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 __rcu
*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 bool 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(sdev
, cdb
, DMA_FROM_DEVICE
, buff
, bufflen
, NULL
,
155 sshdr
, ALUA_FAILOVER_TIMEOUT
* HZ
,
156 ALUA_FAILOVER_RETRIES
, req_flags
, 0, NULL
);
160 * submit_stpg - Issue a SET TARGET PORT GROUP command
162 * Currently we're only setting the current target port group state
163 * to 'active/optimized' and let the array firmware figure out
164 * the states of the remaining groups.
166 static int submit_stpg(struct scsi_device
*sdev
, int group_id
,
167 struct scsi_sense_hdr
*sshdr
)
169 u8 cdb
[COMMAND_SIZE(MAINTENANCE_OUT
)];
170 unsigned char stpg_data
[8];
172 int req_flags
= REQ_FAILFAST_DEV
| REQ_FAILFAST_TRANSPORT
|
175 /* Prepare the data buffer */
176 memset(stpg_data
, 0, stpg_len
);
177 stpg_data
[4] = SCSI_ACCESS_STATE_OPTIMAL
;
178 put_unaligned_be16(group_id
, &stpg_data
[6]);
180 /* Prepare the command. */
181 memset(cdb
, 0x0, COMMAND_SIZE(MAINTENANCE_OUT
));
182 cdb
[0] = MAINTENANCE_OUT
;
183 cdb
[1] = MO_SET_TARGET_PGS
;
184 put_unaligned_be32(stpg_len
, &cdb
[6]);
186 return scsi_execute(sdev
, cdb
, DMA_TO_DEVICE
, stpg_data
, stpg_len
, NULL
,
187 sshdr
, ALUA_FAILOVER_TIMEOUT
* HZ
,
188 ALUA_FAILOVER_RETRIES
, req_flags
, 0, NULL
);
191 static struct alua_port_group
*alua_find_get_pg(char *id_str
, size_t id_size
,
194 struct alua_port_group
*pg
;
196 if (!id_str
|| !id_size
|| !strlen(id_str
))
199 list_for_each_entry(pg
, &port_group_list
, node
) {
200 if (pg
->group_id
!= group_id
)
202 if (!pg
->device_id_len
|| pg
->device_id_len
!= id_size
)
204 if (strncmp(pg
->device_id_str
, id_str
, id_size
))
206 if (!kref_get_unless_zero(&pg
->kref
))
215 * alua_alloc_pg - Allocate a new port_group structure
217 * @h: alua device_handler data
218 * @group_id: port group id
220 * Allocate a new port_group structure for a given
223 static struct alua_port_group
*alua_alloc_pg(struct scsi_device
*sdev
,
224 int group_id
, int tpgs
)
226 struct alua_port_group
*pg
, *tmp_pg
;
228 pg
= kzalloc(sizeof(struct alua_port_group
), GFP_KERNEL
);
230 return ERR_PTR(-ENOMEM
);
232 pg
->device_id_len
= scsi_vpd_lun_id(sdev
, pg
->device_id_str
,
233 sizeof(pg
->device_id_str
));
234 if (pg
->device_id_len
<= 0) {
236 * TPGS supported but no device identification found.
237 * Generate private device identification.
239 sdev_printk(KERN_INFO
, sdev
,
240 "%s: No device descriptors found\n",
242 pg
->device_id_str
[0] = '\0';
243 pg
->device_id_len
= 0;
245 pg
->group_id
= group_id
;
247 pg
->state
= SCSI_ACCESS_STATE_OPTIMAL
;
249 pg
->flags
|= ALUA_OPTIMIZE_STPG
;
250 kref_init(&pg
->kref
);
251 INIT_DELAYED_WORK(&pg
->rtpg_work
, alua_rtpg_work
);
252 INIT_LIST_HEAD(&pg
->rtpg_list
);
253 INIT_LIST_HEAD(&pg
->node
);
254 INIT_LIST_HEAD(&pg
->dh_list
);
255 spin_lock_init(&pg
->lock
);
257 spin_lock(&port_group_lock
);
258 tmp_pg
= alua_find_get_pg(pg
->device_id_str
, pg
->device_id_len
,
261 spin_unlock(&port_group_lock
);
266 list_add(&pg
->node
, &port_group_list
);
267 spin_unlock(&port_group_lock
);
273 * alua_check_tpgs - Evaluate TPGS setting
274 * @sdev: device to be checked
276 * Examine the TPGS setting of the sdev to find out if ALUA
279 static int alua_check_tpgs(struct scsi_device
*sdev
)
281 int tpgs
= TPGS_MODE_NONE
;
284 * ALUA support for non-disk devices is fraught with
285 * difficulties, so disable it for now.
287 if (sdev
->type
!= TYPE_DISK
) {
288 sdev_printk(KERN_INFO
, sdev
,
289 "%s: disable for non-disk devices\n",
294 tpgs
= scsi_device_tpgs(sdev
);
296 case TPGS_MODE_EXPLICIT
|TPGS_MODE_IMPLICIT
:
297 sdev_printk(KERN_INFO
, sdev
,
298 "%s: supports implicit and explicit TPGS\n",
301 case TPGS_MODE_EXPLICIT
:
302 sdev_printk(KERN_INFO
, sdev
, "%s: supports explicit TPGS\n",
305 case TPGS_MODE_IMPLICIT
:
306 sdev_printk(KERN_INFO
, sdev
, "%s: supports implicit TPGS\n",
310 sdev_printk(KERN_INFO
, sdev
, "%s: not supported\n",
314 sdev_printk(KERN_INFO
, sdev
,
315 "%s: unsupported TPGS setting %d\n",
317 tpgs
= TPGS_MODE_NONE
;
325 * alua_check_vpd - Evaluate INQUIRY vpd page 0x83
326 * @sdev: device to be checked
328 * Extract the relative target port and the target port group
329 * descriptor from the list of identificators.
331 static int alua_check_vpd(struct scsi_device
*sdev
, struct alua_dh_data
*h
,
334 int rel_port
= -1, group_id
;
335 struct alua_port_group
*pg
, *old_pg
= NULL
;
336 bool pg_updated
= false;
339 group_id
= scsi_vpd_tpg_id(sdev
, &rel_port
);
342 * Internal error; TPGS supported but required
343 * VPD identification descriptors not present.
344 * Disable ALUA support
346 sdev_printk(KERN_INFO
, sdev
,
347 "%s: No target port descriptors found\n",
349 return SCSI_DH_DEV_UNSUPP
;
352 pg
= alua_alloc_pg(sdev
, group_id
, tpgs
);
354 if (PTR_ERR(pg
) == -ENOMEM
)
355 return SCSI_DH_NOMEM
;
356 return SCSI_DH_DEV_UNSUPP
;
358 if (pg
->device_id_len
)
359 sdev_printk(KERN_INFO
, sdev
,
360 "%s: device %s port group %x rel port %x\n",
361 ALUA_DH_NAME
, pg
->device_id_str
,
364 sdev_printk(KERN_INFO
, sdev
,
365 "%s: port group %x rel port %x\n",
366 ALUA_DH_NAME
, group_id
, rel_port
);
368 /* Check for existing port group references */
369 spin_lock(&h
->pg_lock
);
370 old_pg
= rcu_dereference_protected(h
->pg
, lockdep_is_held(&h
->pg_lock
));
372 /* port group has changed. Update to new port group */
374 spin_lock_irqsave(&old_pg
->lock
, flags
);
375 list_del_rcu(&h
->node
);
376 spin_unlock_irqrestore(&old_pg
->lock
, flags
);
378 rcu_assign_pointer(h
->pg
, pg
);
382 spin_lock_irqsave(&pg
->lock
, flags
);
383 if (sdev
->synchronous_alua
)
384 pg
->flags
|= ALUA_SYNC_STPG
;
386 list_add_rcu(&h
->node
, &pg
->dh_list
);
387 spin_unlock_irqrestore(&pg
->lock
, flags
);
389 alua_rtpg_queue(rcu_dereference_protected(h
->pg
,
390 lockdep_is_held(&h
->pg_lock
)),
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
);
796 kref_put(&pg
->kref
, release_port_group
);
799 if (pg
->flags
& ALUA_SYNC_STPG
)
800 alua_wq
= kaluad_sync_wq
;
801 pg
->flags
|= ALUA_PG_RUNNING
;
802 if (pg
->flags
& ALUA_PG_RUN_RTPG
) {
803 int state
= pg
->state
;
805 pg
->flags
&= ~ALUA_PG_RUN_RTPG
;
806 spin_unlock_irqrestore(&pg
->lock
, flags
);
807 if (state
== SCSI_ACCESS_STATE_TRANSITIONING
) {
808 if (alua_tur(sdev
) == SCSI_DH_RETRY
) {
809 spin_lock_irqsave(&pg
->lock
, flags
);
810 pg
->flags
&= ~ALUA_PG_RUNNING
;
811 pg
->flags
|= ALUA_PG_RUN_RTPG
;
812 spin_unlock_irqrestore(&pg
->lock
, flags
);
813 queue_delayed_work(alua_wq
, &pg
->rtpg_work
,
817 /* Send RTPG on failure or if TUR indicates SUCCESS */
819 err
= alua_rtpg(sdev
, pg
);
820 spin_lock_irqsave(&pg
->lock
, flags
);
821 if (err
== SCSI_DH_RETRY
|| pg
->flags
& ALUA_PG_RUN_RTPG
) {
822 pg
->flags
&= ~ALUA_PG_RUNNING
;
823 pg
->flags
|= ALUA_PG_RUN_RTPG
;
824 spin_unlock_irqrestore(&pg
->lock
, flags
);
825 queue_delayed_work(alua_wq
, &pg
->rtpg_work
,
829 if (err
!= SCSI_DH_OK
)
830 pg
->flags
&= ~ALUA_PG_RUN_STPG
;
832 if (pg
->flags
& ALUA_PG_RUN_STPG
) {
833 pg
->flags
&= ~ALUA_PG_RUN_STPG
;
834 spin_unlock_irqrestore(&pg
->lock
, flags
);
835 err
= alua_stpg(sdev
, pg
);
836 spin_lock_irqsave(&pg
->lock
, flags
);
837 if (err
== SCSI_DH_RETRY
|| pg
->flags
& ALUA_PG_RUN_RTPG
) {
838 pg
->flags
|= ALUA_PG_RUN_RTPG
;
840 pg
->flags
&= ~ALUA_PG_RUNNING
;
841 spin_unlock_irqrestore(&pg
->lock
, flags
);
842 queue_delayed_work(alua_wq
, &pg
->rtpg_work
,
848 list_splice_init(&pg
->rtpg_list
, &qdata_list
);
849 pg
->rtpg_sdev
= NULL
;
850 spin_unlock_irqrestore(&pg
->lock
, flags
);
852 list_for_each_entry_safe(qdata
, tmp
, &qdata_list
, entry
) {
853 list_del(&qdata
->entry
);
854 if (qdata
->callback_fn
)
855 qdata
->callback_fn(qdata
->callback_data
, err
);
858 spin_lock_irqsave(&pg
->lock
, flags
);
859 pg
->flags
&= ~ALUA_PG_RUNNING
;
860 spin_unlock_irqrestore(&pg
->lock
, flags
);
861 scsi_device_put(sdev
);
862 kref_put(&pg
->kref
, release_port_group
);
866 * alua_rtpg_queue() - cause RTPG to be submitted asynchronously
868 * Returns true if and only if alua_rtpg_work() will be called asynchronously.
869 * That function is responsible for calling @qdata->fn().
871 static bool alua_rtpg_queue(struct alua_port_group
*pg
,
872 struct scsi_device
*sdev
,
873 struct alua_queue_data
*qdata
, bool force
)
877 struct workqueue_struct
*alua_wq
= kaluad_wq
;
879 if (WARN_ON_ONCE(!pg
) || scsi_device_get(sdev
))
882 spin_lock_irqsave(&pg
->lock
, flags
);
884 list_add_tail(&qdata
->entry
, &pg
->rtpg_list
);
885 pg
->flags
|= ALUA_PG_RUN_STPG
;
888 if (pg
->rtpg_sdev
== NULL
) {
890 pg
->flags
|= ALUA_PG_RUN_RTPG
;
892 pg
->rtpg_sdev
= sdev
;
894 } else if (!(pg
->flags
& ALUA_PG_RUN_RTPG
) && force
) {
895 pg
->flags
|= ALUA_PG_RUN_RTPG
;
896 /* Do not queue if the worker is already running */
897 if (!(pg
->flags
& ALUA_PG_RUNNING
)) {
903 if (pg
->flags
& ALUA_SYNC_STPG
)
904 alua_wq
= kaluad_sync_wq
;
905 spin_unlock_irqrestore(&pg
->lock
, flags
);
908 if (queue_delayed_work(alua_wq
, &pg
->rtpg_work
,
909 msecs_to_jiffies(ALUA_RTPG_DELAY_MSECS
)))
912 kref_put(&pg
->kref
, release_port_group
);
915 scsi_device_put(sdev
);
921 * alua_initialize - Initialize ALUA state
922 * @sdev: the device to be initialized
924 * For the prep_fn to work correctly we have
925 * to initialize the ALUA state for the device.
927 static int alua_initialize(struct scsi_device
*sdev
, struct alua_dh_data
*h
)
929 int err
= SCSI_DH_DEV_UNSUPP
, tpgs
;
931 mutex_lock(&h
->init_mutex
);
932 tpgs
= alua_check_tpgs(sdev
);
933 if (tpgs
!= TPGS_MODE_NONE
)
934 err
= alua_check_vpd(sdev
, h
, tpgs
);
936 mutex_unlock(&h
->init_mutex
);
940 * alua_set_params - set/unset the optimize flag
941 * @sdev: device on the path to be activated
942 * params - parameters in the following format
943 * "no_of_params\0param1\0param2\0param3\0...\0"
944 * For example, to set the flag pass the following parameters
945 * from multipath.conf
946 * hardware_handler "2 alua 1"
948 static int alua_set_params(struct scsi_device
*sdev
, const char *params
)
950 struct alua_dh_data
*h
= sdev
->handler_data
;
951 struct alua_port_group
*pg
= NULL
;
952 unsigned int optimize
= 0, argc
;
953 const char *p
= params
;
954 int result
= SCSI_DH_OK
;
957 if ((sscanf(params
, "%u", &argc
) != 1) || (argc
!= 1))
962 if ((sscanf(p
, "%u", &optimize
) != 1) || (optimize
> 1))
966 pg
= rcu_dereference(h
->pg
);
971 spin_lock_irqsave(&pg
->lock
, flags
);
973 pg
->flags
|= ALUA_OPTIMIZE_STPG
;
975 pg
->flags
&= ~ALUA_OPTIMIZE_STPG
;
976 spin_unlock_irqrestore(&pg
->lock
, flags
);
983 * alua_activate - activate a path
984 * @sdev: device on the path to be activated
986 * We're currently switching the port group to be activated only and
987 * let the array figure out the rest.
988 * There may be other arrays which require us to switch all port groups
989 * based on a certain policy. But until we actually encounter them it
992 static int alua_activate(struct scsi_device
*sdev
,
993 activate_complete fn
, void *data
)
995 struct alua_dh_data
*h
= sdev
->handler_data
;
996 int err
= SCSI_DH_OK
;
997 struct alua_queue_data
*qdata
;
998 struct alua_port_group
*pg
;
1000 qdata
= kzalloc(sizeof(*qdata
), GFP_KERNEL
);
1002 err
= SCSI_DH_RES_TEMP_UNAVAIL
;
1005 qdata
->callback_fn
= fn
;
1006 qdata
->callback_data
= data
;
1008 mutex_lock(&h
->init_mutex
);
1010 pg
= rcu_dereference(h
->pg
);
1011 if (!pg
|| !kref_get_unless_zero(&pg
->kref
)) {
1014 err
= h
->init_error
;
1015 mutex_unlock(&h
->init_mutex
);
1019 mutex_unlock(&h
->init_mutex
);
1021 if (alua_rtpg_queue(pg
, sdev
, qdata
, true))
1024 err
= SCSI_DH_DEV_OFFLINED
;
1025 kref_put(&pg
->kref
, release_port_group
);
1033 * alua_check - check path status
1034 * @sdev: device on the path to be checked
1036 * Check the device status
1038 static void alua_check(struct scsi_device
*sdev
, bool force
)
1040 struct alua_dh_data
*h
= sdev
->handler_data
;
1041 struct alua_port_group
*pg
;
1044 pg
= rcu_dereference(h
->pg
);
1045 if (!pg
|| !kref_get_unless_zero(&pg
->kref
)) {
1051 alua_rtpg_queue(pg
, sdev
, NULL
, force
);
1052 kref_put(&pg
->kref
, release_port_group
);
1056 * alua_prep_fn - request callback
1058 * Fail I/O to all paths not in state
1059 * active/optimized or active/non-optimized.
1061 static int alua_prep_fn(struct scsi_device
*sdev
, struct request
*req
)
1063 struct alua_dh_data
*h
= sdev
->handler_data
;
1064 struct alua_port_group
*pg
;
1065 unsigned char state
= SCSI_ACCESS_STATE_OPTIMAL
;
1066 int ret
= BLKPREP_OK
;
1069 pg
= rcu_dereference(h
->pg
);
1073 if (state
== SCSI_ACCESS_STATE_TRANSITIONING
)
1074 ret
= BLKPREP_DEFER
;
1075 else if (state
!= SCSI_ACCESS_STATE_OPTIMAL
&&
1076 state
!= SCSI_ACCESS_STATE_ACTIVE
&&
1077 state
!= SCSI_ACCESS_STATE_LBA
) {
1079 req
->rq_flags
|= RQF_QUIET
;
1085 static void alua_rescan(struct scsi_device
*sdev
)
1087 struct alua_dh_data
*h
= sdev
->handler_data
;
1089 alua_initialize(sdev
, h
);
1093 * alua_bus_attach - Attach device handler
1094 * @sdev: device to be attached to
1096 static int alua_bus_attach(struct scsi_device
*sdev
)
1098 struct alua_dh_data
*h
;
1099 int err
, ret
= -EINVAL
;
1101 h
= kzalloc(sizeof(*h
) , GFP_KERNEL
);
1104 spin_lock_init(&h
->pg_lock
);
1105 rcu_assign_pointer(h
->pg
, NULL
);
1106 h
->init_error
= SCSI_DH_OK
;
1108 INIT_LIST_HEAD(&h
->node
);
1110 mutex_init(&h
->init_mutex
);
1111 err
= alua_initialize(sdev
, h
);
1112 if (err
== SCSI_DH_NOMEM
)
1114 if (err
!= SCSI_DH_OK
&& err
!= SCSI_DH_DEV_OFFLINED
)
1117 sdev
->handler_data
= h
;
1125 * alua_bus_detach - Detach device handler
1126 * @sdev: device to be detached from
1128 static void alua_bus_detach(struct scsi_device
*sdev
)
1130 struct alua_dh_data
*h
= sdev
->handler_data
;
1131 struct alua_port_group
*pg
;
1133 spin_lock(&h
->pg_lock
);
1134 pg
= rcu_dereference_protected(h
->pg
, lockdep_is_held(&h
->pg_lock
));
1135 rcu_assign_pointer(h
->pg
, NULL
);
1137 spin_unlock(&h
->pg_lock
);
1139 spin_lock_irq(&pg
->lock
);
1140 list_del_rcu(&h
->node
);
1141 spin_unlock_irq(&pg
->lock
);
1142 kref_put(&pg
->kref
, release_port_group
);
1144 sdev
->handler_data
= NULL
;
1148 static struct scsi_device_handler alua_dh
= {
1149 .name
= ALUA_DH_NAME
,
1150 .module
= THIS_MODULE
,
1151 .attach
= alua_bus_attach
,
1152 .detach
= alua_bus_detach
,
1153 .prep_fn
= alua_prep_fn
,
1154 .check_sense
= alua_check_sense
,
1155 .activate
= alua_activate
,
1156 .rescan
= alua_rescan
,
1157 .set_params
= alua_set_params
,
1160 static int __init
alua_init(void)
1164 kaluad_wq
= alloc_workqueue("kaluad", WQ_MEM_RECLAIM
, 0);
1166 /* Temporary failure, bypass */
1167 return SCSI_DH_DEV_TEMP_BUSY
;
1169 kaluad_sync_wq
= create_workqueue("kaluad_sync");
1170 if (!kaluad_sync_wq
) {
1171 destroy_workqueue(kaluad_wq
);
1172 return SCSI_DH_DEV_TEMP_BUSY
;
1174 r
= scsi_register_device_handler(&alua_dh
);
1176 printk(KERN_ERR
"%s: Failed to register scsi device handler",
1178 destroy_workqueue(kaluad_sync_wq
);
1179 destroy_workqueue(kaluad_wq
);
1184 static void __exit
alua_exit(void)
1186 scsi_unregister_device_handler(&alua_dh
);
1187 destroy_workqueue(kaluad_sync_wq
);
1188 destroy_workqueue(kaluad_wq
);
1191 module_init(alua_init
);
1192 module_exit(alua_exit
);
1194 MODULE_DESCRIPTION("DM Multipath ALUA support");
1195 MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
1196 MODULE_LICENSE("GPL");
1197 MODULE_VERSION(ALUA_DH_VER
);