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_dbg.h>
28 #include <scsi/scsi_eh.h>
29 #include <scsi/scsi_dh.h>
31 #define ALUA_DH_NAME "alua"
32 #define ALUA_DH_VER "1.3"
34 #define TPGS_STATE_OPTIMIZED 0x0
35 #define TPGS_STATE_NONOPTIMIZED 0x1
36 #define TPGS_STATE_STANDBY 0x2
37 #define TPGS_STATE_UNAVAILABLE 0x3
38 #define TPGS_STATE_LBA_DEPENDENT 0x4
39 #define TPGS_STATE_OFFLINE 0xe
40 #define TPGS_STATE_TRANSITIONING 0xf
42 #define TPGS_SUPPORT_NONE 0x00
43 #define TPGS_SUPPORT_OPTIMIZED 0x01
44 #define TPGS_SUPPORT_NONOPTIMIZED 0x02
45 #define TPGS_SUPPORT_STANDBY 0x04
46 #define TPGS_SUPPORT_UNAVAILABLE 0x08
47 #define TPGS_SUPPORT_LBA_DEPENDENT 0x10
48 #define TPGS_SUPPORT_OFFLINE 0x40
49 #define TPGS_SUPPORT_TRANSITION 0x80
51 #define RTPG_FMT_MASK 0x70
52 #define RTPG_FMT_EXT_HDR 0x10
54 #define TPGS_MODE_UNINITIALIZED -1
55 #define TPGS_MODE_NONE 0x0
56 #define TPGS_MODE_IMPLICIT 0x1
57 #define TPGS_MODE_EXPLICIT 0x2
59 #define ALUA_INQUIRY_SIZE 36
60 #define ALUA_FAILOVER_TIMEOUT 60
61 #define ALUA_FAILOVER_RETRIES 5
63 /* device handler flags */
64 #define ALUA_OPTIMIZE_STPG 1
65 #define ALUA_RTPG_EXT_HDR_UNSUPP 2
73 unsigned flags; /* used for optimizing STPG */
74 unsigned char inq[ALUA_INQUIRY_SIZE];
77 unsigned char transition_tmo;
78 unsigned char sense[SCSI_SENSE_BUFFERSIZE];
79 struct scsi_device *sdev;
80 activate_complete callback_fn;
84 #define ALUA_POLICY_SWITCH_CURRENT 0
85 #define ALUA_POLICY_SWITCH_ALL 1
87 static char print_alua_state(int);
89 static int realloc_buffer(struct alua_dh_data *h, unsigned len)
91 if (h->buff && h->buff != h->inq)
94 h->buff = kmalloc(len, GFP_NOIO);
97 h->bufflen = ALUA_INQUIRY_SIZE;
104 static struct request *get_alua_req(struct scsi_device *sdev,
105 void *buffer, unsigned buflen, int rw)
108 struct request_queue *q = sdev->request_queue;
110 rq = blk_get_request(q, rw, GFP_NOIO);
113 sdev_printk(KERN_INFO, sdev,
114 "%s: blk_get_request failed\n", __func__);
117 blk_rq_set_block_pc(rq);
119 if (buflen && blk_rq_map_kern(q, rq, buffer, buflen, GFP_NOIO)) {
121 sdev_printk(KERN_INFO, sdev,
122 "%s: blk_rq_map_kern failed\n", __func__);
126 rq->cmd_flags |= REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
128 rq->retries = ALUA_FAILOVER_RETRIES;
129 rq->timeout = ALUA_FAILOVER_TIMEOUT * HZ;
135 * submit_rtpg - Issue a REPORT TARGET GROUP STATES command
136 * @sdev: sdev the command should be sent to
138 static unsigned submit_rtpg(struct scsi_device *sdev, unsigned char *buff,
139 int bufflen, unsigned char *sense, int flags)
144 rq = get_alua_req(sdev, buff, bufflen, READ);
146 err = DRIVER_BUSY << 24;
150 /* Prepare the command. */
151 rq->cmd[0] = MAINTENANCE_IN;
152 if (!(flags & ALUA_RTPG_EXT_HDR_UNSUPP))
153 rq->cmd[1] = MI_REPORT_TARGET_PGS | MI_EXT_HDR_PARAM_FMT;
155 rq->cmd[1] = MI_REPORT_TARGET_PGS;
156 put_unaligned_be32(bufflen, &rq->cmd[6]);
157 rq->cmd_len = COMMAND_SIZE(MAINTENANCE_IN);
160 memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
163 blk_execute_rq(rq->q, NULL, rq, 1);
172 * submit_stpg - Issue a SET TARGET PORT GROUP command
174 * Currently we're only setting the current target port group state
175 * to 'active/optimized' and let the array firmware figure out
176 * the states of the remaining groups.
178 static unsigned submit_stpg(struct scsi_device *sdev, int group_id,
179 unsigned char *sense)
182 unsigned char stpg_data[8];
186 /* Prepare the data buffer */
187 memset(stpg_data, 0, stpg_len);
188 stpg_data[4] = TPGS_STATE_OPTIMIZED & 0x0f;
189 put_unaligned_be16(group_id, &stpg_data[6]);
191 rq = get_alua_req(sdev, stpg_data, stpg_len, WRITE);
193 return DRIVER_BUSY << 24;
195 /* Prepare the command. */
196 rq->cmd[0] = MAINTENANCE_OUT;
197 rq->cmd[1] = MO_SET_TARGET_PGS;
198 put_unaligned_be32(stpg_len, &rq->cmd[6]);
199 rq->cmd_len = COMMAND_SIZE(MAINTENANCE_OUT);
202 memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
205 blk_execute_rq(rq->q, NULL, rq, 1);
215 * alua_check_tpgs - Evaluate TPGS setting
216 * @sdev: device to be checked
218 * Examine the TPGS setting of the sdev to find out if ALUA
221 static int alua_check_tpgs(struct scsi_device *sdev)
223 int tpgs = TPGS_MODE_NONE;
226 * ALUA support for non-disk devices is fraught with
227 * difficulties, so disable it for now.
229 if (sdev->type != TYPE_DISK) {
230 sdev_printk(KERN_INFO, sdev,
231 "%s: disable for non-disk devices\n",
236 tpgs = scsi_device_tpgs(sdev);
238 case TPGS_MODE_EXPLICIT|TPGS_MODE_IMPLICIT:
239 sdev_printk(KERN_INFO, sdev,
240 "%s: supports implicit and explicit TPGS\n",
243 case TPGS_MODE_EXPLICIT:
244 sdev_printk(KERN_INFO, sdev, "%s: supports explicit TPGS\n",
247 case TPGS_MODE_IMPLICIT:
248 sdev_printk(KERN_INFO, sdev, "%s: supports implicit TPGS\n",
252 sdev_printk(KERN_INFO, sdev, "%s: not supported\n",
256 sdev_printk(KERN_INFO, sdev,
257 "%s: unsupported TPGS setting %d\n",
259 tpgs = TPGS_MODE_NONE;
267 * alua_check_vpd - Evaluate INQUIRY vpd page 0x83
268 * @sdev: device to be checked
270 * Extract the relative target port and the target port group
271 * descriptor from the list of identificators.
273 static int alua_check_vpd(struct scsi_device *sdev, struct alua_dh_data *h)
275 int rel_port = -1, group_id;
277 group_id = scsi_vpd_tpg_id(sdev, &rel_port);
280 * Internal error; TPGS supported but required
281 * VPD identification descriptors not present.
282 * Disable ALUA support
284 sdev_printk(KERN_INFO, sdev,
285 "%s: No target port descriptors found\n",
287 return SCSI_DH_DEV_UNSUPP;
289 h->state = TPGS_STATE_OPTIMIZED;
290 h->group_id = group_id;
292 sdev_printk(KERN_INFO, sdev,
293 "%s: port group %02x rel port %02x\n",
294 ALUA_DH_NAME, h->group_id, h->rel_port);
299 static char print_alua_state(int state)
302 case TPGS_STATE_OPTIMIZED:
304 case TPGS_STATE_NONOPTIMIZED:
306 case TPGS_STATE_STANDBY:
308 case TPGS_STATE_UNAVAILABLE:
310 case TPGS_STATE_LBA_DEPENDENT:
312 case TPGS_STATE_OFFLINE:
314 case TPGS_STATE_TRANSITIONING:
321 static int alua_check_sense(struct scsi_device *sdev,
322 struct scsi_sense_hdr *sense_hdr)
324 switch (sense_hdr->sense_key) {
326 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a)
328 * LUN Not Accessible - ALUA state transition
330 return ADD_TO_MLQUEUE;
333 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x00)
335 * Power On, Reset, or Bus Device Reset, just retry.
337 return ADD_TO_MLQUEUE;
338 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x04)
340 * Device internal reset
342 return ADD_TO_MLQUEUE;
343 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x01)
345 * Mode Parameters Changed
347 return ADD_TO_MLQUEUE;
348 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x06)
352 return ADD_TO_MLQUEUE;
353 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x07)
355 * Implicit ALUA state transition failed
357 return ADD_TO_MLQUEUE;
358 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x03)
360 * Inquiry data has changed
362 return ADD_TO_MLQUEUE;
363 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x0e)
365 * REPORTED_LUNS_DATA_HAS_CHANGED is reported
366 * when switching controllers on targets like
367 * Intel Multi-Flex. We can just retry.
369 return ADD_TO_MLQUEUE;
373 return SCSI_RETURN_NOT_HANDLED;
377 * alua_rtpg - Evaluate REPORT TARGET GROUP STATES
378 * @sdev: the device to be evaluated.
379 * @wait_for_transition: if nonzero, wait ALUA_FAILOVER_TIMEOUT seconds for device to exit transitioning state
381 * Evaluate the Target Port Group State.
382 * Returns SCSI_DH_DEV_OFFLINED if the path is
383 * found to be unusable.
385 static int alua_rtpg(struct scsi_device *sdev, struct alua_dh_data *h, int wait_for_transition)
387 struct scsi_sense_hdr sense_hdr;
388 int len, k, off, valid_states = 0;
390 unsigned err, retval;
391 unsigned long expiry, interval = 0;
392 unsigned int tpg_desc_tbl_off;
393 unsigned char orig_transition_tmo;
395 if (!h->transition_tmo)
396 expiry = round_jiffies_up(jiffies + ALUA_FAILOVER_TIMEOUT * HZ);
398 expiry = round_jiffies_up(jiffies + h->transition_tmo * HZ);
401 retval = submit_rtpg(sdev, h->buff, h->bufflen, h->sense, h->flags);
403 if (!scsi_normalize_sense(h->sense, SCSI_SENSE_BUFFERSIZE,
405 sdev_printk(KERN_INFO, sdev,
406 "%s: rtpg failed, result %d\n",
407 ALUA_DH_NAME, retval);
408 if (driver_byte(retval) == DRIVER_BUSY)
409 return SCSI_DH_DEV_TEMP_BUSY;
414 * submit_rtpg() has failed on existing arrays
415 * when requesting extended header info, and
416 * the array doesn't support extended headers,
417 * even though it shouldn't according to T10.
418 * The retry without rtpg_ext_hdr_req set
421 if (!(h->flags & ALUA_RTPG_EXT_HDR_UNSUPP) &&
422 sense_hdr.sense_key == ILLEGAL_REQUEST &&
423 sense_hdr.asc == 0x24 && sense_hdr.ascq == 0) {
424 h->flags |= ALUA_RTPG_EXT_HDR_UNSUPP;
428 * Retry on ALUA state transition or if any
429 * UNIT ATTENTION occurred.
431 if (sense_hdr.sense_key == NOT_READY &&
432 sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a)
434 else if (sense_hdr.sense_key == UNIT_ATTENTION)
436 if (err == SCSI_DH_RETRY && time_before(jiffies, expiry)) {
437 sdev_printk(KERN_ERR, sdev, "%s: rtpg retry\n",
439 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
442 sdev_printk(KERN_ERR, sdev, "%s: rtpg failed\n",
444 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
448 len = get_unaligned_be32(&h->buff[0]) + 4;
450 if (len > h->bufflen) {
451 /* Resubmit with the correct length */
452 if (realloc_buffer(h, len)) {
453 sdev_printk(KERN_WARNING, sdev,
454 "%s: kmalloc buffer failed\n",__func__);
455 /* Temporary failure, bypass */
456 return SCSI_DH_DEV_TEMP_BUSY;
461 orig_transition_tmo = h->transition_tmo;
462 if ((h->buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR && h->buff[5] != 0)
463 h->transition_tmo = h->buff[5];
465 h->transition_tmo = ALUA_FAILOVER_TIMEOUT;
467 if (wait_for_transition && (orig_transition_tmo != h->transition_tmo)) {
468 sdev_printk(KERN_INFO, sdev,
469 "%s: transition timeout set to %d seconds\n",
470 ALUA_DH_NAME, h->transition_tmo);
471 expiry = jiffies + h->transition_tmo * HZ;
474 if ((h->buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR)
475 tpg_desc_tbl_off = 8;
477 tpg_desc_tbl_off = 4;
479 for (k = tpg_desc_tbl_off, ucp = h->buff + tpg_desc_tbl_off;
481 k += off, ucp += off) {
483 if (h->group_id == get_unaligned_be16(&ucp[2])) {
484 h->state = ucp[0] & 0x0f;
485 h->pref = ucp[0] >> 7;
486 valid_states = ucp[1];
488 off = 8 + (ucp[7] * 4);
491 sdev_printk(KERN_INFO, sdev,
492 "%s: port group %02x state %c %s supports %c%c%c%c%c%c%c\n",
493 ALUA_DH_NAME, h->group_id, print_alua_state(h->state),
494 h->pref ? "preferred" : "non-preferred",
495 valid_states&TPGS_SUPPORT_TRANSITION?'T':'t',
496 valid_states&TPGS_SUPPORT_OFFLINE?'O':'o',
497 valid_states&TPGS_SUPPORT_LBA_DEPENDENT?'L':'l',
498 valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u',
499 valid_states&TPGS_SUPPORT_STANDBY?'S':'s',
500 valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n',
501 valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a');
504 case TPGS_STATE_TRANSITIONING:
505 if (wait_for_transition) {
506 if (time_before(jiffies, expiry)) {
507 /* State transition, retry */
517 /* Transitioning time exceeded, set port to standby */
518 h->state = TPGS_STATE_STANDBY;
520 case TPGS_STATE_OFFLINE:
522 err = SCSI_DH_DEV_OFFLINED;
525 /* Useable path if active */
533 * alua_stpg - Issue a SET TARGET PORT GROUP command
535 * Issue a SET TARGET PORT GROUP command and evaluate the
536 * response. Returns SCSI_DH_RETRY per default to trigger
537 * a re-evaluation of the target group state or SCSI_DH_OK
538 * if no further action needs to be taken.
540 static unsigned alua_stpg(struct scsi_device *sdev, struct alua_dh_data *h)
543 struct scsi_sense_hdr sense_hdr;
545 if (!(h->tpgs & TPGS_MODE_EXPLICIT)) {
546 /* Only implicit ALUA supported, retry */
547 return SCSI_DH_RETRY;
550 case TPGS_STATE_OPTIMIZED:
552 case TPGS_STATE_NONOPTIMIZED:
553 if ((h->flags & ALUA_OPTIMIZE_STPG) &&
555 (h->tpgs & TPGS_MODE_IMPLICIT))
558 case TPGS_STATE_STANDBY:
559 case TPGS_STATE_UNAVAILABLE:
561 case TPGS_STATE_OFFLINE:
563 case TPGS_STATE_TRANSITIONING:
566 sdev_printk(KERN_INFO, sdev,
567 "%s: stpg failed, unhandled TPGS state %d",
568 ALUA_DH_NAME, h->state);
569 return SCSI_DH_NOSYS;
571 retval = submit_stpg(sdev, h->group_id, h->sense);
574 if (!scsi_normalize_sense(h->sense, SCSI_SENSE_BUFFERSIZE,
576 sdev_printk(KERN_INFO, sdev,
577 "%s: stpg failed, result %d",
578 ALUA_DH_NAME, retval);
579 if (driver_byte(retval) == DRIVER_BUSY)
580 return SCSI_DH_DEV_TEMP_BUSY;
582 sdev_printk(KERN_INFO, h->sdev, "%s: stpg failed\n",
584 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
588 return SCSI_DH_RETRY;
592 * alua_initialize - Initialize ALUA state
593 * @sdev: the device to be initialized
595 * For the prep_fn to work correctly we have
596 * to initialize the ALUA state for the device.
598 static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h)
600 int err = SCSI_DH_DEV_UNSUPP;
602 h->tpgs = alua_check_tpgs(sdev);
603 if (h->tpgs == TPGS_MODE_NONE)
606 err = alua_check_vpd(sdev, h);
607 if (err != SCSI_DH_OK)
610 err = alua_rtpg(sdev, h, 0);
611 if (err != SCSI_DH_OK)
618 * alua_set_params - set/unset the optimize flag
619 * @sdev: device on the path to be activated
620 * params - parameters in the following format
621 * "no_of_params\0param1\0param2\0param3\0...\0"
622 * For example, to set the flag pass the following parameters
623 * from multipath.conf
624 * hardware_handler "2 alua 1"
626 static int alua_set_params(struct scsi_device *sdev, const char *params)
628 struct alua_dh_data *h = sdev->handler_data;
629 unsigned int optimize = 0, argc;
630 const char *p = params;
631 int result = SCSI_DH_OK;
633 if ((sscanf(params, "%u", &argc) != 1) || (argc != 1))
638 if ((sscanf(p, "%u", &optimize) != 1) || (optimize > 1))
642 h->flags |= ALUA_OPTIMIZE_STPG;
644 h->flags &= ~ALUA_OPTIMIZE_STPG;
649 static uint optimize_stpg;
650 module_param(optimize_stpg, uint, S_IRUGO|S_IWUSR);
651 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.");
654 * alua_activate - activate a path
655 * @sdev: device on the path to be activated
657 * We're currently switching the port group to be activated only and
658 * let the array figure out the rest.
659 * There may be other arrays which require us to switch all port groups
660 * based on a certain policy. But until we actually encounter them it
663 static int alua_activate(struct scsi_device *sdev,
664 activate_complete fn, void *data)
666 struct alua_dh_data *h = sdev->handler_data;
667 int err = SCSI_DH_OK;
669 err = alua_rtpg(sdev, h, 1);
670 if (err != SCSI_DH_OK)
674 h->flags |= ALUA_OPTIMIZE_STPG;
676 err = alua_stpg(sdev, h);
677 if (err == SCSI_DH_RETRY)
678 err = alua_rtpg(sdev, h, 1);
686 * alua_prep_fn - request callback
688 * Fail I/O to all paths not in state
689 * active/optimized or active/non-optimized.
691 static int alua_prep_fn(struct scsi_device *sdev, struct request *req)
693 struct alua_dh_data *h = sdev->handler_data;
694 int ret = BLKPREP_OK;
696 if (h->state == TPGS_STATE_TRANSITIONING)
698 else if (h->state != TPGS_STATE_OPTIMIZED &&
699 h->state != TPGS_STATE_NONOPTIMIZED &&
700 h->state != TPGS_STATE_LBA_DEPENDENT) {
702 req->cmd_flags |= REQ_QUIET;
709 * alua_bus_attach - Attach device handler
710 * @sdev: device to be attached to
712 static int alua_bus_attach(struct scsi_device *sdev)
714 struct alua_dh_data *h;
717 h = kzalloc(sizeof(*h) , GFP_KERNEL);
720 h->tpgs = TPGS_MODE_UNINITIALIZED;
721 h->state = TPGS_STATE_OPTIMIZED;
725 h->bufflen = ALUA_INQUIRY_SIZE;
728 err = alua_initialize(sdev, h);
729 if (err != SCSI_DH_OK && err != SCSI_DH_DEV_OFFLINED)
732 sdev->handler_data = h;
740 * alua_bus_detach - Detach device handler
741 * @sdev: device to be detached from
743 static void alua_bus_detach(struct scsi_device *sdev)
745 struct alua_dh_data *h = sdev->handler_data;
747 if (h->buff && h->inq != h->buff)
749 sdev->handler_data = NULL;
753 static struct scsi_device_handler alua_dh = {
754 .name = ALUA_DH_NAME,
755 .module = THIS_MODULE,
756 .attach = alua_bus_attach,
757 .detach = alua_bus_detach,
758 .prep_fn = alua_prep_fn,
759 .check_sense = alua_check_sense,
760 .activate = alua_activate,
761 .set_params = alua_set_params,
764 static int __init alua_init(void)
768 r = scsi_register_device_handler(&alua_dh);
770 printk(KERN_ERR "%s: Failed to register scsi device handler",
775 static void __exit alua_exit(void)
777 scsi_unregister_device_handler(&alua_dh);
780 module_init(alua_init);
781 module_exit(alua_exit);
783 MODULE_DESCRIPTION("DM Multipath ALUA support");
784 MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
785 MODULE_LICENSE("GPL");
786 MODULE_VERSION(ALUA_DH_VER);