock)),
379 spin_unlock(&h->pg_lock);
382 kref_put(&old_pg->kref, release_port_group);
387 static char print_alua_state(unsigned char state)
390 case SCSI_ACCESS_STATE_OPTIMAL:
392 case SCSI_ACCESS_STATE_ACTIVE:
394 case SCSI_ACCESS_STATE_STANDBY:
396 case SCSI_ACCESS_STATE_UNAVAILABLE:
398 case SCSI_ACCESS_STATE_LBA:
400 case SCSI_ACCESS_STATE_OFFLINE:
402 case SCSI_ACCESS_STATE_TRANSITIONING:
409 static enum scsi_disposition alua_check_sense(struct scsi_device *sdev,
410 struct scsi_sense_hdr *sense_hdr)
412 struct alua_dh_data *h = sdev->handler_data;
413 struct alua_port_group *pg;
415 switch (sense_hdr->sense_key) {
417 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a) {
419 * LUN Not Accessible - ALUA state transition
422 pg = rcu_dereference(h->pg);
424 pg->state = SCSI_ACCESS_STATE_TRANSITIONING;
426 alua_check(sdev, false);
431 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x00) {
433 * Power On, Reset, or Bus Device Reset.
434 * Might have obscured a state transition,
435 * so schedule a recheck.
437 alua_check(sdev, true);
438 return ADD_TO_MLQUEUE;
440 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x04)
442 * Device internal reset
444 return ADD_TO_MLQUEUE;
445 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x01)
447 * Mode Parameters Changed
449 return ADD_TO_MLQUEUE;
450 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x06) {
454 alua_check(sdev, true);
455 return ADD_TO_MLQUEUE;
457 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x07) {
459 * Implicit ALUA state transition failed
461 alua_check(sdev, true);
462 return ADD_TO_MLQUEUE;
464 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x03)
466 * Inquiry data has changed
468 return ADD_TO_MLQUEUE;
469 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x0e)
471 * REPORTED_LUNS_DATA_HAS_CHANGED is reported
472 * when switching controllers on targets like
473 * Intel Multi-Flex. We can just retry.
475 return ADD_TO_MLQUEUE;
479 return SCSI_RETURN_NOT_HANDLED;
483 * alua_tur - Send a TEST UNIT READY
484 * @sdev: device to which the TEST UNIT READY command should be send
486 * Send a TEST UNIT READY to @sdev to figure out the device state
487 * Returns SCSI_DH_RETRY if the sense code is NOT READY/ALUA TRANSITIONING,
488 * SCSI_DH_OK if no error occurred, and SCSI_DH_IO otherwise.
490 static int alua_tur(struct scsi_device *sdev)
492 struct scsi_sense_hdr sense_hdr;
495 retval = scsi_test_unit_ready(sdev, ALUA_FAILOVER_TIMEOUT * HZ,
496 ALUA_FAILOVER_RETRIES, &sense_hdr);
497 if (sense_hdr.sense_key == NOT_READY &&
498 sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a)
499 return SCSI_DH_RETRY;
507 * alua_rtpg - Evaluate REPORT TARGET GROUP STATES
508 * @sdev: the device to be evaluated.
510 * Evaluate the Target Port Group State.
511 * Returns SCSI_DH_DEV_OFFLINED if the path is
512 * found to be unusable.
514 static int alua_rtpg(struct scsi_device *sdev, struct alua_port_group *pg)
516 struct scsi_sense_hdr sense_hdr;
517 struct alua_port_group *tmp_pg;
518 int len, k, off, bufflen = ALUA_RTPG_SIZE;
519 int group_id_old, state_old, pref_old, valid_states_old;
520 unsigned char *desc, *buff;
523 unsigned int tpg_desc_tbl_off;
524 unsigned char orig_transition_tmo;
526 bool transitioning_sense = false;
528 group_id_old = pg->group_id;
529 state_old = pg->state;
531 valid_states_old = pg->valid_states;
534 unsigned long transition_tmo = ALUA_FAILOVER_TIMEOUT * HZ;
536 if (pg->transition_tmo)
537 transition_tmo = pg->transition_tmo * HZ;
539 pg->expiry = round_jiffies_up(jiffies + transition_tmo);
542 buff = kzalloc(bufflen, GFP_KERNEL);
544 return SCSI_DH_DEV_TEMP_BUSY;
548 retval = submit_rtpg(sdev, buff, bufflen, &sense_hdr, pg->flags);
552 * Some (broken) implementations have a habit of returning
553 * an error during things like firmware update etc.
554 * But if the target only supports active/optimized there's
555 * not much we can do; it's not that we can switch paths
557 * So ignore any errors to avoid spurious failures during
560 if ((pg->valid_states & ~TPGS_SUPPORT_OPTIMIZED) == 0) {
561 sdev_printk(KERN_INFO, sdev,
562 "%s: ignoring rtpg result %d\n",
563 ALUA_DH_NAME, retval);
567 if (retval < 0 || !scsi_sense_valid(&sense_hdr)) {
568 sdev_printk(KERN_INFO, sdev,
569 "%s: rtpg failed, result %d\n",
570 ALUA_DH_NAME, retval);
573 return SCSI_DH_DEV_TEMP_BUSY;
574 if (host_byte(retval) == DID_NO_CONNECT)
575 return SCSI_DH_RES_TEMP_UNAVAIL;
580 * submit_rtpg() has failed on existing arrays
581 * when requesting extended header info, and
582 * the array doesn't support extended headers,
583 * even though it shouldn't according to T10.
584 * The retry without rtpg_ext_hdr_req set
586 * Note: some arrays return a sense key of ILLEGAL_REQUEST
587 * with ASC 00h if they don't support the extended header.
589 if (!(pg->flags & ALUA_RTPG_EXT_HDR_UNSUPP) &&
590 sense_hdr.sense_key == ILLEGAL_REQUEST) {
591 pg->flags |= ALUA_RTPG_EXT_HDR_UNSUPP;
595 * If the array returns with 'ALUA state transition'
596 * sense code here it cannot return RTPG data during
597 * transition. So set the state to 'transitioning' directly.
599 if (sense_hdr.sense_key == NOT_READY &&
600 sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a) {
601 transitioning_sense = true;
605 * Retry on any other UNIT ATTENTION occurred.
607 if (sense_hdr.sense_key == UNIT_ATTENTION)
609 if (err == SCSI_DH_RETRY &&
610 pg->expiry != 0 && time_before(jiffies, pg->expiry)) {
611 sdev_printk(KERN_ERR, sdev, "%s: rtpg retry\n",
613 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
617 sdev_printk(KERN_ERR, sdev, "%s: rtpg failed\n",
619 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
625 len = get_unaligned_be32(&buff[0]) + 4;
628 /* Resubmit with the correct length */
631 buff = kmalloc(bufflen, GFP_KERNEL);
633 sdev_printk(KERN_WARNING, sdev,
634 "%s: kmalloc buffer failed\n",__func__);
635 /* Temporary failure, bypass */
637 return SCSI_DH_DEV_TEMP_BUSY;
642 orig_transition_tmo = pg->transition_tmo;
643 if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR && buff[5] != 0)
644 pg->transition_tmo = buff[5];
646 pg->transition_tmo = ALUA_FAILOVER_TIMEOUT;
648 if (orig_transition_tmo != pg->transition_tmo) {
649 sdev_printk(KERN_INFO, sdev,
650 "%s: transition timeout set to %d seconds\n",
651 ALUA_DH_NAME, pg->transition_tmo);
652 pg->expiry = jiffies + pg->transition_tmo * HZ;
655 if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR)
656 tpg_desc_tbl_off = 8;
658 tpg_desc_tbl_off = 4;
660 for (k = tpg_desc_tbl_off, desc = buff + tpg_desc_tbl_off;
662 k += off, desc += off) {
663 u16 group_id = get_unaligned_be16(&desc[2]);
665 spin_lock_irqsave(&port_group_lock, flags);
666 tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len,
668 spin_unlock_irqrestore(&port_group_lock, flags);
670 if (spin_trylock_irqsave(&tmp_pg->lock, flags)) {
671 if ((tmp_pg == pg) ||
672 !(tmp_pg->flags & ALUA_PG_RUNNING)) {
673 struct alua_dh_data *h;
675 tmp_pg->state = desc[0] & 0x0f;
676 tmp_pg->pref = desc[0] >> 7;
678 list_for_each_entry_rcu(h,
679 &tmp_pg->dh_list, node) {
682 h->sdev->access_state = desc[0];
687 tmp_pg->valid_states = desc[1];
688 spin_unlock_irqrestore(&tmp_pg->lock, flags);
690 kref_put(&tmp_pg->kref, release_port_group);
692 off = 8 + (desc[7] * 4);
696 spin_lock_irqsave(&pg->lock, flags);
697 if (transitioning_sense)
698 pg->state = SCSI_ACCESS_STATE_TRANSITIONING;
700 if (group_id_old != pg->group_id || state_old != pg->state ||
701 pref_old != pg->pref || valid_states_old != pg->valid_states)
702 sdev_printk(KERN_INFO, sdev,
703 "%s: port group %02x state %c %s supports %c%c%c%c%c%c%c\n",
704 ALUA_DH_NAME, pg->group_id, print_alua_state(pg->state),
705 pg->pref ? "preferred" : "non-preferred",
706 pg->valid_states&TPGS_SUPPORT_TRANSITION?'T':'t',
707 pg->valid_states&TPGS_SUPPORT_OFFLINE?'O':'o',
708 pg->valid_states&TPGS_SUPPORT_LBA_DEPENDENT?'L':'l',
709 pg->valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u',
710 pg->valid_states&TPGS_SUPPORT_STANDBY?'S':'s',
711 pg->valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n',
712 pg->valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a');
715 case SCSI_ACCESS_STATE_TRANSITIONING:
716 if (time_before(jiffies, pg->expiry)) {
717 /* State transition, retry */
718 pg->interval = ALUA_RTPG_RETRY_DELAY;
721 struct alua_dh_data *h;
723 /* Transitioning time exceeded, set port to standby */
725 pg->state = SCSI_ACCESS_STATE_STANDBY;
728 list_for_each_entry_rcu(h, &pg->dh_list, node) {
731 h->sdev->access_state =
732 (pg->state & SCSI_ACCESS_STATE_MASK);
734 h->sdev->access_state |=
735 SCSI_ACCESS_STATE_PREFERRED;
740 case SCSI_ACCESS_STATE_OFFLINE:
742 err = SCSI_DH_DEV_OFFLINED;
746 /* Useable path if active */
751 spin_unlock_irqrestore(&pg->lock, flags);
757 * alua_stpg - Issue a SET TARGET PORT GROUP command
759 * Issue a SET TARGET PORT GROUP command and evaluate the
760 * response. Returns SCSI_DH_RETRY per default to trigger
761 * a re-evaluation of the target group state or SCSI_DH_OK
762 * if no further action needs to be taken.
764 static unsigned alua_stpg(struct scsi_device *sdev, struct alua_port_group *pg)
767 struct scsi_sense_hdr sense_hdr;
769 if (!(pg->tpgs & TPGS_MODE_EXPLICIT)) {
770 /* Only implicit ALUA supported, retry */
771 return SCSI_DH_RETRY;
774 case SCSI_ACCESS_STATE_OPTIMAL:
776 case SCSI_ACCESS_STATE_ACTIVE:
777 if ((pg->flags & ALUA_OPTIMIZE_STPG) &&
779 (pg->tpgs & TPGS_MODE_IMPLICIT))
782 case SCSI_ACCESS_STATE_STANDBY:
783 case SCSI_ACCESS_STATE_UNAVAILABLE:
785 case SCSI_ACCESS_STATE_OFFLINE:
787 case SCSI_ACCESS_STATE_TRANSITIONING:
790 sdev_printk(KERN_INFO, sdev,
791 "%s: stpg failed, unhandled TPGS state %d",
792 ALUA_DH_NAME, pg->state);
793 return SCSI_DH_NOSYS;
795 retval = submit_stpg(sdev, pg->group_id, &sense_hdr);
798 if (retval < 0 || !scsi_sense_valid(&sense_hdr)) {
799 sdev_printk(KERN_INFO, sdev,
800 "%s: stpg failed, result %d",
801 ALUA_DH_NAME, retval);
803 return SCSI_DH_DEV_TEMP_BUSY;
805 sdev_printk(KERN_INFO, sdev, "%s: stpg failed\n",
807 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
811 return SCSI_DH_RETRY;
814 static bool alua_rtpg_select_sdev(struct alua_port_group *pg)
816 struct alua_dh_data *h;
817 struct scsi_device *sdev = NULL;
819 lockdep_assert_held(&pg->lock);
820 if (WARN_ON(!pg->rtpg_sdev))
824 * RCU protection isn't necessary for dh_list here
825 * as we hold pg->lock, but for access to h->pg.
828 list_for_each_entry_rcu(h, &pg->dh_list, node) {
831 if (h->sdev == pg->rtpg_sdev) {
835 if (rcu_dereference(h->pg) == pg &&
837 !scsi_device_get(h->sdev)) {
845 pr_warn("%s: no device found for rtpg\n",
847 (char *)pg->device_id_str : "(nameless PG)"));
851 sdev_printk(KERN_INFO, sdev, "rtpg retry on different device\n");
853 scsi_device_put(pg->rtpg_sdev);
854 pg->rtpg_sdev = sdev;
859 static void alua_rtpg_work(struct work_struct *work)
861 struct alua_port_group *pg =
862 container_of(work, struct alua_port_group, rtpg_work.work);
863 struct scsi_device *sdev;
864 LIST_HEAD(qdata_list);
865 int err = SCSI_DH_OK;
866 struct alua_queue_data *qdata, *tmp;
867 struct alua_dh_data *h;
870 spin_lock_irqsave(&pg->lock, flags);
871 sdev = pg->rtpg_sdev;
873 WARN_ON(pg->flags & ALUA_PG_RUN_RTPG);
874 WARN_ON(pg->flags & ALUA_PG_RUN_STPG);
875 spin_unlock_irqrestore(&pg->lock, flags);
876 kref_put(&pg->kref, release_port_group);
879 pg->flags |= ALUA_PG_RUNNING;
880 if (pg->flags & ALUA_PG_RUN_RTPG) {
881 int state = pg->state;
883 pg->flags &= ~ALUA_PG_RUN_RTPG;
884 spin_unlock_irqrestore(&pg->lock, flags);
885 if (state == SCSI_ACCESS_STATE_TRANSITIONING) {
886 if (alua_tur(sdev) == SCSI_DH_RETRY) {
887 spin_lock_irqsave(&pg->lock, flags);
888 pg->flags &= ~ALUA_PG_RUNNING;
889 pg->flags |= ALUA_PG_RUN_RTPG;
891 pg->interval = ALUA_RTPG_RETRY_DELAY;
892 spin_unlock_irqrestore(&pg->lock, flags);
893 queue_delayed_work(kaluad_wq, &pg->rtpg_work,
897 /* Send RTPG on failure or if TUR indicates SUCCESS */
899 err = alua_rtpg(sdev, pg);
900 spin_lock_irqsave(&pg->lock, flags);
902 /* If RTPG failed on the current device, try using another */
903 if (err == SCSI_DH_RES_TEMP_UNAVAIL &&
904 alua_rtpg_select_sdev(pg))
905 err = SCSI_DH_IMM_RETRY;
907 if (err == SCSI_DH_RETRY || err == SCSI_DH_IMM_RETRY ||
908 pg->flags & ALUA_PG_RUN_RTPG) {
909 pg->flags &= ~ALUA_PG_RUNNING;
910 if (err == SCSI_DH_IMM_RETRY)
912 else if (!pg->interval && !(pg->flags & ALUA_PG_RUN_RTPG))
913 pg->interval = ALUA_RTPG_RETRY_DELAY;
914 pg->flags |= ALUA_PG_RUN_RTPG;
915 spin_unlock_irqrestore(&pg->lock, flags);
916 queue_delayed_work(kaluad_wq, &pg->rtpg_work,
920 if (err != SCSI_DH_OK)
921 pg->flags &= ~ALUA_PG_RUN_STPG;
923 if (pg->flags & ALUA_PG_RUN_STPG) {
924 pg->flags &= ~ALUA_PG_RUN_STPG;
925 spin_unlock_irqrestore(&pg->lock, flags);
926 err = alua_stpg(sdev, pg);
927 spin_lock_irqsave(&pg->lock, flags);
928 if (err == SCSI_DH_RETRY || pg->flags & ALUA_PG_RUN_RTPG) {
929 pg->flags |= ALUA_PG_RUN_RTPG;
931 pg->flags &= ~ALUA_PG_RUNNING;
932 spin_unlock_irqrestore(&pg->lock, flags);
933 queue_delayed_work(kaluad_wq, &pg->rtpg_work,
939 list_splice_init(&pg->rtpg_list, &qdata_list);
941 * We went through an RTPG, for good or bad.
942 * Re-enable all devices for the next attempt.
944 list_for_each_entry(h, &pg->dh_list, node)
946 pg->rtpg_sdev = NULL;
947 spin_unlock_irqrestore(&pg->lock, flags);
949 list_for_each_entry_safe(qdata, tmp, &qdata_list, entry) {
950 list_del(&qdata->entry);
951 if (qdata->callback_fn)
952 qdata->callback_fn(qdata->callback_data, err);
955 spin_lock_irqsave(&pg->lock, flags);
956 pg->flags &= ~ALUA_PG_RUNNING;
957 spin_unlock_irqrestore(&pg->lock, flags);
958 scsi_device_put(sdev);
959 kref_put(&pg->kref, release_port_group);
963 * alua_rtpg_queue() - cause RTPG to be submitted asynchronously
964 * @pg: ALUA port group associated with @sdev.
965 * @sdev: SCSI device for which to submit an RTPG.
966 * @qdata: Information about the callback to invoke after the RTPG.
967 * @force: Whether or not to submit an RTPG if a work item that will submit an
968 * RTPG already has been scheduled.
970 * Returns true if and only if alua_rtpg_work() will be called asynchronously.
971 * That function is responsible for calling @qdata->fn().
973 static bool alua_rtpg_queue(struct alua_port_group *pg,
974 struct scsi_device *sdev,
975 struct alua_queue_data *qdata, bool force)
979 if (WARN_ON_ONCE(!pg) || scsi_device_get(sdev))
982 spin_lock_irqsave(&pg->lock, flags);
984 list_add_tail(&qdata->entry, &pg->rtpg_list);
985 pg->flags |= ALUA_PG_RUN_STPG;
988 if (pg->rtpg_sdev == NULL) {
990 pg->flags |= ALUA_PG_RUN_RTPG;
992 pg->rtpg_sdev = sdev;
994 } else if (!(pg->flags & ALUA_PG_RUN_RTPG) && force) {
995 pg->flags |= ALUA_PG_RUN_RTPG;
996 /* Do not queue if the worker is already running */
997 if (!(pg->flags & ALUA_PG_RUNNING)) {
1003 spin_unlock_irqrestore(&pg->lock, flags);
1006 if (queue_delayed_work(kaluad_wq, &pg->rtpg_work,
1007 msecs_to_jiffies(ALUA_RTPG_DELAY_MSECS)))
1010 kref_put(&pg->kref, release_port_group);
1013 scsi_device_put(sdev);
1019 * alua_initialize - Initialize ALUA state
1020 * @sdev: the device to be initialized
1022 * For the prep_fn to work correctly we have
1023 * to initialize the ALUA state for the device.
1025 static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h)
1027 int err = SCSI_DH_DEV_UNSUPP, tpgs;
1029 mutex_lock(&h->init_mutex);
1030 h->disabled = false;
1031 tpgs = alua_check_tpgs(sdev);
1032 if (tpgs != TPGS_MODE_NONE)
1033 err = alua_check_vpd(sdev, h, tpgs);
1034 h->init_error = err;
1035 mutex_unlock(&h->init_mutex);
1039 * alua_set_params - set/unset the optimize flag
1040 * @sdev: device on the path to be activated
1041 * params - parameters in the following format
1042 * "no_of_params\0param1\0param2\0param3\0...\0"
1043 * For example, to set the flag pass the following parameters
1044 * from multipath.conf
1045 * hardware_handler "2 alua 1"
1047 static int alua_set_params(struct scsi_device *sdev, const char *params)
1049 struct alua_dh_data *h = sdev->handler_data;
1050 struct alua_port_group *pg = NULL;
1051 unsigned int optimize = 0, argc;
1052 const char *p = params;
1053 int result = SCSI_DH_OK;
1054 unsigned long flags;
1056 if ((sscanf(params, "%u", &argc) != 1) || (argc != 1))
1061 if ((sscanf(p, "%u", &optimize) != 1) || (optimize > 1))
1065 pg = rcu_dereference(h->pg);
1070 spin_lock_irqsave(&pg->lock, flags);
1072 pg->flags |= ALUA_OPTIMIZE_STPG;
1074 pg->flags &= ~ALUA_OPTIMIZE_STPG;
1075 spin_unlock_irqrestore(&pg->lock, flags);
1082 * alua_activate - activate a path
1083 * @sdev: device on the path to be activated
1085 * We're currently switching the port group to be activated only and
1086 * let the array figure out the rest.
1087 * There may be other arrays which require us to switch all port groups
1088 * based on a certain policy. But until we actually encounter them it
1091 static int alua_activate(struct scsi_device *sdev,
1092 activate_complete fn, void *data)
1094 struct alua_dh_data *h = sdev->handler_data;
1095 int err = SCSI_DH_OK;
1096 struct alua_queue_data *qdata;
1097 struct alua_port_group *pg;
1099 qdata = kzalloc(sizeof(*qdata), GFP_KERNEL);
1101 err = SCSI_DH_RES_TEMP_UNAVAIL;
1104 qdata->callback_fn = fn;
1105 qdata->callback_data = data;
1107 mutex_lock(&h->init_mutex);
1109 pg = rcu_dereference(h->pg);
1110 if (!pg || !kref_get_unless_zero(&pg->kref)) {
1113 err = h->init_error;
1114 mutex_unlock(&h->init_mutex);
1118 mutex_unlock(&h->init_mutex);
1120 if (alua_rtpg_queue(pg, sdev, qdata, true))
1123 err = SCSI_DH_DEV_OFFLINED;
1124 kref_put(&pg->kref, release_port_group);
1132 * alua_check - check path status
1133 * @sdev: device on the path to be checked
1135 * Check the device status
1137 static void alua_check(struct scsi_device *sdev, bool force)
1139 struct alua_dh_data *h = sdev->handler_data;
1140 struct alua_port_group *pg;
1143 pg = rcu_dereference(h->pg);
1144 if (!pg || !kref_get_unless_zero(&pg->kref)) {
1149 alua_rtpg_queue(pg, sdev, NULL, force);
1150 kref_put(&pg->kref, release_port_group);
1154 * alua_prep_fn - request callback
1156 * Fail I/O to all paths not in state
1157 * active/optimized or active/non-optimized.
1159 static blk_status_t alua_prep_fn(struct scsi_device *sdev, struct request *req)
1161 struct alua_dh_data *h = sdev->handler_data;
1162 struct alua_port_group *pg;
1163 unsigned char state = SCSI_ACCESS_STATE_OPTIMAL;
1166 pg = rcu_dereference(h->pg);
1172 case SCSI_ACCESS_STATE_OPTIMAL:
1173 case SCSI_ACCESS_STATE_ACTIVE:
1174 case SCSI_ACCESS_STATE_LBA:
1175 case SCSI_ACCESS_STATE_TRANSITIONING:
1178 req->rq_flags |= RQF_QUIET;
1179 return BLK_STS_IOERR;
1183 static void alua_rescan(struct scsi_device *sdev)
1185 struct alua_dh_data *h = sdev->handler_data;
1187 alua_initialize(sdev, h);
1191 * alua_bus_attach - Attach device handler
1192 * @sdev: device to be attached to
1194 static int alua_bus_attach(struct scsi_device *sdev)
1196 struct alua_dh_data *h;
1199 h = kzalloc(sizeof(*h) , GFP_KERNEL);
1201 return SCSI_DH_NOMEM;
1202 spin_lock_init(&h->pg_lock);
1203 rcu_assign_pointer(h->pg, NULL);
1204 h->init_error = SCSI_DH_OK;
1206 INIT_LIST_HEAD(&h->node);
1208 mutex_init(&h->init_mutex);
1209 err = alua_initialize(sdev, h);
1210 if (err != SCSI_DH_OK && err != SCSI_DH_DEV_OFFLINED)
1213 sdev->handler_data = h;
1221 * alua_bus_detach - Detach device handler
1222 * @sdev: device to be detached from
1224 static void alua_bus_detach(struct scsi_device *sdev)
1226 struct alua_dh_data *h = sdev->handler_data;
1227 struct alua_port_group *pg;
1229 spin_lock(&h->pg_lock);
1230 pg = rcu_dereference_protected(h->pg, lockdep_is_held(&h->pg_lock));
1231 rcu_assign_pointer(h->pg, NULL);
1232 spin_unlock(&h->pg_lock);
1234 spin_lock_irq(&pg->lock);
1235 list_del_rcu(&h->node);
1236 spin_unlock_irq(&pg->lock);
1237 kref_put(&pg->kref, release_port_group);
1239 sdev->handler_data = NULL;
1244 static struct scsi_device_handler alua_dh = {
1245 .name = ALUA_DH_NAME,
1246 .module = THIS_MODULE,
1247 .attach = alua_bus_attach,
1248 .detach = alua_bus_detach,
1249 .prep_fn = alua_prep_fn,
1250 .check_sense = alua_check_sense,
1251 .activate = alua_activate,
1252 .rescan = alua_rescan,
1253 .set_params = alua_set_params,
1256 static int __init alua_init(void)
1260 kaluad_wq = alloc_workqueue("kaluad", WQ_MEM_RECLAIM, 0);
1264 r = scsi_register_device_handler(&alua_dh);
1266 printk(KERN_ERR "%s: Failed to register scsi device handler",
1268 destroy_workqueue(kaluad_wq);
1273 static void __exit alua_exit(void)
1275 scsi_unregister_device_handler(&alua_dh);
1276 destroy_workqueue(kaluad_wq);
1279 module_init(alua_init);
1280 module_exit(alua_exit);
1282 MODULE_DESCRIPTION("DM Multipath ALUA support");
1283 MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
1284 MODULE_LICENSE("GPL");
1285 MODULE_VERSION(ALUA_DH_VER);