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 blk_opf_t opf = REQ_OP_DRV_IN | REQ_FAILFAST_DEV |
131 REQ_FAILFAST_TRANSPORT | REQ_FAILFAST_DRIVER;
132 const struct scsi_exec_args exec_args = {
136 /* Prepare the command. */
137 memset(cdb, 0x0, MAX_COMMAND_SIZE);
138 cdb[0] = MAINTENANCE_IN;
139 if (!(flags & ALUA_RTPG_EXT_HDR_UNSUPP))
140 cdb[1] = MI_REPORT_TARGET_PGS | MI_EXT_HDR_PARAM_FMT;
142 cdb[1] = MI_REPORT_TARGET_PGS;
143 put_unaligned_be32(bufflen, &cdb[6]);
145 return scsi_execute_cmd(sdev, cdb, opf, buff, bufflen,
146 ALUA_FAILOVER_TIMEOUT * HZ,
147 ALUA_FAILOVER_RETRIES, &exec_args);
151 * submit_stpg - Issue a SET TARGET PORT GROUP command
153 * Currently we're only setting the current target port group state
154 * to 'active/optimized' and let the array firmware figure out
155 * the states of the remaining groups.
157 static int submit_stpg(struct scsi_device *sdev, int group_id,
158 struct scsi_sense_hdr *sshdr)
160 u8 cdb[MAX_COMMAND_SIZE];
161 unsigned char stpg_data[8];
163 blk_opf_t opf = REQ_OP_DRV_OUT | REQ_FAILFAST_DEV |
164 REQ_FAILFAST_TRANSPORT | REQ_FAILFAST_DRIVER;
165 const struct scsi_exec_args exec_args = {
169 /* Prepare the data buffer */
170 memset(stpg_data, 0, stpg_len);
171 stpg_data[4] = SCSI_ACCESS_STATE_OPTIMAL;
172 put_unaligned_be16(group_id, &stpg_data[6]);
174 /* Prepare the command. */
175 memset(cdb, 0x0, MAX_COMMAND_SIZE);
176 cdb[0] = MAINTENANCE_OUT;
177 cdb[1] = MO_SET_TARGET_PGS;
178 put_unaligned_be32(stpg_len, &cdb[6]);
180 return scsi_execute_cmd(sdev, cdb, opf, stpg_data,
181 stpg_len, ALUA_FAILOVER_TIMEOUT * HZ,
182 ALUA_FAILOVER_RETRIES, &exec_args);
185 static struct alua_port_group *alua_find_get_pg(char *id_str, size_t id_size,
188 struct alua_port_group *pg;
190 if (!id_str || !id_size || !strlen(id_str))
193 list_for_each_entry(pg, &port_group_list, node) {
194 if (pg->group_id != group_id)
196 if (!pg->device_id_len || pg->device_id_len != id_size)
198 if (strncmp(pg->device_id_str, id_str, id_size))
200 if (!kref_get_unless_zero(&pg->kref))
209 * alua_alloc_pg - Allocate a new port_group structure
211 * @group_id: port group id
212 * @tpgs: target port group settings
214 * Allocate a new port_group structure for a given
217 static struct alua_port_group *alua_alloc_pg(struct scsi_device *sdev,
218 int group_id, int tpgs)
220 struct alua_port_group *pg, *tmp_pg;
222 pg = kzalloc(sizeof(struct alua_port_group), GFP_KERNEL);
224 return ERR_PTR(-ENOMEM);
226 pg->device_id_len = scsi_vpd_lun_id(sdev, pg->device_id_str,
227 sizeof(pg->device_id_str));
228 if (pg->device_id_len <= 0) {
230 * TPGS supported but no device identification found.
231 * Generate private device identification.
233 sdev_printk(KERN_INFO, sdev,
234 "%s: No device descriptors found\n",
236 pg->device_id_str[0] = '\0';
237 pg->device_id_len = 0;
239 pg->group_id = group_id;
241 pg->state = SCSI_ACCESS_STATE_OPTIMAL;
242 pg->valid_states = TPGS_SUPPORT_ALL;
244 pg->flags |= ALUA_OPTIMIZE_STPG;
245 kref_init(&pg->kref);
246 INIT_DELAYED_WORK(&pg->rtpg_work, alua_rtpg_work);
247 INIT_LIST_HEAD(&pg->rtpg_list);
248 INIT_LIST_HEAD(&pg->node);
249 INIT_LIST_HEAD(&pg->dh_list);
250 spin_lock_init(&pg->lock);
252 spin_lock(&port_group_lock);
253 tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len,
256 spin_unlock(&port_group_lock);
261 list_add(&pg->node, &port_group_list);
262 spin_unlock(&port_group_lock);
268 * alua_check_tpgs - Evaluate TPGS setting
269 * @sdev: device to be checked
271 * Examine the TPGS setting of the sdev to find out if ALUA
274 static int alua_check_tpgs(struct scsi_device *sdev)
276 int tpgs = TPGS_MODE_NONE;
279 * ALUA support for non-disk devices is fraught with
280 * difficulties, so disable it for now.
282 if (sdev->type != TYPE_DISK) {
283 sdev_printk(KERN_INFO, sdev,
284 "%s: disable for non-disk devices\n",
289 tpgs = scsi_device_tpgs(sdev);
291 case TPGS_MODE_EXPLICIT|TPGS_MODE_IMPLICIT:
292 sdev_printk(KERN_INFO, sdev,
293 "%s: supports implicit and explicit TPGS\n",
296 case TPGS_MODE_EXPLICIT:
297 sdev_printk(KERN_INFO, sdev, "%s: supports explicit TPGS\n",
300 case TPGS_MODE_IMPLICIT:
301 sdev_printk(KERN_INFO, sdev, "%s: supports implicit TPGS\n",
305 sdev_printk(KERN_INFO, sdev, "%s: not supported\n",
309 sdev_printk(KERN_INFO, sdev,
310 "%s: unsupported TPGS setting %d\n",
312 tpgs = TPGS_MODE_NONE;
320 * alua_check_vpd - Evaluate INQUIRY vpd page 0x83
321 * @sdev: device to be checked
323 * Extract the relative target port and the target port group
324 * descriptor from the list of identificators.
326 static int alua_check_vpd(struct scsi_device *sdev, struct alua_dh_data *h,
329 int rel_port = -1, group_id;
330 struct alua_port_group *pg, *old_pg = NULL;
331 bool pg_updated = false;
334 group_id = scsi_vpd_tpg_id(sdev, &rel_port);
337 * Internal error; TPGS supported but required
338 * VPD identification descriptors not present.
339 * Disable ALUA support
341 sdev_printk(KERN_INFO, sdev,
342 "%s: No target port descriptors found\n",
344 return SCSI_DH_DEV_UNSUPP;
347 pg = alua_alloc_pg(sdev, group_id, tpgs);
349 if (PTR_ERR(pg) == -ENOMEM)
350 return SCSI_DH_NOMEM;
351 return SCSI_DH_DEV_UNSUPP;
353 if (pg->device_id_len)
354 sdev_printk(KERN_INFO, sdev,
355 "%s: device %s port group %x rel port %x\n",
356 ALUA_DH_NAME, pg->device_id_str,
359 sdev_printk(KERN_INFO, sdev,
360 "%s: port group %x rel port %x\n",
361 ALUA_DH_NAME, group_id, rel_port);
365 /* Check for existing port group references */
366 spin_lock(&h->pg_lock);
367 old_pg = rcu_dereference_protected(h->pg, lockdep_is_held(&h->pg_lock));
369 /* port group has changed. Update to new port group */
371 spin_lock_irqsave(&old_pg->lock, flags);
372 list_del_rcu(&h->node);
373 spin_unlock_irqrestore(&old_pg->lock, flags);
375 rcu_assign_pointer(h->pg, pg);
379 spin_lock_irqsave(&pg->lock, flags);
381 list_add_rcu(&h->node, &pg->dh_list);
382 spin_unlock_irqrestore(&pg->lock, flags);
384 spin_unlock(&h->pg_lock);
386 alua_rtpg_queue(pg, sdev, NULL, true);
387 kref_put(&pg->kref, release_port_group);
390 kref_put(&old_pg->kref, release_port_group);
395 static char print_alua_state(unsigned char state)
398 case SCSI_ACCESS_STATE_OPTIMAL:
400 case SCSI_ACCESS_STATE_ACTIVE:
402 case SCSI_ACCESS_STATE_STANDBY:
404 case SCSI_ACCESS_STATE_UNAVAILABLE:
406 case SCSI_ACCESS_STATE_LBA:
408 case SCSI_ACCESS_STATE_OFFLINE:
410 case SCSI_ACCESS_STATE_TRANSITIONING:
417 static enum scsi_disposition alua_check_sense(struct scsi_device *sdev,
418 struct scsi_sense_hdr *sense_hdr)
420 struct alua_dh_data *h = sdev->handler_data;
421 struct alua_port_group *pg;
423 switch (sense_hdr->sense_key) {
425 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a) {
427 * LUN Not Accessible - ALUA state transition
430 pg = rcu_dereference(h->pg);
432 pg->state = SCSI_ACCESS_STATE_TRANSITIONING;
434 alua_check(sdev, false);
439 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x00) {
441 * Power On, Reset, or Bus Device Reset.
442 * Might have obscured a state transition,
443 * so schedule a recheck.
445 alua_check(sdev, true);
446 return ADD_TO_MLQUEUE;
448 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x04)
450 * Device internal reset
452 return ADD_TO_MLQUEUE;
453 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x01)
455 * Mode Parameters Changed
457 return ADD_TO_MLQUEUE;
458 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x06) {
462 alua_check(sdev, true);
463 return ADD_TO_MLQUEUE;
465 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x07) {
467 * Implicit ALUA state transition failed
469 alua_check(sdev, true);
470 return ADD_TO_MLQUEUE;
472 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x03)
474 * Inquiry data has changed
476 return ADD_TO_MLQUEUE;
477 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x0e)
479 * REPORTED_LUNS_DATA_HAS_CHANGED is reported
480 * when switching controllers on targets like
481 * Intel Multi-Flex. We can just retry.
483 return ADD_TO_MLQUEUE;
487 return SCSI_RETURN_NOT_HANDLED;
491 * alua_tur - Send a TEST UNIT READY
492 * @sdev: device to which the TEST UNIT READY command should be send
494 * Send a TEST UNIT READY to @sdev to figure out the device state
495 * Returns SCSI_DH_RETRY if the sense code is NOT READY/ALUA TRANSITIONING,
496 * SCSI_DH_OK if no error occurred, and SCSI_DH_IO otherwise.
498 static int alua_tur(struct scsi_device *sdev)
500 struct scsi_sense_hdr sense_hdr;
503 retval = scsi_test_unit_ready(sdev, ALUA_FAILOVER_TIMEOUT * HZ,
504 ALUA_FAILOVER_RETRIES, &sense_hdr);
505 if (sense_hdr.sense_key == NOT_READY &&
506 sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a)
507 return SCSI_DH_RETRY;
515 * alua_rtpg - Evaluate REPORT TARGET GROUP STATES
516 * @sdev: the device to be evaluated.
518 * Evaluate the Target Port Group State.
519 * Returns SCSI_DH_DEV_OFFLINED if the path is
520 * found to be unusable.
522 static int alua_rtpg(struct scsi_device *sdev, struct alua_port_group *pg)
524 struct scsi_sense_hdr sense_hdr;
525 struct alua_port_group *tmp_pg;
526 int len, k, off, bufflen = ALUA_RTPG_SIZE;
527 int group_id_old, state_old, pref_old, valid_states_old;
528 unsigned char *desc, *buff;
531 unsigned int tpg_desc_tbl_off;
532 unsigned char orig_transition_tmo;
534 bool transitioning_sense = false;
536 group_id_old = pg->group_id;
537 state_old = pg->state;
539 valid_states_old = pg->valid_states;
542 unsigned long transition_tmo = ALUA_FAILOVER_TIMEOUT * HZ;
544 if (pg->transition_tmo)
545 transition_tmo = pg->transition_tmo * HZ;
547 pg->expiry = round_jiffies_up(jiffies + transition_tmo);
550 buff = kzalloc(bufflen, GFP_KERNEL);
552 return SCSI_DH_DEV_TEMP_BUSY;
556 retval = submit_rtpg(sdev, buff, bufflen, &sense_hdr, pg->flags);
560 * Some (broken) implementations have a habit of returning
561 * an error during things like firmware update etc.
562 * But if the target only supports active/optimized there's
563 * not much we can do; it's not that we can switch paths
565 * So ignore any errors to avoid spurious failures during
568 if ((pg->valid_states & ~TPGS_SUPPORT_OPTIMIZED) == 0) {
569 sdev_printk(KERN_INFO, sdev,
570 "%s: ignoring rtpg result %d\n",
571 ALUA_DH_NAME, retval);
575 if (retval < 0 || !scsi_sense_valid(&sense_hdr)) {
576 sdev_printk(KERN_INFO, sdev,
577 "%s: rtpg failed, result %d\n",
578 ALUA_DH_NAME, retval);
581 return SCSI_DH_DEV_TEMP_BUSY;
582 if (host_byte(retval) == DID_NO_CONNECT)
583 return SCSI_DH_RES_TEMP_UNAVAIL;
588 * submit_rtpg() has failed on existing arrays
589 * when requesting extended header info, and
590 * the array doesn't support extended headers,
591 * even though it shouldn't according to T10.
592 * The retry without rtpg_ext_hdr_req set
594 * Note: some arrays return a sense key of ILLEGAL_REQUEST
595 * with ASC 00h if they don't support the extended header.
597 if (!(pg->flags & ALUA_RTPG_EXT_HDR_UNSUPP) &&
598 sense_hdr.sense_key == ILLEGAL_REQUEST) {
599 pg->flags |= ALUA_RTPG_EXT_HDR_UNSUPP;
603 * If the array returns with 'ALUA state transition'
604 * sense code here it cannot return RTPG data during
605 * transition. So set the state to 'transitioning' directly.
607 if (sense_hdr.sense_key == NOT_READY &&
608 sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a) {
609 transitioning_sense = true;
613 * Retry on any other UNIT ATTENTION occurred.
615 if (sense_hdr.sense_key == UNIT_ATTENTION)
617 if (err == SCSI_DH_RETRY &&
618 pg->expiry != 0 && time_before(jiffies, pg->expiry)) {
619 sdev_printk(KERN_ERR, sdev, "%s: rtpg retry\n",
621 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
625 sdev_printk(KERN_ERR, sdev, "%s: rtpg failed\n",
627 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
633 len = get_unaligned_be32(&buff[0]) + 4;
636 /* Resubmit with the correct length */
639 buff = kmalloc(bufflen, GFP_KERNEL);
641 sdev_printk(KERN_WARNING, sdev,
642 "%s: kmalloc buffer failed\n",__func__);
643 /* Temporary failure, bypass */
645 return SCSI_DH_DEV_TEMP_BUSY;
650 orig_transition_tmo = pg->transition_tmo;
651 if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR && buff[5] != 0)
652 pg->transition_tmo = buff[5];
654 pg->transition_tmo = ALUA_FAILOVER_TIMEOUT;
656 if (orig_transition_tmo != pg->transition_tmo) {
657 sdev_printk(KERN_INFO, sdev,
658 "%s: transition timeout set to %d seconds\n",
659 ALUA_DH_NAME, pg->transition_tmo);
660 pg->expiry = jiffies + pg->transition_tmo * HZ;
663 if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR)
664 tpg_desc_tbl_off = 8;
666 tpg_desc_tbl_off = 4;
668 for (k = tpg_desc_tbl_off, desc = buff + tpg_desc_tbl_off;
670 k += off, desc += off) {
671 u16 group_id = get_unaligned_be16(&desc[2]);
673 spin_lock_irqsave(&port_group_lock, flags);
674 tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len,
676 spin_unlock_irqrestore(&port_group_lock, flags);
678 if (spin_trylock_irqsave(&tmp_pg->lock, flags)) {
679 if ((tmp_pg == pg) ||
680 !(tmp_pg->flags & ALUA_PG_RUNNING)) {
681 struct alua_dh_data *h;
683 tmp_pg->state = desc[0] & 0x0f;
684 tmp_pg->pref = desc[0] >> 7;
686 list_for_each_entry_rcu(h,
687 &tmp_pg->dh_list, node) {
690 h->sdev->access_state = desc[0];
695 tmp_pg->valid_states = desc[1];
696 spin_unlock_irqrestore(&tmp_pg->lock, flags);
698 kref_put(&tmp_pg->kref, release_port_group);
700 off = 8 + (desc[7] * 4);
704 spin_lock_irqsave(&pg->lock, flags);
705 if (transitioning_sense)
706 pg->state = SCSI_ACCESS_STATE_TRANSITIONING;
708 if (group_id_old != pg->group_id || state_old != pg->state ||
709 pref_old != pg->pref || valid_states_old != pg->valid_states)
710 sdev_printk(KERN_INFO, sdev,
711 "%s: port group %02x state %c %s supports %c%c%c%c%c%c%c\n",
712 ALUA_DH_NAME, pg->group_id, print_alua_state(pg->state),
713 pg->pref ? "preferred" : "non-preferred",
714 pg->valid_states&TPGS_SUPPORT_TRANSITION?'T':'t',
715 pg->valid_states&TPGS_SUPPORT_OFFLINE?'O':'o',
716 pg->valid_states&TPGS_SUPPORT_LBA_DEPENDENT?'L':'l',
717 pg->valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u',
718 pg->valid_states&TPGS_SUPPORT_STANDBY?'S':'s',
719 pg->valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n',
720 pg->valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a');
723 case SCSI_ACCESS_STATE_TRANSITIONING:
724 if (time_before(jiffies, pg->expiry)) {
725 /* State transition, retry */
726 pg->interval = ALUA_RTPG_RETRY_DELAY;
729 struct alua_dh_data *h;
731 /* Transitioning time exceeded, set port to standby */
733 pg->state = SCSI_ACCESS_STATE_STANDBY;
736 list_for_each_entry_rcu(h, &pg->dh_list, node) {
739 h->sdev->access_state =
740 (pg->state & SCSI_ACCESS_STATE_MASK);
742 h->sdev->access_state |=
743 SCSI_ACCESS_STATE_PREFERRED;
748 case SCSI_ACCESS_STATE_OFFLINE:
750 err = SCSI_DH_DEV_OFFLINED;
754 /* Useable path if active */
759 spin_unlock_irqrestore(&pg->lock, flags);
765 * alua_stpg - Issue a SET TARGET PORT GROUP command
767 * Issue a SET TARGET PORT GROUP command and evaluate the
768 * response. Returns SCSI_DH_RETRY per default to trigger
769 * a re-evaluation of the target group state or SCSI_DH_OK
770 * if no further action needs to be taken.
772 static unsigned alua_stpg(struct scsi_device *sdev, struct alua_port_group *pg)
775 struct scsi_sense_hdr sense_hdr;
777 if (!(pg->tpgs & TPGS_MODE_EXPLICIT)) {
778 /* Only implicit ALUA supported, retry */
779 return SCSI_DH_RETRY;
782 case SCSI_ACCESS_STATE_OPTIMAL:
784 case SCSI_ACCESS_STATE_ACTIVE:
785 if ((pg->flags & ALUA_OPTIMIZE_STPG) &&
787 (pg->tpgs & TPGS_MODE_IMPLICIT))
790 case SCSI_ACCESS_STATE_STANDBY:
791 case SCSI_ACCESS_STATE_UNAVAILABLE:
793 case SCSI_ACCESS_STATE_OFFLINE:
795 case SCSI_ACCESS_STATE_TRANSITIONING:
798 sdev_printk(KERN_INFO, sdev,
799 "%s: stpg failed, unhandled TPGS state %d",
800 ALUA_DH_NAME, pg->state);
801 return SCSI_DH_NOSYS;
803 retval = submit_stpg(sdev, pg->group_id, &sense_hdr);
806 if (retval < 0 || !scsi_sense_valid(&sense_hdr)) {
807 sdev_printk(KERN_INFO, sdev,
808 "%s: stpg failed, result %d",
809 ALUA_DH_NAME, retval);
811 return SCSI_DH_DEV_TEMP_BUSY;
813 sdev_printk(KERN_INFO, sdev, "%s: stpg failed\n",
815 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
819 return SCSI_DH_RETRY;
823 * The caller must call scsi_device_put() on the returned pointer if it is not
826 static struct scsi_device * __must_check
827 alua_rtpg_select_sdev(struct alua_port_group *pg)
829 struct alua_dh_data *h;
830 struct scsi_device *sdev = NULL, *prev_sdev;
832 lockdep_assert_held(&pg->lock);
833 if (WARN_ON(!pg->rtpg_sdev))
837 * RCU protection isn't necessary for dh_list here
838 * as we hold pg->lock, but for access to h->pg.
841 list_for_each_entry_rcu(h, &pg->dh_list, node) {
844 if (h->sdev == pg->rtpg_sdev) {
848 if (rcu_dereference(h->pg) == pg &&
850 !scsi_device_get(h->sdev)) {
858 pr_warn("%s: no device found for rtpg\n",
860 (char *)pg->device_id_str : "(nameless PG)"));
864 sdev_printk(KERN_INFO, sdev, "rtpg retry on different device\n");
866 prev_sdev = pg->rtpg_sdev;
867 pg->rtpg_sdev = sdev;
872 static void alua_rtpg_work(struct work_struct *work)
874 struct alua_port_group *pg =
875 container_of(work, struct alua_port_group, rtpg_work.work);
876 struct scsi_device *sdev, *prev_sdev = NULL;
877 LIST_HEAD(qdata_list);
878 int err = SCSI_DH_OK;
879 struct alua_queue_data *qdata, *tmp;
880 struct alua_dh_data *h;
883 spin_lock_irqsave(&pg->lock, flags);
884 sdev = pg->rtpg_sdev;
886 WARN_ON(pg->flags & ALUA_PG_RUN_RTPG);
887 WARN_ON(pg->flags & ALUA_PG_RUN_STPG);
888 spin_unlock_irqrestore(&pg->lock, flags);
889 kref_put(&pg->kref, release_port_group);
892 pg->flags |= ALUA_PG_RUNNING;
893 if (pg->flags & ALUA_PG_RUN_RTPG) {
894 int state = pg->state;
896 pg->flags &= ~ALUA_PG_RUN_RTPG;
897 spin_unlock_irqrestore(&pg->lock, flags);
898 if (state == SCSI_ACCESS_STATE_TRANSITIONING) {
899 if (alua_tur(sdev) == SCSI_DH_RETRY) {
900 spin_lock_irqsave(&pg->lock, flags);
901 pg->flags &= ~ALUA_PG_RUNNING;
902 pg->flags |= ALUA_PG_RUN_RTPG;
904 pg->interval = ALUA_RTPG_RETRY_DELAY;
905 spin_unlock_irqrestore(&pg->lock, flags);
906 queue_delayed_work(kaluad_wq, &pg->rtpg_work,
910 /* Send RTPG on failure or if TUR indicates SUCCESS */
912 err = alua_rtpg(sdev, pg);
913 spin_lock_irqsave(&pg->lock, flags);
915 /* If RTPG failed on the current device, try using another */
916 if (err == SCSI_DH_RES_TEMP_UNAVAIL &&
917 (prev_sdev = alua_rtpg_select_sdev(pg)))
918 err = SCSI_DH_IMM_RETRY;
920 if (err == SCSI_DH_RETRY || err == SCSI_DH_IMM_RETRY ||
921 pg->flags & ALUA_PG_RUN_RTPG) {
922 pg->flags &= ~ALUA_PG_RUNNING;
923 if (err == SCSI_DH_IMM_RETRY)
925 else if (!pg->interval && !(pg->flags & ALUA_PG_RUN_RTPG))
926 pg->interval = ALUA_RTPG_RETRY_DELAY;
927 pg->flags |= ALUA_PG_RUN_RTPG;
928 spin_unlock_irqrestore(&pg->lock, flags);
931 if (err != SCSI_DH_OK)
932 pg->flags &= ~ALUA_PG_RUN_STPG;
934 if (pg->flags & ALUA_PG_RUN_STPG) {
935 pg->flags &= ~ALUA_PG_RUN_STPG;
936 spin_unlock_irqrestore(&pg->lock, flags);
937 err = alua_stpg(sdev, pg);
938 spin_lock_irqsave(&pg->lock, flags);
939 if (err == SCSI_DH_RETRY || pg->flags & ALUA_PG_RUN_RTPG) {
940 pg->flags |= ALUA_PG_RUN_RTPG;
942 pg->flags &= ~ALUA_PG_RUNNING;
943 spin_unlock_irqrestore(&pg->lock, flags);
948 list_splice_init(&pg->rtpg_list, &qdata_list);
950 * We went through an RTPG, for good or bad.
951 * Re-enable all devices for the next attempt.
953 list_for_each_entry(h, &pg->dh_list, node)
955 pg->rtpg_sdev = NULL;
956 spin_unlock_irqrestore(&pg->lock, flags);
959 scsi_device_put(prev_sdev);
961 list_for_each_entry_safe(qdata, tmp, &qdata_list, entry) {
962 list_del(&qdata->entry);
963 if (qdata->callback_fn)
964 qdata->callback_fn(qdata->callback_data, err);
967 spin_lock_irqsave(&pg->lock, flags);
968 pg->flags &= ~ALUA_PG_RUNNING;
969 spin_unlock_irqrestore(&pg->lock, flags);
970 scsi_device_put(sdev);
971 kref_put(&pg->kref, release_port_group);
976 scsi_device_put(prev_sdev);
977 queue_delayed_work(kaluad_wq, &pg->rtpg_work, pg->interval * HZ);
981 * alua_rtpg_queue() - cause RTPG to be submitted asynchronously
982 * @pg: ALUA port group associated with @sdev.
983 * @sdev: SCSI device for which to submit an RTPG.
984 * @qdata: Information about the callback to invoke after the RTPG.
985 * @force: Whether or not to submit an RTPG if a work item that will submit an
986 * RTPG already has been scheduled.
988 * Returns true if and only if alua_rtpg_work() will be called asynchronously.
989 * That function is responsible for calling @qdata->fn().
991 * Context: may be called from atomic context (alua_check()) only if the caller
992 * holds an sdev reference.
994 static bool alua_rtpg_queue(struct alua_port_group *pg,
995 struct scsi_device *sdev,
996 struct alua_queue_data *qdata, bool force)
1001 if (WARN_ON_ONCE(!pg) || scsi_device_get(sdev))
1004 spin_lock_irqsave(&pg->lock, flags);
1006 list_add_tail(&qdata->entry, &pg->rtpg_list);
1007 pg->flags |= ALUA_PG_RUN_STPG;
1010 if (pg->rtpg_sdev == NULL) {
1011 struct alua_dh_data *h = sdev->handler_data;
1014 if (h && rcu_dereference(h->pg) == pg) {
1016 pg->flags |= ALUA_PG_RUN_RTPG;
1017 kref_get(&pg->kref);
1018 pg->rtpg_sdev = sdev;
1022 } else if (!(pg->flags & ALUA_PG_RUN_RTPG) && force) {
1023 pg->flags |= ALUA_PG_RUN_RTPG;
1024 /* Do not queue if the worker is already running */
1025 if (!(pg->flags & ALUA_PG_RUNNING)) {
1026 kref_get(&pg->kref);
1031 spin_unlock_irqrestore(&pg->lock, flags);
1034 if (queue_delayed_work(kaluad_wq, &pg->rtpg_work,
1035 msecs_to_jiffies(ALUA_RTPG_DELAY_MSECS)))
1038 kref_put(&pg->kref, release_port_group);
1041 scsi_device_put(sdev);
1047 * alua_initialize - Initialize ALUA state
1048 * @sdev: the device to be initialized
1050 * For the prep_fn to work correctly we have
1051 * to initialize the ALUA state for the device.
1053 static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h)
1055 int err = SCSI_DH_DEV_UNSUPP, tpgs;
1057 mutex_lock(&h->init_mutex);
1058 h->disabled = false;
1059 tpgs = alua_check_tpgs(sdev);
1060 if (tpgs != TPGS_MODE_NONE)
1061 err = alua_check_vpd(sdev, h, tpgs);
1062 h->init_error = err;
1063 mutex_unlock(&h->init_mutex);
1067 * alua_set_params - set/unset the optimize flag
1068 * @sdev: device on the path to be activated
1069 * params - parameters in the following format
1070 * "no_of_params\0param1\0param2\0param3\0...\0"
1071 * For example, to set the flag pass the following parameters
1072 * from multipath.conf
1073 * hardware_handler "2 alua 1"
1075 static int alua_set_params(struct scsi_device *sdev, const char *params)
1077 struct alua_dh_data *h = sdev->handler_data;
1078 struct alua_port_group *pg = NULL;
1079 unsigned int optimize = 0, argc;
1080 const char *p = params;
1081 int result = SCSI_DH_OK;
1082 unsigned long flags;
1084 if ((sscanf(params, "%u", &argc) != 1) || (argc != 1))
1089 if ((sscanf(p, "%u", &optimize) != 1) || (optimize > 1))
1093 pg = rcu_dereference(h->pg);
1098 spin_lock_irqsave(&pg->lock, flags);
1100 pg->flags |= ALUA_OPTIMIZE_STPG;
1102 pg->flags &= ~ALUA_OPTIMIZE_STPG;
1103 spin_unlock_irqrestore(&pg->lock, flags);
1110 * alua_activate - activate a path
1111 * @sdev: device on the path to be activated
1113 * We're currently switching the port group to be activated only and
1114 * let the array figure out the rest.
1115 * There may be other arrays which require us to switch all port groups
1116 * based on a certain policy. But until we actually encounter them it
1119 static int alua_activate(struct scsi_device *sdev,
1120 activate_complete fn, void *data)
1122 struct alua_dh_data *h = sdev->handler_data;
1123 int err = SCSI_DH_OK;
1124 struct alua_queue_data *qdata;
1125 struct alua_port_group *pg;
1127 qdata = kzalloc(sizeof(*qdata), GFP_KERNEL);
1129 err = SCSI_DH_RES_TEMP_UNAVAIL;
1132 qdata->callback_fn = fn;
1133 qdata->callback_data = data;
1135 mutex_lock(&h->init_mutex);
1137 pg = rcu_dereference(h->pg);
1138 if (!pg || !kref_get_unless_zero(&pg->kref)) {
1141 err = h->init_error;
1142 mutex_unlock(&h->init_mutex);
1146 mutex_unlock(&h->init_mutex);
1148 if (alua_rtpg_queue(pg, sdev, qdata, true)) {
1152 err = SCSI_DH_DEV_OFFLINED;
1154 kref_put(&pg->kref, release_port_group);
1162 * alua_check - check path status
1163 * @sdev: device on the path to be checked
1165 * Check the device status
1167 static void alua_check(struct scsi_device *sdev, bool force)
1169 struct alua_dh_data *h = sdev->handler_data;
1170 struct alua_port_group *pg;
1173 pg = rcu_dereference(h->pg);
1174 if (!pg || !kref_get_unless_zero(&pg->kref)) {
1179 alua_rtpg_queue(pg, sdev, NULL, force);
1180 kref_put(&pg->kref, release_port_group);
1184 * alua_prep_fn - request callback
1186 * Fail I/O to all paths not in state
1187 * active/optimized or active/non-optimized.
1189 static blk_status_t alua_prep_fn(struct scsi_device *sdev, struct request *req)
1191 struct alua_dh_data *h = sdev->handler_data;
1192 struct alua_port_group *pg;
1193 unsigned char state = SCSI_ACCESS_STATE_OPTIMAL;
1196 pg = rcu_dereference(h->pg);
1202 case SCSI_ACCESS_STATE_OPTIMAL:
1203 case SCSI_ACCESS_STATE_ACTIVE:
1204 case SCSI_ACCESS_STATE_LBA:
1205 case SCSI_ACCESS_STATE_TRANSITIONING:
1208 req->rq_flags |= RQF_QUIET;
1209 return BLK_STS_IOERR;
1213 static void alua_rescan(struct scsi_device *sdev)
1215 struct alua_dh_data *h = sdev->handler_data;
1217 alua_initialize(sdev, h);
1221 * alua_bus_attach - Attach device handler
1222 * @sdev: device to be attached to
1224 static int alua_bus_attach(struct scsi_device *sdev)
1226 struct alua_dh_data *h;
1229 h = kzalloc(sizeof(*h) , GFP_KERNEL);
1231 return SCSI_DH_NOMEM;
1232 spin_lock_init(&h->pg_lock);
1233 rcu_assign_pointer(h->pg, NULL);
1234 h->init_error = SCSI_DH_OK;
1236 INIT_LIST_HEAD(&h->node);
1238 mutex_init(&h->init_mutex);
1239 err = alua_initialize(sdev, h);
1240 if (err != SCSI_DH_OK && err != SCSI_DH_DEV_OFFLINED)
1243 sdev->handler_data = h;
1251 * alua_bus_detach - Detach device handler
1252 * @sdev: device to be detached from
1254 static void alua_bus_detach(struct scsi_device *sdev)
1256 struct alua_dh_data *h = sdev->handler_data;
1257 struct alua_port_group *pg;
1259 spin_lock(&h->pg_lock);
1260 pg = rcu_dereference_protected(h->pg, lockdep_is_held(&h->pg_lock));
1261 rcu_assign_pointer(h->pg, NULL);
1262 spin_unlock(&h->pg_lock);
1264 spin_lock_irq(&pg->lock);
1265 list_del_rcu(&h->node);
1266 spin_unlock_irq(&pg->lock);
1267 kref_put(&pg->kref, release_port_group);
1269 sdev->handler_data = NULL;
1274 static struct scsi_device_handler alua_dh = {
1275 .name = ALUA_DH_NAME,
1276 .module = THIS_MODULE,
1277 .attach = alua_bus_attach,
1278 .detach = alua_bus_detach,
1279 .prep_fn = alua_prep_fn,
1280 .check_sense = alua_check_sense,
1281 .activate = alua_activate,
1282 .rescan = alua_rescan,
1283 .set_params = alua_set_params,
1286 static int __init alua_init(void)
1290 kaluad_wq = alloc_workqueue("kaluad", WQ_MEM_RECLAIM, 0);
1294 r = scsi_register_device_handler(&alua_dh);
1296 printk(KERN_ERR "%s: Failed to register scsi device handler",
1298 destroy_workqueue(kaluad_wq);
1303 static void __exit alua_exit(void)
1305 scsi_unregister_device_handler(&alua_dh);
1306 destroy_workqueue(kaluad_wq);
1309 module_init(alua_init);
1310 module_exit(alua_exit);
1312 MODULE_DESCRIPTION("DM Multipath ALUA support");
1313 MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
1314 MODULE_LICENSE("GPL");
1315 MODULE_VERSION(ALUA_DH_VER);