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 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 blk_opf_t 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);
359 /* Check for existing port group references */
360 spin_lock(&h->pg_lock);
361 old_pg = rcu_dereference_protected(h->pg, lockdep_is_held(&h->pg_lock));
363 /* port group has changed. Update to new port group */
365 spin_lock_irqsave(&old_pg->lock, flags);
366 list_del_rcu(&h->node);
367 spin_unlock_irqrestore(&old_pg->lock, flags);
369 rcu_assign_pointer(h->pg, pg);
373 spin_lock_irqsave(&pg->lock, flags);
375 list_add_rcu(&h->node, &pg->dh_list);
376 spin_unlock_irqrestore(&pg->lock, flags);
378 spin_unlock(&h->pg_lock);
380 alua_rtpg_queue(pg, sdev, NULL, true);
381 kref_put(&pg->kref, release_port_group);
384 kref_put(&old_pg->kref, release_port_group);
389 static char print_alua_state(unsigned char state)
392 case SCSI_ACCESS_STATE_OPTIMAL:
394 case SCSI_ACCESS_STATE_ACTIVE:
396 case SCSI_ACCESS_STATE_STANDBY:
398 case SCSI_ACCESS_STATE_UNAVAILABLE:
400 case SCSI_ACCESS_STATE_LBA:
402 case SCSI_ACCESS_STATE_OFFLINE:
404 case SCSI_ACCESS_STATE_TRANSITIONING:
411 static enum scsi_disposition alua_check_sense(struct scsi_device *sdev,
412 struct scsi_sense_hdr *sense_hdr)
414 struct alua_dh_data *h = sdev->handler_data;
415 struct alua_port_group *pg;
417 switch (sense_hdr->sense_key) {
419 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a) {
421 * LUN Not Accessible - ALUA state transition
424 pg = rcu_dereference(h->pg);
426 pg->state = SCSI_ACCESS_STATE_TRANSITIONING;
428 alua_check(sdev, false);
433 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x00) {
435 * Power On, Reset, or Bus Device Reset.
436 * Might have obscured a state transition,
437 * so schedule a recheck.
439 alua_check(sdev, true);
440 return ADD_TO_MLQUEUE;
442 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x04)
444 * Device internal reset
446 return ADD_TO_MLQUEUE;
447 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x01)
449 * Mode Parameters Changed
451 return ADD_TO_MLQUEUE;
452 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x06) {
456 alua_check(sdev, true);
457 return ADD_TO_MLQUEUE;
459 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x07) {
461 * Implicit ALUA state transition failed
463 alua_check(sdev, true);
464 return ADD_TO_MLQUEUE;
466 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x03)
468 * Inquiry data has changed
470 return ADD_TO_MLQUEUE;
471 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x0e)
473 * REPORTED_LUNS_DATA_HAS_CHANGED is reported
474 * when switching controllers on targets like
475 * Intel Multi-Flex. We can just retry.
477 return ADD_TO_MLQUEUE;
481 return SCSI_RETURN_NOT_HANDLED;
485 * alua_tur - Send a TEST UNIT READY
486 * @sdev: device to which the TEST UNIT READY command should be send
488 * Send a TEST UNIT READY to @sdev to figure out the device state
489 * Returns SCSI_DH_RETRY if the sense code is NOT READY/ALUA TRANSITIONING,
490 * SCSI_DH_OK if no error occurred, and SCSI_DH_IO otherwise.
492 static int alua_tur(struct scsi_device *sdev)
494 struct scsi_sense_hdr sense_hdr;
497 retval = scsi_test_unit_ready(sdev, ALUA_FAILOVER_TIMEOUT * HZ,
498 ALUA_FAILOVER_RETRIES, &sense_hdr);
499 if (sense_hdr.sense_key == NOT_READY &&
500 sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a)
501 return SCSI_DH_RETRY;
509 * alua_rtpg - Evaluate REPORT TARGET GROUP STATES
510 * @sdev: the device to be evaluated.
512 * Evaluate the Target Port Group State.
513 * Returns SCSI_DH_DEV_OFFLINED if the path is
514 * found to be unusable.
516 static int alua_rtpg(struct scsi_device *sdev, struct alua_port_group *pg)
518 struct scsi_sense_hdr sense_hdr;
519 struct alua_port_group *tmp_pg;
520 int len, k, off, bufflen = ALUA_RTPG_SIZE;
521 int group_id_old, state_old, pref_old, valid_states_old;
522 unsigned char *desc, *buff;
525 unsigned int tpg_desc_tbl_off;
526 unsigned char orig_transition_tmo;
528 bool transitioning_sense = false;
530 group_id_old = pg->group_id;
531 state_old = pg->state;
533 valid_states_old = pg->valid_states;
536 unsigned long transition_tmo = ALUA_FAILOVER_TIMEOUT * HZ;
538 if (pg->transition_tmo)
539 transition_tmo = pg->transition_tmo * HZ;
541 pg->expiry = round_jiffies_up(jiffies + transition_tmo);
544 buff = kzalloc(bufflen, GFP_KERNEL);
546 return SCSI_DH_DEV_TEMP_BUSY;
550 retval = submit_rtpg(sdev, buff, bufflen, &sense_hdr, pg->flags);
554 * Some (broken) implementations have a habit of returning
555 * an error during things like firmware update etc.
556 * But if the target only supports active/optimized there's
557 * not much we can do; it's not that we can switch paths
559 * So ignore any errors to avoid spurious failures during
562 if ((pg->valid_states & ~TPGS_SUPPORT_OPTIMIZED) == 0) {
563 sdev_printk(KERN_INFO, sdev,
564 "%s: ignoring rtpg result %d\n",
565 ALUA_DH_NAME, retval);
569 if (retval < 0 || !scsi_sense_valid(&sense_hdr)) {
570 sdev_printk(KERN_INFO, sdev,
571 "%s: rtpg failed, result %d\n",
572 ALUA_DH_NAME, retval);
575 return SCSI_DH_DEV_TEMP_BUSY;
576 if (host_byte(retval) == DID_NO_CONNECT)
577 return SCSI_DH_RES_TEMP_UNAVAIL;
582 * submit_rtpg() has failed on existing arrays
583 * when requesting extended header info, and
584 * the array doesn't support extended headers,
585 * even though it shouldn't according to T10.
586 * The retry without rtpg_ext_hdr_req set
588 * Note: some arrays return a sense key of ILLEGAL_REQUEST
589 * with ASC 00h if they don't support the extended header.
591 if (!(pg->flags & ALUA_RTPG_EXT_HDR_UNSUPP) &&
592 sense_hdr.sense_key == ILLEGAL_REQUEST) {
593 pg->flags |= ALUA_RTPG_EXT_HDR_UNSUPP;
597 * If the array returns with 'ALUA state transition'
598 * sense code here it cannot return RTPG data during
599 * transition. So set the state to 'transitioning' directly.
601 if (sense_hdr.sense_key == NOT_READY &&
602 sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a) {
603 transitioning_sense = true;
607 * Retry on any other UNIT ATTENTION occurred.
609 if (sense_hdr.sense_key == UNIT_ATTENTION)
611 if (err == SCSI_DH_RETRY &&
612 pg->expiry != 0 && time_before(jiffies, pg->expiry)) {
613 sdev_printk(KERN_ERR, sdev, "%s: rtpg retry\n",
615 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
619 sdev_printk(KERN_ERR, sdev, "%s: rtpg failed\n",
621 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
627 len = get_unaligned_be32(&buff[0]) + 4;
630 /* Resubmit with the correct length */
633 buff = kmalloc(bufflen, GFP_KERNEL);
635 sdev_printk(KERN_WARNING, sdev,
636 "%s: kmalloc buffer failed\n",__func__);
637 /* Temporary failure, bypass */
639 return SCSI_DH_DEV_TEMP_BUSY;
644 orig_transition_tmo = pg->transition_tmo;
645 if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR && buff[5] != 0)
646 pg->transition_tmo = buff[5];
648 pg->transition_tmo = ALUA_FAILOVER_TIMEOUT;
650 if (orig_transition_tmo != pg->transition_tmo) {
651 sdev_printk(KERN_INFO, sdev,
652 "%s: transition timeout set to %d seconds\n",
653 ALUA_DH_NAME, pg->transition_tmo);
654 pg->expiry = jiffies + pg->transition_tmo * HZ;
657 if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR)
658 tpg_desc_tbl_off = 8;
660 tpg_desc_tbl_off = 4;
662 for (k = tpg_desc_tbl_off, desc = buff + tpg_desc_tbl_off;
664 k += off, desc += off) {
665 u16 group_id = get_unaligned_be16(&desc[2]);
667 spin_lock_irqsave(&port_group_lock, flags);
668 tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len,
670 spin_unlock_irqrestore(&port_group_lock, flags);
672 if (spin_trylock_irqsave(&tmp_pg->lock, flags)) {
673 if ((tmp_pg == pg) ||
674 !(tmp_pg->flags & ALUA_PG_RUNNING)) {
675 struct alua_dh_data *h;
677 tmp_pg->state = desc[0] & 0x0f;
678 tmp_pg->pref = desc[0] >> 7;
680 list_for_each_entry_rcu(h,
681 &tmp_pg->dh_list, node) {
684 h->sdev->access_state = desc[0];
689 tmp_pg->valid_states = desc[1];
690 spin_unlock_irqrestore(&tmp_pg->lock, flags);
692 kref_put(&tmp_pg->kref, release_port_group);
694 off = 8 + (desc[7] * 4);
698 spin_lock_irqsave(&pg->lock, flags);
699 if (transitioning_sense)
700 pg->state = SCSI_ACCESS_STATE_TRANSITIONING;
702 if (group_id_old != pg->group_id || state_old != pg->state ||
703 pref_old != pg->pref || valid_states_old != pg->valid_states)
704 sdev_printk(KERN_INFO, sdev,
705 "%s: port group %02x state %c %s supports %c%c%c%c%c%c%c\n",
706 ALUA_DH_NAME, pg->group_id, print_alua_state(pg->state),
707 pg->pref ? "preferred" : "non-preferred",
708 pg->valid_states&TPGS_SUPPORT_TRANSITION?'T':'t',
709 pg->valid_states&TPGS_SUPPORT_OFFLINE?'O':'o',
710 pg->valid_states&TPGS_SUPPORT_LBA_DEPENDENT?'L':'l',
711 pg->valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u',
712 pg->valid_states&TPGS_SUPPORT_STANDBY?'S':'s',
713 pg->valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n',
714 pg->valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a');
717 case SCSI_ACCESS_STATE_TRANSITIONING:
718 if (time_before(jiffies, pg->expiry)) {
719 /* State transition, retry */
720 pg->interval = ALUA_RTPG_RETRY_DELAY;
723 struct alua_dh_data *h;
725 /* Transitioning time exceeded, set port to standby */
727 pg->state = SCSI_ACCESS_STATE_STANDBY;
730 list_for_each_entry_rcu(h, &pg->dh_list, node) {
733 h->sdev->access_state =
734 (pg->state & SCSI_ACCESS_STATE_MASK);
736 h->sdev->access_state |=
737 SCSI_ACCESS_STATE_PREFERRED;
742 case SCSI_ACCESS_STATE_OFFLINE:
744 err = SCSI_DH_DEV_OFFLINED;
748 /* Useable path if active */
753 spin_unlock_irqrestore(&pg->lock, flags);
759 * alua_stpg - Issue a SET TARGET PORT GROUP command
761 * Issue a SET TARGET PORT GROUP command and evaluate the
762 * response. Returns SCSI_DH_RETRY per default to trigger
763 * a re-evaluation of the target group state or SCSI_DH_OK
764 * if no further action needs to be taken.
766 static unsigned alua_stpg(struct scsi_device *sdev, struct alua_port_group *pg)
769 struct scsi_sense_hdr sense_hdr;
771 if (!(pg->tpgs & TPGS_MODE_EXPLICIT)) {
772 /* Only implicit ALUA supported, retry */
773 return SCSI_DH_RETRY;
776 case SCSI_ACCESS_STATE_OPTIMAL:
778 case SCSI_ACCESS_STATE_ACTIVE:
779 if ((pg->flags & ALUA_OPTIMIZE_STPG) &&
781 (pg->tpgs & TPGS_MODE_IMPLICIT))
784 case SCSI_ACCESS_STATE_STANDBY:
785 case SCSI_ACCESS_STATE_UNAVAILABLE:
787 case SCSI_ACCESS_STATE_OFFLINE:
789 case SCSI_ACCESS_STATE_TRANSITIONING:
792 sdev_printk(KERN_INFO, sdev,
793 "%s: stpg failed, unhandled TPGS state %d",
794 ALUA_DH_NAME, pg->state);
795 return SCSI_DH_NOSYS;
797 retval = submit_stpg(sdev, pg->group_id, &sense_hdr);
800 if (retval < 0 || !scsi_sense_valid(&sense_hdr)) {
801 sdev_printk(KERN_INFO, sdev,
802 "%s: stpg failed, result %d",
803 ALUA_DH_NAME, retval);
805 return SCSI_DH_DEV_TEMP_BUSY;
807 sdev_printk(KERN_INFO, sdev, "%s: stpg failed\n",
809 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
813 return SCSI_DH_RETRY;
817 * The caller must call scsi_device_put() on the returned pointer if it is not
820 static struct scsi_device * __must_check
821 alua_rtpg_select_sdev(struct alua_port_group *pg)
823 struct alua_dh_data *h;
824 struct scsi_device *sdev = NULL, *prev_sdev;
826 lockdep_assert_held(&pg->lock);
827 if (WARN_ON(!pg->rtpg_sdev))
831 * RCU protection isn't necessary for dh_list here
832 * as we hold pg->lock, but for access to h->pg.
835 list_for_each_entry_rcu(h, &pg->dh_list, node) {
838 if (h->sdev == pg->rtpg_sdev) {
842 if (rcu_dereference(h->pg) == pg &&
844 !scsi_device_get(h->sdev)) {
852 pr_warn("%s: no device found for rtpg\n",
854 (char *)pg->device_id_str : "(nameless PG)"));
858 sdev_printk(KERN_INFO, sdev, "rtpg retry on different device\n");
860 prev_sdev = pg->rtpg_sdev;
861 pg->rtpg_sdev = sdev;
866 static void alua_rtpg_work(struct work_struct *work)
868 struct alua_port_group *pg =
869 container_of(work, struct alua_port_group, rtpg_work.work);
870 struct scsi_device *sdev, *prev_sdev = NULL;
871 LIST_HEAD(qdata_list);
872 int err = SCSI_DH_OK;
873 struct alua_queue_data *qdata, *tmp;
874 struct alua_dh_data *h;
877 spin_lock_irqsave(&pg->lock, flags);
878 sdev = pg->rtpg_sdev;
880 WARN_ON(pg->flags & ALUA_PG_RUN_RTPG);
881 WARN_ON(pg->flags & ALUA_PG_RUN_STPG);
882 spin_unlock_irqrestore(&pg->lock, flags);
883 kref_put(&pg->kref, release_port_group);
886 pg->flags |= ALUA_PG_RUNNING;
887 if (pg->flags & ALUA_PG_RUN_RTPG) {
888 int state = pg->state;
890 pg->flags &= ~ALUA_PG_RUN_RTPG;
891 spin_unlock_irqrestore(&pg->lock, flags);
892 if (state == SCSI_ACCESS_STATE_TRANSITIONING) {
893 if (alua_tur(sdev) == SCSI_DH_RETRY) {
894 spin_lock_irqsave(&pg->lock, flags);
895 pg->flags &= ~ALUA_PG_RUNNING;
896 pg->flags |= ALUA_PG_RUN_RTPG;
898 pg->interval = ALUA_RTPG_RETRY_DELAY;
899 spin_unlock_irqrestore(&pg->lock, flags);
900 queue_delayed_work(kaluad_wq, &pg->rtpg_work,
904 /* Send RTPG on failure or if TUR indicates SUCCESS */
906 err = alua_rtpg(sdev, pg);
907 spin_lock_irqsave(&pg->lock, flags);
909 /* If RTPG failed on the current device, try using another */
910 if (err == SCSI_DH_RES_TEMP_UNAVAIL &&
911 (prev_sdev = alua_rtpg_select_sdev(pg)))
912 err = SCSI_DH_IMM_RETRY;
914 if (err == SCSI_DH_RETRY || err == SCSI_DH_IMM_RETRY ||
915 pg->flags & ALUA_PG_RUN_RTPG) {
916 pg->flags &= ~ALUA_PG_RUNNING;
917 if (err == SCSI_DH_IMM_RETRY)
919 else if (!pg->interval && !(pg->flags & ALUA_PG_RUN_RTPG))
920 pg->interval = ALUA_RTPG_RETRY_DELAY;
921 pg->flags |= ALUA_PG_RUN_RTPG;
922 spin_unlock_irqrestore(&pg->lock, flags);
925 if (err != SCSI_DH_OK)
926 pg->flags &= ~ALUA_PG_RUN_STPG;
928 if (pg->flags & ALUA_PG_RUN_STPG) {
929 pg->flags &= ~ALUA_PG_RUN_STPG;
930 spin_unlock_irqrestore(&pg->lock, flags);
931 err = alua_stpg(sdev, pg);
932 spin_lock_irqsave(&pg->lock, flags);
933 if (err == SCSI_DH_RETRY || pg->flags & ALUA_PG_RUN_RTPG) {
934 pg->flags |= ALUA_PG_RUN_RTPG;
936 pg->flags &= ~ALUA_PG_RUNNING;
937 spin_unlock_irqrestore(&pg->lock, flags);
942 list_splice_init(&pg->rtpg_list, &qdata_list);
944 * We went through an RTPG, for good or bad.
945 * Re-enable all devices for the next attempt.
947 list_for_each_entry(h, &pg->dh_list, node)
949 pg->rtpg_sdev = NULL;
950 spin_unlock_irqrestore(&pg->lock, flags);
953 scsi_device_put(prev_sdev);
955 list_for_each_entry_safe(qdata, tmp, &qdata_list, entry) {
956 list_del(&qdata->entry);
957 if (qdata->callback_fn)
958 qdata->callback_fn(qdata->callback_data, err);
961 spin_lock_irqsave(&pg->lock, flags);
962 pg->flags &= ~ALUA_PG_RUNNING;
963 spin_unlock_irqrestore(&pg->lock, flags);
964 scsi_device_put(sdev);
965 kref_put(&pg->kref, release_port_group);
970 scsi_device_put(prev_sdev);
971 queue_delayed_work(kaluad_wq, &pg->rtpg_work, pg->interval * HZ);
975 * alua_rtpg_queue() - cause RTPG to be submitted asynchronously
976 * @pg: ALUA port group associated with @sdev.
977 * @sdev: SCSI device for which to submit an RTPG.
978 * @qdata: Information about the callback to invoke after the RTPG.
979 * @force: Whether or not to submit an RTPG if a work item that will submit an
980 * RTPG already has been scheduled.
982 * Returns true if and only if alua_rtpg_work() will be called asynchronously.
983 * That function is responsible for calling @qdata->fn().
985 * Context: may be called from atomic context (alua_check()) only if the caller
986 * holds an sdev reference.
988 static bool alua_rtpg_queue(struct alua_port_group *pg,
989 struct scsi_device *sdev,
990 struct alua_queue_data *qdata, bool force)
995 if (WARN_ON_ONCE(!pg) || scsi_device_get(sdev))
998 spin_lock_irqsave(&pg->lock, flags);
1000 list_add_tail(&qdata->entry, &pg->rtpg_list);
1001 pg->flags |= ALUA_PG_RUN_STPG;
1004 if (pg->rtpg_sdev == NULL) {
1005 struct alua_dh_data *h = sdev->handler_data;
1008 if (h && rcu_dereference(h->pg) == pg) {
1010 pg->flags |= ALUA_PG_RUN_RTPG;
1011 kref_get(&pg->kref);
1012 pg->rtpg_sdev = sdev;
1016 } else if (!(pg->flags & ALUA_PG_RUN_RTPG) && force) {
1017 pg->flags |= ALUA_PG_RUN_RTPG;
1018 /* Do not queue if the worker is already running */
1019 if (!(pg->flags & ALUA_PG_RUNNING)) {
1020 kref_get(&pg->kref);
1025 spin_unlock_irqrestore(&pg->lock, flags);
1028 if (queue_delayed_work(kaluad_wq, &pg->rtpg_work,
1029 msecs_to_jiffies(ALUA_RTPG_DELAY_MSECS)))
1032 kref_put(&pg->kref, release_port_group);
1035 scsi_device_put(sdev);
1041 * alua_initialize - Initialize ALUA state
1042 * @sdev: the device to be initialized
1044 * For the prep_fn to work correctly we have
1045 * to initialize the ALUA state for the device.
1047 static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h)
1049 int err = SCSI_DH_DEV_UNSUPP, tpgs;
1051 mutex_lock(&h->init_mutex);
1052 h->disabled = false;
1053 tpgs = alua_check_tpgs(sdev);
1054 if (tpgs != TPGS_MODE_NONE)
1055 err = alua_check_vpd(sdev, h, tpgs);
1056 h->init_error = err;
1057 mutex_unlock(&h->init_mutex);
1061 * alua_set_params - set/unset the optimize flag
1062 * @sdev: device on the path to be activated
1063 * params - parameters in the following format
1064 * "no_of_params\0param1\0param2\0param3\0...\0"
1065 * For example, to set the flag pass the following parameters
1066 * from multipath.conf
1067 * hardware_handler "2 alua 1"
1069 static int alua_set_params(struct scsi_device *sdev, const char *params)
1071 struct alua_dh_data *h = sdev->handler_data;
1072 struct alua_port_group *pg = NULL;
1073 unsigned int optimize = 0, argc;
1074 const char *p = params;
1075 int result = SCSI_DH_OK;
1076 unsigned long flags;
1078 if ((sscanf(params, "%u", &argc) != 1) || (argc != 1))
1083 if ((sscanf(p, "%u", &optimize) != 1) || (optimize > 1))
1087 pg = rcu_dereference(h->pg);
1092 spin_lock_irqsave(&pg->lock, flags);
1094 pg->flags |= ALUA_OPTIMIZE_STPG;
1096 pg->flags &= ~ALUA_OPTIMIZE_STPG;
1097 spin_unlock_irqrestore(&pg->lock, flags);
1104 * alua_activate - activate a path
1105 * @sdev: device on the path to be activated
1107 * We're currently switching the port group to be activated only and
1108 * let the array figure out the rest.
1109 * There may be other arrays which require us to switch all port groups
1110 * based on a certain policy. But until we actually encounter them it
1113 static int alua_activate(struct scsi_device *sdev,
1114 activate_complete fn, void *data)
1116 struct alua_dh_data *h = sdev->handler_data;
1117 int err = SCSI_DH_OK;
1118 struct alua_queue_data *qdata;
1119 struct alua_port_group *pg;
1121 qdata = kzalloc(sizeof(*qdata), GFP_KERNEL);
1123 err = SCSI_DH_RES_TEMP_UNAVAIL;
1126 qdata->callback_fn = fn;
1127 qdata->callback_data = data;
1129 mutex_lock(&h->init_mutex);
1131 pg = rcu_dereference(h->pg);
1132 if (!pg || !kref_get_unless_zero(&pg->kref)) {
1135 err = h->init_error;
1136 mutex_unlock(&h->init_mutex);
1140 mutex_unlock(&h->init_mutex);
1142 if (alua_rtpg_queue(pg, sdev, qdata, true))
1145 err = SCSI_DH_DEV_OFFLINED;
1146 kref_put(&pg->kref, release_port_group);
1154 * alua_check - check path status
1155 * @sdev: device on the path to be checked
1157 * Check the device status
1159 static void alua_check(struct scsi_device *sdev, bool force)
1161 struct alua_dh_data *h = sdev->handler_data;
1162 struct alua_port_group *pg;
1165 pg = rcu_dereference(h->pg);
1166 if (!pg || !kref_get_unless_zero(&pg->kref)) {
1171 alua_rtpg_queue(pg, sdev, NULL, force);
1172 kref_put(&pg->kref, release_port_group);
1176 * alua_prep_fn - request callback
1178 * Fail I/O to all paths not in state
1179 * active/optimized or active/non-optimized.
1181 static blk_status_t alua_prep_fn(struct scsi_device *sdev, struct request *req)
1183 struct alua_dh_data *h = sdev->handler_data;
1184 struct alua_port_group *pg;
1185 unsigned char state = SCSI_ACCESS_STATE_OPTIMAL;
1188 pg = rcu_dereference(h->pg);
1194 case SCSI_ACCESS_STATE_OPTIMAL:
1195 case SCSI_ACCESS_STATE_ACTIVE:
1196 case SCSI_ACCESS_STATE_LBA:
1197 case SCSI_ACCESS_STATE_TRANSITIONING:
1200 req->rq_flags |= RQF_QUIET;
1201 return BLK_STS_IOERR;
1205 static void alua_rescan(struct scsi_device *sdev)
1207 struct alua_dh_data *h = sdev->handler_data;
1209 alua_initialize(sdev, h);
1213 * alua_bus_attach - Attach device handler
1214 * @sdev: device to be attached to
1216 static int alua_bus_attach(struct scsi_device *sdev)
1218 struct alua_dh_data *h;
1221 h = kzalloc(sizeof(*h) , GFP_KERNEL);
1223 return SCSI_DH_NOMEM;
1224 spin_lock_init(&h->pg_lock);
1225 rcu_assign_pointer(h->pg, NULL);
1226 h->init_error = SCSI_DH_OK;
1228 INIT_LIST_HEAD(&h->node);
1230 mutex_init(&h->init_mutex);
1231 err = alua_initialize(sdev, h);
1232 if (err != SCSI_DH_OK && err != SCSI_DH_DEV_OFFLINED)
1235 sdev->handler_data = h;
1243 * alua_bus_detach - Detach device handler
1244 * @sdev: device to be detached from
1246 static void alua_bus_detach(struct scsi_device *sdev)
1248 struct alua_dh_data *h = sdev->handler_data;
1249 struct alua_port_group *pg;
1251 spin_lock(&h->pg_lock);
1252 pg = rcu_dereference_protected(h->pg, lockdep_is_held(&h->pg_lock));
1253 rcu_assign_pointer(h->pg, NULL);
1254 spin_unlock(&h->pg_lock);
1256 spin_lock_irq(&pg->lock);
1257 list_del_rcu(&h->node);
1258 spin_unlock_irq(&pg->lock);
1259 kref_put(&pg->kref, release_port_group);
1261 sdev->handler_data = NULL;
1266 static struct scsi_device_handler alua_dh = {
1267 .name = ALUA_DH_NAME,
1268 .module = THIS_MODULE,
1269 .attach = alua_bus_attach,
1270 .detach = alua_bus_detach,
1271 .prep_fn = alua_prep_fn,
1272 .check_sense = alua_check_sense,
1273 .activate = alua_activate,
1274 .rescan = alua_rescan,
1275 .set_params = alua_set_params,
1278 static int __init alua_init(void)
1282 kaluad_wq = alloc_workqueue("kaluad", WQ_MEM_RECLAIM, 0);
1286 r = scsi_register_device_handler(&alua_dh);
1288 printk(KERN_ERR "%s: Failed to register scsi device handler",
1290 destroy_workqueue(kaluad_wq);
1295 static void __exit alua_exit(void)
1297 scsi_unregister_device_handler(&alua_dh);
1298 destroy_workqueue(kaluad_wq);
1301 module_init(alua_init);
1302 module_exit(alua_exit);
1304 MODULE_DESCRIPTION("DM Multipath ALUA support");
1305 MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
1306 MODULE_LICENSE("GPL");
1307 MODULE_VERSION(ALUA_DH_VER);