2 * Generic SCSI-3 ALUA SCSI Device Handler
4 * Copyright (C) 2007-2010 Hannes Reinecke, SUSE Linux Products GmbH.
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
22 #include <linux/slab.h>
23 #include <linux/delay.h>
24 #include <linux/module.h>
25 #include <asm/unaligned.h>
26 #include <scsi/scsi.h>
27 #include <scsi/scsi_proto.h>
28 #include <scsi/scsi_dbg.h>
29 #include <scsi/scsi_eh.h>
30 #include <scsi/scsi_dh.h>
32 #define ALUA_DH_NAME "alua"
33 #define ALUA_DH_VER "2.0"
35 #define TPGS_SUPPORT_NONE 0x00
36 #define TPGS_SUPPORT_OPTIMIZED 0x01
37 #define TPGS_SUPPORT_NONOPTIMIZED 0x02
38 #define TPGS_SUPPORT_STANDBY 0x04
39 #define TPGS_SUPPORT_UNAVAILABLE 0x08
40 #define TPGS_SUPPORT_LBA_DEPENDENT 0x10
41 #define TPGS_SUPPORT_OFFLINE 0x40
42 #define TPGS_SUPPORT_TRANSITION 0x80
44 #define RTPG_FMT_MASK 0x70
45 #define RTPG_FMT_EXT_HDR 0x10
47 #define TPGS_MODE_UNINITIALIZED -1
48 #define TPGS_MODE_NONE 0x0
49 #define TPGS_MODE_IMPLICIT 0x1
50 #define TPGS_MODE_EXPLICIT 0x2
52 #define ALUA_RTPG_SIZE 128
53 #define ALUA_FAILOVER_TIMEOUT 60
54 #define ALUA_FAILOVER_RETRIES 5
55 #define ALUA_RTPG_DELAY_MSECS 5
57 /* device handler flags */
58 #define ALUA_OPTIMIZE_STPG 0x01
59 #define ALUA_RTPG_EXT_HDR_UNSUPP 0x02
60 #define ALUA_SYNC_STPG 0x04
61 /* State machine flags */
62 #define ALUA_PG_RUN_RTPG 0x10
63 #define ALUA_PG_RUN_STPG 0x20
64 #define ALUA_PG_RUNNING 0x40
66 static uint optimize_stpg;
67 module_param(optimize_stpg, uint, S_IRUGO|S_IWUSR);
68 MODULE_PARM_DESC(optimize_stpg, "Allow use of a non-optimized path, rather than sending a STPG, when implicit TPGS is supported (0=No,1=Yes). Default is 0.");
70 static LIST_HEAD(port_group_list);
71 static DEFINE_SPINLOCK(port_group_lock);
72 static struct workqueue_struct *kaluad_wq;
73 static struct workqueue_struct *kaluad_sync_wq;
75 struct alua_port_group {
78 struct list_head node;
79 struct list_head dh_list;
80 unsigned char device_id_str[256];
86 unsigned flags; /* used for optimizing STPG */
87 unsigned char transition_tmo;
89 unsigned long interval;
90 struct delayed_work rtpg_work;
92 struct list_head rtpg_list;
93 struct scsi_device *rtpg_sdev;
97 struct list_head node;
98 struct alua_port_group *pg;
101 struct scsi_device *sdev;
103 struct mutex init_mutex;
106 struct alua_queue_data {
107 struct list_head entry;
108 activate_complete callback_fn;
112 #define ALUA_POLICY_SWITCH_CURRENT 0
113 #define ALUA_POLICY_SWITCH_ALL 1
115 static void alua_rtpg_work(struct work_struct *work);
116 static void alua_rtpg_queue(struct alua_port_group *pg,
117 struct scsi_device *sdev,
118 struct alua_queue_data *qdata, bool force);
119 static void alua_check(struct scsi_device *sdev, bool force);
121 static void release_port_group(struct kref *kref)
123 struct alua_port_group *pg;
125 pg = container_of(kref, struct alua_port_group, kref);
127 flush_delayed_work(&pg->rtpg_work);
128 spin_lock(&port_group_lock);
130 spin_unlock(&port_group_lock);
135 * submit_rtpg - Issue a REPORT TARGET GROUP STATES command
136 * @sdev: sdev the command should be sent to
138 static int submit_rtpg(struct scsi_device *sdev, unsigned char *buff,
139 int bufflen, struct scsi_sense_hdr *sshdr, int flags)
141 u8 cdb[COMMAND_SIZE(MAINTENANCE_IN)];
142 int req_flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
145 /* Prepare the command. */
146 memset(cdb, 0x0, COMMAND_SIZE(MAINTENANCE_IN));
147 cdb[0] = MAINTENANCE_IN;
148 if (!(flags & ALUA_RTPG_EXT_HDR_UNSUPP))
149 cdb[1] = MI_REPORT_TARGET_PGS | MI_EXT_HDR_PARAM_FMT;
151 cdb[1] = MI_REPORT_TARGET_PGS;
152 put_unaligned_be32(bufflen, &cdb[6]);
154 return scsi_execute_req_flags(sdev, cdb, DMA_FROM_DEVICE,
155 buff, bufflen, sshdr,
156 ALUA_FAILOVER_TIMEOUT * HZ,
157 ALUA_FAILOVER_RETRIES, NULL, req_flags);
161 * submit_stpg - Issue a SET TARGET PORT GROUP command
163 * Currently we're only setting the current target port group state
164 * to 'active/optimized' and let the array firmware figure out
165 * the states of the remaining groups.
167 static int submit_stpg(struct scsi_device *sdev, int group_id,
168 struct scsi_sense_hdr *sshdr)
170 u8 cdb[COMMAND_SIZE(MAINTENANCE_OUT)];
171 unsigned char stpg_data[8];
173 int req_flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
176 /* Prepare the data buffer */
177 memset(stpg_data, 0, stpg_len);
178 stpg_data[4] = SCSI_ACCESS_STATE_OPTIMAL;
179 put_unaligned_be16(group_id, &stpg_data[6]);
181 /* Prepare the command. */
182 memset(cdb, 0x0, COMMAND_SIZE(MAINTENANCE_OUT));
183 cdb[0] = MAINTENANCE_OUT;
184 cdb[1] = MO_SET_TARGET_PGS;
185 put_unaligned_be32(stpg_len, &cdb[6]);
187 return scsi_execute_req_flags(sdev, cdb, DMA_TO_DEVICE,
189 sshdr, ALUA_FAILOVER_TIMEOUT * HZ,
190 ALUA_FAILOVER_RETRIES, NULL, req_flags);
193 static struct alua_port_group *alua_find_get_pg(char *id_str, size_t id_size,
196 struct alua_port_group *pg;
198 if (!id_str || !id_size || !strlen(id_str))
201 list_for_each_entry(pg, &port_group_list, node) {
202 if (pg->group_id != group_id)
204 if (!pg->device_id_len || pg->device_id_len != id_size)
206 if (strncmp(pg->device_id_str, id_str, id_size))
208 if (!kref_get_unless_zero(&pg->kref))
217 * alua_alloc_pg - Allocate a new port_group structure
219 * @h: alua device_handler data
220 * @group_id: port group id
222 * Allocate a new port_group structure for a given
225 static struct alua_port_group *alua_alloc_pg(struct scsi_device *sdev,
226 int group_id, int tpgs)
228 struct alua_port_group *pg, *tmp_pg;
230 pg = kzalloc(sizeof(struct alua_port_group), GFP_KERNEL);
232 return ERR_PTR(-ENOMEM);
234 pg->device_id_len = scsi_vpd_lun_id(sdev, pg->device_id_str,
235 sizeof(pg->device_id_str));
236 if (pg->device_id_len <= 0) {
238 * TPGS supported but no device identification found.
239 * Generate private device identification.
241 sdev_printk(KERN_INFO, sdev,
242 "%s: No device descriptors found\n",
244 pg->device_id_str[0] = '\0';
245 pg->device_id_len = 0;
247 pg->group_id = group_id;
249 pg->state = SCSI_ACCESS_STATE_OPTIMAL;
251 pg->flags |= ALUA_OPTIMIZE_STPG;
252 kref_init(&pg->kref);
253 INIT_DELAYED_WORK(&pg->rtpg_work, alua_rtpg_work);
254 INIT_LIST_HEAD(&pg->rtpg_list);
255 INIT_LIST_HEAD(&pg->node);
256 INIT_LIST_HEAD(&pg->dh_list);
257 spin_lock_init(&pg->lock);
259 spin_lock(&port_group_lock);
260 tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len,
263 spin_unlock(&port_group_lock);
268 list_add(&pg->node, &port_group_list);
269 spin_unlock(&port_group_lock);
275 * alua_check_tpgs - Evaluate TPGS setting
276 * @sdev: device to be checked
278 * Examine the TPGS setting of the sdev to find out if ALUA
281 static int alua_check_tpgs(struct scsi_device *sdev)
283 int tpgs = TPGS_MODE_NONE;
286 * ALUA support for non-disk devices is fraught with
287 * difficulties, so disable it for now.
289 if (sdev->type != TYPE_DISK) {
290 sdev_printk(KERN_INFO, sdev,
291 "%s: disable for non-disk devices\n",
296 tpgs = scsi_device_tpgs(sdev);
298 case TPGS_MODE_EXPLICIT|TPGS_MODE_IMPLICIT:
299 sdev_printk(KERN_INFO, sdev,
300 "%s: supports implicit and explicit TPGS\n",
303 case TPGS_MODE_EXPLICIT:
304 sdev_printk(KERN_INFO, sdev, "%s: supports explicit TPGS\n",
307 case TPGS_MODE_IMPLICIT:
308 sdev_printk(KERN_INFO, sdev, "%s: supports implicit TPGS\n",
312 sdev_printk(KERN_INFO, sdev, "%s: not supported\n",
316 sdev_printk(KERN_INFO, sdev,
317 "%s: unsupported TPGS setting %d\n",
319 tpgs = TPGS_MODE_NONE;
327 * alua_check_vpd - Evaluate INQUIRY vpd page 0x83
328 * @sdev: device to be checked
330 * Extract the relative target port and the target port group
331 * descriptor from the list of identificators.
333 static int alua_check_vpd(struct scsi_device *sdev, struct alua_dh_data *h,
336 int rel_port = -1, group_id;
337 struct alua_port_group *pg, *old_pg = NULL;
338 bool pg_updated = false;
341 group_id = scsi_vpd_tpg_id(sdev, &rel_port);
344 * Internal error; TPGS supported but required
345 * VPD identification descriptors not present.
346 * Disable ALUA support
348 sdev_printk(KERN_INFO, sdev,
349 "%s: No target port descriptors found\n",
351 return SCSI_DH_DEV_UNSUPP;
354 pg = alua_alloc_pg(sdev, group_id, tpgs);
356 if (PTR_ERR(pg) == -ENOMEM)
357 return SCSI_DH_NOMEM;
358 return SCSI_DH_DEV_UNSUPP;
360 if (pg->device_id_len)
361 sdev_printk(KERN_INFO, sdev,
362 "%s: device %s port group %x rel port %x\n",
363 ALUA_DH_NAME, pg->device_id_str,
366 sdev_printk(KERN_INFO, sdev,
367 "%s: port group %x rel port %x\n",
368 ALUA_DH_NAME, group_id, rel_port);
370 /* Check for existing port group references */
371 spin_lock(&h->pg_lock);
374 /* port group has changed. Update to new port group */
376 spin_lock_irqsave(&old_pg->lock, flags);
377 list_del_rcu(&h->node);
378 spin_unlock_irqrestore(&old_pg->lock, flags);
380 rcu_assign_pointer(h->pg, pg);
384 spin_lock_irqsave(&pg->lock, flags);
385 if (sdev->synchronous_alua)
386 pg->flags |= ALUA_SYNC_STPG;
388 list_add_rcu(&h->node, &pg->dh_list);
389 spin_unlock_irqrestore(&pg->lock, flags);
391 alua_rtpg_queue(h->pg, sdev, NULL, true);
392 spin_unlock(&h->pg_lock);
395 kref_put(&old_pg->kref, release_port_group);
400 static char print_alua_state(unsigned char state)
403 case SCSI_ACCESS_STATE_OPTIMAL:
405 case SCSI_ACCESS_STATE_ACTIVE:
407 case SCSI_ACCESS_STATE_STANDBY:
409 case SCSI_ACCESS_STATE_UNAVAILABLE:
411 case SCSI_ACCESS_STATE_LBA:
413 case SCSI_ACCESS_STATE_OFFLINE:
415 case SCSI_ACCESS_STATE_TRANSITIONING:
422 static int alua_check_sense(struct scsi_device *sdev,
423 struct scsi_sense_hdr *sense_hdr)
425 switch (sense_hdr->sense_key) {
427 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a) {
429 * LUN Not Accessible - ALUA state transition
431 alua_check(sdev, false);
436 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x00) {
438 * Power On, Reset, or Bus Device Reset.
439 * Might have obscured a state transition,
440 * so schedule a recheck.
442 alua_check(sdev, true);
443 return ADD_TO_MLQUEUE;
445 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x04)
447 * Device internal reset
449 return ADD_TO_MLQUEUE;
450 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x01)
452 * Mode Parameters Changed
454 return ADD_TO_MLQUEUE;
455 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x06) {
459 alua_check(sdev, true);
460 return ADD_TO_MLQUEUE;
462 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x07) {
464 * Implicit ALUA state transition failed
466 alua_check(sdev, true);
467 return ADD_TO_MLQUEUE;
469 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x03)
471 * Inquiry data has changed
473 return ADD_TO_MLQUEUE;
474 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x0e)
476 * REPORTED_LUNS_DATA_HAS_CHANGED is reported
477 * when switching controllers on targets like
478 * Intel Multi-Flex. We can just retry.
480 return ADD_TO_MLQUEUE;
484 return SCSI_RETURN_NOT_HANDLED;
488 * alua_tur - Send a TEST UNIT READY
489 * @sdev: device to which the TEST UNIT READY command should be send
491 * Send a TEST UNIT READY to @sdev to figure out the device state
492 * Returns SCSI_DH_RETRY if the sense code is NOT READY/ALUA TRANSITIONING,
493 * SCSI_DH_OK if no error occurred, and SCSI_DH_IO otherwise.
495 static int alua_tur(struct scsi_device *sdev)
497 struct scsi_sense_hdr sense_hdr;
500 retval = scsi_test_unit_ready(sdev, ALUA_FAILOVER_TIMEOUT * HZ,
501 ALUA_FAILOVER_RETRIES, &sense_hdr);
502 if (sense_hdr.sense_key == NOT_READY &&
503 sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a)
504 return SCSI_DH_RETRY;
512 * alua_rtpg - Evaluate REPORT TARGET GROUP STATES
513 * @sdev: the device to be evaluated.
515 * Evaluate the Target Port Group State.
516 * Returns SCSI_DH_DEV_OFFLINED if the path is
517 * found to be unusable.
519 static int alua_rtpg(struct scsi_device *sdev, struct alua_port_group *pg)
521 struct scsi_sense_hdr sense_hdr;
522 struct alua_port_group *tmp_pg;
523 int len, k, off, valid_states = 0, bufflen = ALUA_RTPG_SIZE;
524 unsigned char *desc, *buff;
525 unsigned err, retval;
526 unsigned int tpg_desc_tbl_off;
527 unsigned char orig_transition_tmo;
531 unsigned long transition_tmo = ALUA_FAILOVER_TIMEOUT * HZ;
533 if (pg->transition_tmo)
534 transition_tmo = pg->transition_tmo * HZ;
536 pg->expiry = round_jiffies_up(jiffies + transition_tmo);
539 buff = kzalloc(bufflen, GFP_KERNEL);
541 return SCSI_DH_DEV_TEMP_BUSY;
545 retval = submit_rtpg(sdev, buff, bufflen, &sense_hdr, pg->flags);
548 if (!scsi_sense_valid(&sense_hdr)) {
549 sdev_printk(KERN_INFO, sdev,
550 "%s: rtpg failed, result %d\n",
551 ALUA_DH_NAME, retval);
553 if (driver_byte(retval) == DRIVER_ERROR)
554 return SCSI_DH_DEV_TEMP_BUSY;
559 * submit_rtpg() has failed on existing arrays
560 * when requesting extended header info, and
561 * the array doesn't support extended headers,
562 * even though it shouldn't according to T10.
563 * The retry without rtpg_ext_hdr_req set
566 if (!(pg->flags & ALUA_RTPG_EXT_HDR_UNSUPP) &&
567 sense_hdr.sense_key == ILLEGAL_REQUEST &&
568 sense_hdr.asc == 0x24 && sense_hdr.ascq == 0) {
569 pg->flags |= ALUA_RTPG_EXT_HDR_UNSUPP;
573 * Retry on ALUA state transition or if any
574 * UNIT ATTENTION occurred.
576 if (sense_hdr.sense_key == NOT_READY &&
577 sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a)
579 else if (sense_hdr.sense_key == UNIT_ATTENTION)
581 if (err == SCSI_DH_RETRY &&
582 pg->expiry != 0 && time_before(jiffies, pg->expiry)) {
583 sdev_printk(KERN_ERR, sdev, "%s: rtpg retry\n",
585 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
589 sdev_printk(KERN_ERR, sdev, "%s: rtpg failed\n",
591 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
597 len = get_unaligned_be32(&buff[0]) + 4;
600 /* Resubmit with the correct length */
603 buff = kmalloc(bufflen, GFP_KERNEL);
605 sdev_printk(KERN_WARNING, sdev,
606 "%s: kmalloc buffer failed\n",__func__);
607 /* Temporary failure, bypass */
609 return SCSI_DH_DEV_TEMP_BUSY;
614 orig_transition_tmo = pg->transition_tmo;
615 if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR && buff[5] != 0)
616 pg->transition_tmo = buff[5];
618 pg->transition_tmo = ALUA_FAILOVER_TIMEOUT;
620 if (orig_transition_tmo != pg->transition_tmo) {
621 sdev_printk(KERN_INFO, sdev,
622 "%s: transition timeout set to %d seconds\n",
623 ALUA_DH_NAME, pg->transition_tmo);
624 pg->expiry = jiffies + pg->transition_tmo * HZ;
627 if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR)
628 tpg_desc_tbl_off = 8;
630 tpg_desc_tbl_off = 4;
632 for (k = tpg_desc_tbl_off, desc = buff + tpg_desc_tbl_off;
634 k += off, desc += off) {
635 u16 group_id = get_unaligned_be16(&desc[2]);
637 spin_lock_irqsave(&port_group_lock, flags);
638 tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len,
640 spin_unlock_irqrestore(&port_group_lock, flags);
642 if (spin_trylock_irqsave(&tmp_pg->lock, flags)) {
643 if ((tmp_pg == pg) ||
644 !(tmp_pg->flags & ALUA_PG_RUNNING)) {
645 struct alua_dh_data *h;
647 tmp_pg->state = desc[0] & 0x0f;
648 tmp_pg->pref = desc[0] >> 7;
650 list_for_each_entry_rcu(h,
651 &tmp_pg->dh_list, node) {
652 /* h->sdev should always be valid */
654 h->sdev->access_state = desc[0];
659 valid_states = desc[1];
660 spin_unlock_irqrestore(&tmp_pg->lock, flags);
662 kref_put(&tmp_pg->kref, release_port_group);
664 off = 8 + (desc[7] * 4);
667 spin_lock_irqsave(&pg->lock, flags);
668 sdev_printk(KERN_INFO, sdev,
669 "%s: port group %02x state %c %s supports %c%c%c%c%c%c%c\n",
670 ALUA_DH_NAME, pg->group_id, print_alua_state(pg->state),
671 pg->pref ? "preferred" : "non-preferred",
672 valid_states&TPGS_SUPPORT_TRANSITION?'T':'t',
673 valid_states&TPGS_SUPPORT_OFFLINE?'O':'o',
674 valid_states&TPGS_SUPPORT_LBA_DEPENDENT?'L':'l',
675 valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u',
676 valid_states&TPGS_SUPPORT_STANDBY?'S':'s',
677 valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n',
678 valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a');
681 case SCSI_ACCESS_STATE_TRANSITIONING:
682 if (time_before(jiffies, pg->expiry)) {
683 /* State transition, retry */
687 struct alua_dh_data *h;
689 /* Transitioning time exceeded, set port to standby */
691 pg->state = SCSI_ACCESS_STATE_STANDBY;
694 list_for_each_entry_rcu(h, &pg->dh_list, node) {
696 h->sdev->access_state =
697 (pg->state & SCSI_ACCESS_STATE_MASK);
699 h->sdev->access_state |=
700 SCSI_ACCESS_STATE_PREFERRED;
705 case SCSI_ACCESS_STATE_OFFLINE:
707 err = SCSI_DH_DEV_OFFLINED;
711 /* Useable path if active */
716 spin_unlock_irqrestore(&pg->lock, flags);
722 * alua_stpg - Issue a SET TARGET PORT GROUP command
724 * Issue a SET TARGET PORT GROUP command and evaluate the
725 * response. Returns SCSI_DH_RETRY per default to trigger
726 * a re-evaluation of the target group state or SCSI_DH_OK
727 * if no further action needs to be taken.
729 static unsigned alua_stpg(struct scsi_device *sdev, struct alua_port_group *pg)
732 struct scsi_sense_hdr sense_hdr;
734 if (!(pg->tpgs & TPGS_MODE_EXPLICIT)) {
735 /* Only implicit ALUA supported, retry */
736 return SCSI_DH_RETRY;
739 case SCSI_ACCESS_STATE_OPTIMAL:
741 case SCSI_ACCESS_STATE_ACTIVE:
742 if ((pg->flags & ALUA_OPTIMIZE_STPG) &&
744 (pg->tpgs & TPGS_MODE_IMPLICIT))
747 case SCSI_ACCESS_STATE_STANDBY:
748 case SCSI_ACCESS_STATE_UNAVAILABLE:
750 case SCSI_ACCESS_STATE_OFFLINE:
752 case SCSI_ACCESS_STATE_TRANSITIONING:
755 sdev_printk(KERN_INFO, sdev,
756 "%s: stpg failed, unhandled TPGS state %d",
757 ALUA_DH_NAME, pg->state);
758 return SCSI_DH_NOSYS;
760 retval = submit_stpg(sdev, pg->group_id, &sense_hdr);
763 if (!scsi_sense_valid(&sense_hdr)) {
764 sdev_printk(KERN_INFO, sdev,
765 "%s: stpg failed, result %d",
766 ALUA_DH_NAME, retval);
767 if (driver_byte(retval) == DRIVER_ERROR)
768 return SCSI_DH_DEV_TEMP_BUSY;
770 sdev_printk(KERN_INFO, sdev, "%s: stpg failed\n",
772 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
776 return SCSI_DH_RETRY;
779 static void alua_rtpg_work(struct work_struct *work)
781 struct alua_port_group *pg =
782 container_of(work, struct alua_port_group, rtpg_work.work);
783 struct scsi_device *sdev;
784 LIST_HEAD(qdata_list);
785 int err = SCSI_DH_OK;
786 struct alua_queue_data *qdata, *tmp;
788 struct workqueue_struct *alua_wq = kaluad_wq;
790 spin_lock_irqsave(&pg->lock, flags);
791 sdev = pg->rtpg_sdev;
793 WARN_ON(pg->flags & ALUA_PG_RUN_RTPG);
794 WARN_ON(pg->flags & ALUA_PG_RUN_STPG);
795 spin_unlock_irqrestore(&pg->lock, flags);
798 if (pg->flags & ALUA_SYNC_STPG)
799 alua_wq = kaluad_sync_wq;
800 pg->flags |= ALUA_PG_RUNNING;
801 if (pg->flags & ALUA_PG_RUN_RTPG) {
802 int state = pg->state;
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 spin_unlock_irqrestore(&pg->lock, flags);
812 queue_delayed_work(alua_wq, &pg->rtpg_work,
816 /* Send RTPG on failure or if TUR indicates SUCCESS */
818 err = alua_rtpg(sdev, pg);
819 spin_lock_irqsave(&pg->lock, flags);
820 if (err == SCSI_DH_RETRY || pg->flags & ALUA_PG_RUN_RTPG) {
821 pg->flags &= ~ALUA_PG_RUNNING;
822 pg->flags |= ALUA_PG_RUN_RTPG;
823 spin_unlock_irqrestore(&pg->lock, flags);
824 queue_delayed_work(alua_wq, &pg->rtpg_work,
828 if (err != SCSI_DH_OK)
829 pg->flags &= ~ALUA_PG_RUN_STPG;
831 if (pg->flags & ALUA_PG_RUN_STPG) {
832 pg->flags &= ~ALUA_PG_RUN_STPG;
833 spin_unlock_irqrestore(&pg->lock, flags);
834 err = alua_stpg(sdev, pg);
835 spin_lock_irqsave(&pg->lock, flags);
836 if (err == SCSI_DH_RETRY || pg->flags & ALUA_PG_RUN_RTPG) {
837 pg->flags |= ALUA_PG_RUN_RTPG;
839 pg->flags &= ~ALUA_PG_RUNNING;
840 spin_unlock_irqrestore(&pg->lock, flags);
841 queue_delayed_work(alua_wq, &pg->rtpg_work,
847 list_splice_init(&pg->rtpg_list, &qdata_list);
848 pg->rtpg_sdev = NULL;
849 spin_unlock_irqrestore(&pg->lock, flags);
851 list_for_each_entry_safe(qdata, tmp, &qdata_list, entry) {
852 list_del(&qdata->entry);
853 if (qdata->callback_fn)
854 qdata->callback_fn(qdata->callback_data, err);
857 spin_lock_irqsave(&pg->lock, flags);
858 pg->flags &= ~ALUA_PG_RUNNING;
859 spin_unlock_irqrestore(&pg->lock, flags);
860 scsi_device_put(sdev);
861 kref_put(&pg->kref, release_port_group);
864 static void alua_rtpg_queue(struct alua_port_group *pg,
865 struct scsi_device *sdev,
866 struct alua_queue_data *qdata, bool force)
870 struct workqueue_struct *alua_wq = kaluad_wq;
875 spin_lock_irqsave(&pg->lock, flags);
877 list_add_tail(&qdata->entry, &pg->rtpg_list);
878 pg->flags |= ALUA_PG_RUN_STPG;
881 if (pg->rtpg_sdev == NULL) {
883 pg->flags |= ALUA_PG_RUN_RTPG;
885 pg->rtpg_sdev = sdev;
886 scsi_device_get(sdev);
888 } else if (!(pg->flags & ALUA_PG_RUN_RTPG) && force) {
889 pg->flags |= ALUA_PG_RUN_RTPG;
890 /* Do not queue if the worker is already running */
891 if (!(pg->flags & ALUA_PG_RUNNING)) {
897 if (pg->flags & ALUA_SYNC_STPG)
898 alua_wq = kaluad_sync_wq;
899 spin_unlock_irqrestore(&pg->lock, flags);
902 !queue_delayed_work(alua_wq, &pg->rtpg_work,
903 msecs_to_jiffies(ALUA_RTPG_DELAY_MSECS))) {
904 scsi_device_put(sdev);
905 kref_put(&pg->kref, release_port_group);
910 * alua_initialize - Initialize ALUA state
911 * @sdev: the device to be initialized
913 * For the prep_fn to work correctly we have
914 * to initialize the ALUA state for the device.
916 static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h)
918 int err = SCSI_DH_DEV_UNSUPP, tpgs;
920 mutex_lock(&h->init_mutex);
921 tpgs = alua_check_tpgs(sdev);
922 if (tpgs != TPGS_MODE_NONE)
923 err = alua_check_vpd(sdev, h, tpgs);
925 mutex_unlock(&h->init_mutex);
929 * alua_set_params - set/unset the optimize flag
930 * @sdev: device on the path to be activated
931 * params - parameters in the following format
932 * "no_of_params\0param1\0param2\0param3\0...\0"
933 * For example, to set the flag pass the following parameters
934 * from multipath.conf
935 * hardware_handler "2 alua 1"
937 static int alua_set_params(struct scsi_device *sdev, const char *params)
939 struct alua_dh_data *h = sdev->handler_data;
940 struct alua_port_group __rcu *pg = NULL;
941 unsigned int optimize = 0, argc;
942 const char *p = params;
943 int result = SCSI_DH_OK;
946 if ((sscanf(params, "%u", &argc) != 1) || (argc != 1))
951 if ((sscanf(p, "%u", &optimize) != 1) || (optimize > 1))
955 pg = rcu_dereference(h->pg);
960 spin_lock_irqsave(&pg->lock, flags);
962 pg->flags |= ALUA_OPTIMIZE_STPG;
964 pg->flags &= ~ALUA_OPTIMIZE_STPG;
965 spin_unlock_irqrestore(&pg->lock, flags);
972 * alua_activate - activate a path
973 * @sdev: device on the path to be activated
975 * We're currently switching the port group to be activated only and
976 * let the array figure out the rest.
977 * There may be other arrays which require us to switch all port groups
978 * based on a certain policy. But until we actually encounter them it
981 static int alua_activate(struct scsi_device *sdev,
982 activate_complete fn, void *data)
984 struct alua_dh_data *h = sdev->handler_data;
985 int err = SCSI_DH_OK;
986 struct alua_queue_data *qdata;
987 struct alua_port_group __rcu *pg;
989 qdata = kzalloc(sizeof(*qdata), GFP_KERNEL);
991 err = SCSI_DH_RES_TEMP_UNAVAIL;
994 qdata->callback_fn = fn;
995 qdata->callback_data = data;
997 mutex_lock(&h->init_mutex);
999 pg = rcu_dereference(h->pg);
1000 if (!pg || !kref_get_unless_zero(&pg->kref)) {
1003 err = h->init_error;
1004 mutex_unlock(&h->init_mutex);
1009 mutex_unlock(&h->init_mutex);
1011 alua_rtpg_queue(pg, sdev, qdata, true);
1012 kref_put(&pg->kref, release_port_group);
1020 * alua_check - check path status
1021 * @sdev: device on the path to be checked
1023 * Check the device status
1025 static void alua_check(struct scsi_device *sdev, bool force)
1027 struct alua_dh_data *h = sdev->handler_data;
1028 struct alua_port_group *pg;
1031 pg = rcu_dereference(h->pg);
1032 if (!pg || !kref_get_unless_zero(&pg->kref)) {
1038 alua_rtpg_queue(pg, sdev, NULL, force);
1039 kref_put(&pg->kref, release_port_group);
1043 * alua_prep_fn - request callback
1045 * Fail I/O to all paths not in state
1046 * active/optimized or active/non-optimized.
1048 static int alua_prep_fn(struct scsi_device *sdev, struct request *req)
1050 struct alua_dh_data *h = sdev->handler_data;
1051 struct alua_port_group __rcu *pg;
1052 unsigned char state = SCSI_ACCESS_STATE_OPTIMAL;
1053 int ret = BLKPREP_OK;
1056 pg = rcu_dereference(h->pg);
1060 if (state == SCSI_ACCESS_STATE_TRANSITIONING)
1061 ret = BLKPREP_DEFER;
1062 else if (state != SCSI_ACCESS_STATE_OPTIMAL &&
1063 state != SCSI_ACCESS_STATE_ACTIVE &&
1064 state != SCSI_ACCESS_STATE_LBA) {
1066 req->cmd_flags |= REQ_QUIET;
1072 static void alua_rescan(struct scsi_device *sdev)
1074 struct alua_dh_data *h = sdev->handler_data;
1076 alua_initialize(sdev, h);
1080 * alua_bus_attach - Attach device handler
1081 * @sdev: device to be attached to
1083 static int alua_bus_attach(struct scsi_device *sdev)
1085 struct alua_dh_data *h;
1086 int err, ret = -EINVAL;
1088 h = kzalloc(sizeof(*h) , GFP_KERNEL);
1091 spin_lock_init(&h->pg_lock);
1092 rcu_assign_pointer(h->pg, NULL);
1093 h->init_error = SCSI_DH_OK;
1095 INIT_LIST_HEAD(&h->node);
1097 mutex_init(&h->init_mutex);
1098 err = alua_initialize(sdev, h);
1099 if (err == SCSI_DH_NOMEM)
1101 if (err != SCSI_DH_OK && err != SCSI_DH_DEV_OFFLINED)
1104 sdev->handler_data = h;
1112 * alua_bus_detach - Detach device handler
1113 * @sdev: device to be detached from
1115 static void alua_bus_detach(struct scsi_device *sdev)
1117 struct alua_dh_data *h = sdev->handler_data;
1118 struct alua_port_group *pg;
1120 spin_lock(&h->pg_lock);
1122 rcu_assign_pointer(h->pg, NULL);
1124 spin_unlock(&h->pg_lock);
1126 spin_lock_irq(&pg->lock);
1127 list_del_rcu(&h->node);
1128 spin_unlock_irq(&pg->lock);
1129 kref_put(&pg->kref, release_port_group);
1131 sdev->handler_data = NULL;
1135 static struct scsi_device_handler alua_dh = {
1136 .name = ALUA_DH_NAME,
1137 .module = THIS_MODULE,
1138 .attach = alua_bus_attach,
1139 .detach = alua_bus_detach,
1140 .prep_fn = alua_prep_fn,
1141 .check_sense = alua_check_sense,
1142 .activate = alua_activate,
1143 .rescan = alua_rescan,
1144 .set_params = alua_set_params,
1147 static int __init alua_init(void)
1151 kaluad_wq = alloc_workqueue("kaluad", WQ_MEM_RECLAIM, 0);
1153 /* Temporary failure, bypass */
1154 return SCSI_DH_DEV_TEMP_BUSY;
1156 kaluad_sync_wq = create_workqueue("kaluad_sync");
1157 if (!kaluad_sync_wq) {
1158 destroy_workqueue(kaluad_wq);
1159 return SCSI_DH_DEV_TEMP_BUSY;
1161 r = scsi_register_device_handler(&alua_dh);
1163 printk(KERN_ERR "%s: Failed to register scsi device handler",
1165 destroy_workqueue(kaluad_sync_wq);
1166 destroy_workqueue(kaluad_wq);
1171 static void __exit alua_exit(void)
1173 scsi_unregister_device_handler(&alua_dh);
1174 destroy_workqueue(kaluad_sync_wq);
1175 destroy_workqueue(kaluad_wq);
1178 module_init(alua_init);
1179 module_exit(alua_exit);
1181 MODULE_DESCRIPTION("DM Multipath ALUA support");
1182 MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
1183 MODULE_LICENSE("GPL");
1184 MODULE_VERSION(ALUA_DH_VER);