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 <scsi/scsi.h>
25 #include <scsi/scsi_eh.h>
26 #include <scsi/scsi_dh.h>
28 #define ALUA_DH_NAME "alua"
29 #define ALUA_DH_VER "1.3"
31 #define TPGS_STATE_OPTIMIZED 0x0
32 #define TPGS_STATE_NONOPTIMIZED 0x1
33 #define TPGS_STATE_STANDBY 0x2
34 #define TPGS_STATE_UNAVAILABLE 0x3
35 #define TPGS_STATE_LBA_DEPENDENT 0x4
36 #define TPGS_STATE_OFFLINE 0xe
37 #define TPGS_STATE_TRANSITIONING 0xf
39 #define TPGS_SUPPORT_NONE 0x00
40 #define TPGS_SUPPORT_OPTIMIZED 0x01
41 #define TPGS_SUPPORT_NONOPTIMIZED 0x02
42 #define TPGS_SUPPORT_STANDBY 0x04
43 #define TPGS_SUPPORT_UNAVAILABLE 0x08
44 #define TPGS_SUPPORT_LBA_DEPENDENT 0x10
45 #define TPGS_SUPPORT_OFFLINE 0x40
46 #define TPGS_SUPPORT_TRANSITION 0x80
48 #define TPGS_MODE_UNINITIALIZED -1
49 #define TPGS_MODE_NONE 0x0
50 #define TPGS_MODE_IMPLICIT 0x1
51 #define TPGS_MODE_EXPLICIT 0x2
53 #define ALUA_INQUIRY_SIZE 36
54 #define ALUA_FAILOVER_TIMEOUT (60 * HZ)
55 #define ALUA_FAILOVER_RETRIES 5
62 unsigned char inq[ALUA_INQUIRY_SIZE];
65 unsigned char sense[SCSI_SENSE_BUFFERSIZE];
67 struct scsi_device *sdev;
68 activate_complete callback_fn;
72 #define ALUA_POLICY_SWITCH_CURRENT 0
73 #define ALUA_POLICY_SWITCH_ALL 1
75 static char print_alua_state(int);
76 static int alua_check_sense(struct scsi_device *, struct scsi_sense_hdr *);
78 static inline struct alua_dh_data *get_alua_data(struct scsi_device *sdev)
80 struct scsi_dh_data *scsi_dh_data = sdev->scsi_dh_data;
81 BUG_ON(scsi_dh_data == NULL);
82 return ((struct alua_dh_data *) scsi_dh_data->buf);
85 static int realloc_buffer(struct alua_dh_data *h, unsigned len)
87 if (h->buff && h->buff != h->inq)
90 h->buff = kmalloc(len, GFP_NOIO);
93 h->bufflen = ALUA_INQUIRY_SIZE;
100 static struct request *get_alua_req(struct scsi_device *sdev,
101 void *buffer, unsigned buflen, int rw)
104 struct request_queue *q = sdev->request_queue;
106 rq = blk_get_request(q, rw, GFP_NOIO);
109 sdev_printk(KERN_INFO, sdev,
110 "%s: blk_get_request failed\n", __func__);
114 if (buflen && blk_rq_map_kern(q, rq, buffer, buflen, GFP_NOIO)) {
116 sdev_printk(KERN_INFO, sdev,
117 "%s: blk_rq_map_kern failed\n", __func__);
121 rq->cmd_type = REQ_TYPE_BLOCK_PC;
122 rq->cmd_flags |= REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
124 rq->retries = ALUA_FAILOVER_RETRIES;
125 rq->timeout = ALUA_FAILOVER_TIMEOUT;
131 * submit_std_inquiry - Issue a standard INQUIRY command
132 * @sdev: sdev the command should be send to
134 static int submit_std_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
137 int err = SCSI_DH_RES_TEMP_UNAVAIL;
139 rq = get_alua_req(sdev, h->inq, ALUA_INQUIRY_SIZE, READ);
143 /* Prepare the command. */
144 rq->cmd[0] = INQUIRY;
147 rq->cmd[4] = ALUA_INQUIRY_SIZE;
148 rq->cmd_len = COMMAND_SIZE(INQUIRY);
150 rq->sense = h->sense;
151 memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
152 rq->sense_len = h->senselen = 0;
154 err = blk_execute_rq(rq->q, NULL, rq, 1);
156 sdev_printk(KERN_INFO, sdev,
157 "%s: std inquiry failed with %x\n",
158 ALUA_DH_NAME, rq->errors);
159 h->senselen = rq->sense_len;
168 * submit_vpd_inquiry - Issue an INQUIRY VPD page 0x83 command
169 * @sdev: sdev the command should be sent to
171 static int submit_vpd_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
174 int err = SCSI_DH_RES_TEMP_UNAVAIL;
176 rq = get_alua_req(sdev, h->buff, h->bufflen, READ);
180 /* Prepare the command. */
181 rq->cmd[0] = INQUIRY;
184 rq->cmd[4] = h->bufflen;
185 rq->cmd_len = COMMAND_SIZE(INQUIRY);
187 rq->sense = h->sense;
188 memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
189 rq->sense_len = h->senselen = 0;
191 err = blk_execute_rq(rq->q, NULL, rq, 1);
193 sdev_printk(KERN_INFO, sdev,
194 "%s: evpd inquiry failed with %x\n",
195 ALUA_DH_NAME, rq->errors);
196 h->senselen = rq->sense_len;
205 * submit_rtpg - Issue a REPORT TARGET GROUP STATES command
206 * @sdev: sdev the command should be sent to
208 static unsigned submit_rtpg(struct scsi_device *sdev, struct alua_dh_data *h)
211 int err = SCSI_DH_RES_TEMP_UNAVAIL;
213 rq = get_alua_req(sdev, h->buff, h->bufflen, READ);
217 /* Prepare the command. */
218 rq->cmd[0] = MAINTENANCE_IN;
219 rq->cmd[1] = MI_REPORT_TARGET_PGS;
220 rq->cmd[6] = (h->bufflen >> 24) & 0xff;
221 rq->cmd[7] = (h->bufflen >> 16) & 0xff;
222 rq->cmd[8] = (h->bufflen >> 8) & 0xff;
223 rq->cmd[9] = h->bufflen & 0xff;
224 rq->cmd_len = COMMAND_SIZE(MAINTENANCE_IN);
226 rq->sense = h->sense;
227 memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
228 rq->sense_len = h->senselen = 0;
230 err = blk_execute_rq(rq->q, NULL, rq, 1);
232 sdev_printk(KERN_INFO, sdev,
233 "%s: rtpg failed with %x\n",
234 ALUA_DH_NAME, rq->errors);
235 h->senselen = rq->sense_len;
244 * alua_stpg - Evaluate SET TARGET GROUP STATES
245 * @sdev: the device to be evaluated
246 * @state: the new target group state
248 * Send a SET TARGET GROUP STATES command to the device.
249 * We only have to test here if we should resubmit the command;
250 * any other error is assumed as a failure.
252 static void stpg_endio(struct request *req, int error)
254 struct alua_dh_data *h = req->end_io_data;
255 struct scsi_sense_hdr sense_hdr;
256 unsigned err = SCSI_DH_OK;
258 if (error || host_byte(req->errors) != DID_OK ||
259 msg_byte(req->errors) != COMMAND_COMPLETE) {
264 if (h->senselen > 0) {
265 err = scsi_normalize_sense(h->sense, SCSI_SENSE_BUFFERSIZE,
271 err = alua_check_sense(h->sdev, &sense_hdr);
272 if (err == ADD_TO_MLQUEUE) {
276 sdev_printk(KERN_INFO, h->sdev,
277 "%s: stpg sense code: %02x/%02x/%02x\n",
278 ALUA_DH_NAME, sense_hdr.sense_key,
279 sense_hdr.asc, sense_hdr.ascq);
282 if (err == SCSI_DH_OK) {
283 h->state = TPGS_STATE_OPTIMIZED;
284 sdev_printk(KERN_INFO, h->sdev,
285 "%s: port group %02x switched to state %c\n",
286 ALUA_DH_NAME, h->group_id,
287 print_alua_state(h->state));
290 req->end_io_data = NULL;
291 __blk_put_request(req->q, req);
292 if (h->callback_fn) {
293 h->callback_fn(h->callback_data, err);
294 h->callback_fn = h->callback_data = NULL;
300 * submit_stpg - Issue a SET TARGET GROUP STATES command
302 * Currently we're only setting the current target port group state
303 * to 'active/optimized' and let the array firmware figure out
304 * the states of the remaining groups.
306 static unsigned submit_stpg(struct alua_dh_data *h)
310 struct scsi_device *sdev = h->sdev;
312 /* Prepare the data buffer */
313 memset(h->buff, 0, stpg_len);
314 h->buff[4] = TPGS_STATE_OPTIMIZED & 0x0f;
315 h->buff[6] = (h->group_id >> 8) & 0xff;
316 h->buff[7] = h->group_id & 0xff;
318 rq = get_alua_req(sdev, h->buff, stpg_len, WRITE);
320 return SCSI_DH_RES_TEMP_UNAVAIL;
322 /* Prepare the command. */
323 rq->cmd[0] = MAINTENANCE_OUT;
324 rq->cmd[1] = MO_SET_TARGET_PGS;
325 rq->cmd[6] = (stpg_len >> 24) & 0xff;
326 rq->cmd[7] = (stpg_len >> 16) & 0xff;
327 rq->cmd[8] = (stpg_len >> 8) & 0xff;
328 rq->cmd[9] = stpg_len & 0xff;
329 rq->cmd_len = COMMAND_SIZE(MAINTENANCE_OUT);
331 rq->sense = h->sense;
332 memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
333 rq->sense_len = h->senselen = 0;
336 blk_execute_rq_nowait(rq->q, NULL, rq, 1, stpg_endio);
341 * alua_std_inquiry - Evaluate standard INQUIRY command
342 * @sdev: device to be checked
344 * Just extract the TPGS setting to find out if ALUA
347 static int alua_std_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
351 err = submit_std_inquiry(sdev, h);
353 if (err != SCSI_DH_OK)
356 /* Check TPGS setting */
357 h->tpgs = (h->inq[5] >> 4) & 0x3;
359 case TPGS_MODE_EXPLICIT|TPGS_MODE_IMPLICIT:
360 sdev_printk(KERN_INFO, sdev,
361 "%s: supports implicit and explicit TPGS\n",
364 case TPGS_MODE_EXPLICIT:
365 sdev_printk(KERN_INFO, sdev, "%s: supports explicit TPGS\n",
368 case TPGS_MODE_IMPLICIT:
369 sdev_printk(KERN_INFO, sdev, "%s: supports implicit TPGS\n",
373 h->tpgs = TPGS_MODE_NONE;
374 sdev_printk(KERN_INFO, sdev, "%s: not supported\n",
376 err = SCSI_DH_DEV_UNSUPP;
384 * alua_vpd_inquiry - Evaluate INQUIRY vpd page 0x83
385 * @sdev: device to be checked
387 * Extract the relative target port and the target port group
388 * descriptor from the list of identificators.
390 static int alua_vpd_inquiry(struct scsi_device *sdev, struct alua_dh_data *h)
397 err = submit_vpd_inquiry(sdev, h);
399 if (err != SCSI_DH_OK)
402 /* Check if vpd page exceeds initial buffer */
403 len = (h->buff[2] << 8) + h->buff[3] + 4;
404 if (len > h->bufflen) {
405 /* Resubmit with the correct length */
406 if (realloc_buffer(h, len)) {
407 sdev_printk(KERN_WARNING, sdev,
408 "%s: kmalloc buffer failed\n",
410 /* Temporary failure, bypass */
411 return SCSI_DH_DEV_TEMP_BUSY;
417 * Now look for the correct descriptor.
420 while (d < h->buff + len) {
421 switch (d[1] & 0xf) {
423 /* Relative target port */
424 h->rel_port = (d[6] << 8) + d[7];
427 /* Target port group */
428 h->group_id = (d[6] << 8) + d[7];
436 if (h->group_id == -1) {
438 * Internal error; TPGS supported but required
439 * VPD identification descriptors not present.
440 * Disable ALUA support
442 sdev_printk(KERN_INFO, sdev,
443 "%s: No target port descriptors found\n",
445 h->state = TPGS_STATE_OPTIMIZED;
446 h->tpgs = TPGS_MODE_NONE;
447 err = SCSI_DH_DEV_UNSUPP;
449 sdev_printk(KERN_INFO, sdev,
450 "%s: port group %02x rel port %02x\n",
451 ALUA_DH_NAME, h->group_id, h->rel_port);
457 static char print_alua_state(int state)
460 case TPGS_STATE_OPTIMIZED:
462 case TPGS_STATE_NONOPTIMIZED:
464 case TPGS_STATE_STANDBY:
466 case TPGS_STATE_UNAVAILABLE:
468 case TPGS_STATE_LBA_DEPENDENT:
470 case TPGS_STATE_OFFLINE:
472 case TPGS_STATE_TRANSITIONING:
479 static int alua_check_sense(struct scsi_device *sdev,
480 struct scsi_sense_hdr *sense_hdr)
482 switch (sense_hdr->sense_key) {
484 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a)
486 * LUN Not Accessible - ALUA state transition
488 return ADD_TO_MLQUEUE;
489 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0b)
491 * LUN Not Accessible -- Target port in standby state
494 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0c)
496 * LUN Not Accessible -- Target port in unavailable state
499 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x12)
501 * LUN Not Ready -- Offline
506 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x00)
508 * Power On, Reset, or Bus Device Reset, just retry.
510 return ADD_TO_MLQUEUE;
511 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x06) {
515 return ADD_TO_MLQUEUE;
517 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x07) {
519 * Implicit ALUA state transition failed
521 return ADD_TO_MLQUEUE;
523 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x0e) {
525 * REPORTED_LUNS_DATA_HAS_CHANGED is reported
526 * when switching controllers on targets like
527 * Intel Multi-Flex. We can just retry.
529 return ADD_TO_MLQUEUE;
535 return SCSI_RETURN_NOT_HANDLED;
539 * alua_rtpg - Evaluate REPORT TARGET GROUP STATES
540 * @sdev: the device to be evaluated.
542 * Evaluate the Target Port Group State.
543 * Returns SCSI_DH_DEV_OFFLINED if the path is
544 * found to be unusable.
546 static int alua_rtpg(struct scsi_device *sdev, struct alua_dh_data *h)
548 struct scsi_sense_hdr sense_hdr;
549 int len, k, off, valid_states = 0;
552 unsigned long expiry, interval = 10;
554 expiry = round_jiffies_up(jiffies + ALUA_FAILOVER_TIMEOUT);
556 err = submit_rtpg(sdev, h);
558 if (err == SCSI_DH_IO && h->senselen > 0) {
559 err = scsi_normalize_sense(h->sense, SCSI_SENSE_BUFFERSIZE,
564 err = alua_check_sense(sdev, &sense_hdr);
565 if (err == ADD_TO_MLQUEUE && time_before(jiffies, expiry))
567 sdev_printk(KERN_INFO, sdev,
568 "%s: rtpg sense code %02x/%02x/%02x\n",
569 ALUA_DH_NAME, sense_hdr.sense_key,
570 sense_hdr.asc, sense_hdr.ascq);
573 if (err != SCSI_DH_OK)
576 len = (h->buff[0] << 24) + (h->buff[1] << 16) +
577 (h->buff[2] << 8) + h->buff[3] + 4;
579 if (len > h->bufflen) {
580 /* Resubmit with the correct length */
581 if (realloc_buffer(h, len)) {
582 sdev_printk(KERN_WARNING, sdev,
583 "%s: kmalloc buffer failed\n",__func__);
584 /* Temporary failure, bypass */
585 return SCSI_DH_DEV_TEMP_BUSY;
590 for (k = 4, ucp = h->buff + 4; k < len; k += off, ucp += off) {
591 if (h->group_id == (ucp[2] << 8) + ucp[3]) {
592 h->state = ucp[0] & 0x0f;
593 valid_states = ucp[1];
595 off = 8 + (ucp[7] * 4);
598 sdev_printk(KERN_INFO, sdev,
599 "%s: port group %02x state %c supports %c%c%c%c%c%c%c\n",
600 ALUA_DH_NAME, h->group_id, print_alua_state(h->state),
601 valid_states&TPGS_SUPPORT_TRANSITION?'T':'t',
602 valid_states&TPGS_SUPPORT_OFFLINE?'O':'o',
603 valid_states&TPGS_SUPPORT_LBA_DEPENDENT?'L':'l',
604 valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u',
605 valid_states&TPGS_SUPPORT_STANDBY?'S':'s',
606 valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n',
607 valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a');
610 case TPGS_STATE_TRANSITIONING:
611 if (time_before(jiffies, expiry)) {
612 /* State transition, retry */
617 /* Transitioning time exceeded, set port to standby */
619 h->state = TPGS_STATE_STANDBY;
621 case TPGS_STATE_OFFLINE:
622 case TPGS_STATE_UNAVAILABLE:
623 /* Path unusable for unavailable/offline */
624 err = SCSI_DH_DEV_OFFLINED;
627 /* Useable path if active */
635 * alua_initialize - Initialize ALUA state
636 * @sdev: the device to be initialized
638 * For the prep_fn to work correctly we have
639 * to initialize the ALUA state for the device.
641 static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h)
645 err = alua_std_inquiry(sdev, h);
646 if (err != SCSI_DH_OK)
649 err = alua_vpd_inquiry(sdev, h);
650 if (err != SCSI_DH_OK)
653 err = alua_rtpg(sdev, h);
654 if (err != SCSI_DH_OK)
662 * alua_activate - activate a path
663 * @sdev: device on the path to be activated
665 * We're currently switching the port group to be activated only and
666 * let the array figure out the rest.
667 * There may be other arrays which require us to switch all port groups
668 * based on a certain policy. But until we actually encounter them it
671 static int alua_activate(struct scsi_device *sdev,
672 activate_complete fn, void *data)
674 struct alua_dh_data *h = get_alua_data(sdev);
675 int err = SCSI_DH_OK;
677 if (h->group_id != -1) {
678 err = alua_rtpg(sdev, h);
679 if (err != SCSI_DH_OK)
683 if (h->tpgs & TPGS_MODE_EXPLICIT &&
684 h->state != TPGS_STATE_OPTIMIZED &&
685 h->state != TPGS_STATE_LBA_DEPENDENT) {
687 h->callback_data = data;
688 err = submit_stpg(h);
689 if (err == SCSI_DH_OK)
691 h->callback_fn = h->callback_data = NULL;
701 * alua_prep_fn - request callback
703 * Fail I/O to all paths not in state
704 * active/optimized or active/non-optimized.
706 static int alua_prep_fn(struct scsi_device *sdev, struct request *req)
708 struct alua_dh_data *h = get_alua_data(sdev);
709 int ret = BLKPREP_OK;
711 if (h->state == TPGS_STATE_TRANSITIONING)
713 else if (h->state != TPGS_STATE_OPTIMIZED &&
714 h->state != TPGS_STATE_NONOPTIMIZED &&
715 h->state != TPGS_STATE_LBA_DEPENDENT) {
717 req->cmd_flags |= REQ_QUIET;
723 static const struct scsi_dh_devlist alua_dev_list[] = {
724 {"HP", "MSA VOLUME" },
732 {"Pillar", "Axiom" },
733 {"Intel", "Multi-Flex"},
735 {"NETAPP", "LUN C-Mode"},
737 {"Promise", "VTrak"},
741 static int alua_bus_attach(struct scsi_device *sdev);
742 static void alua_bus_detach(struct scsi_device *sdev);
744 static struct scsi_device_handler alua_dh = {
745 .name = ALUA_DH_NAME,
746 .module = THIS_MODULE,
747 .devlist = alua_dev_list,
748 .attach = alua_bus_attach,
749 .detach = alua_bus_detach,
750 .prep_fn = alua_prep_fn,
751 .check_sense = alua_check_sense,
752 .activate = alua_activate,
756 * alua_bus_attach - Attach device handler
757 * @sdev: device to be attached to
759 static int alua_bus_attach(struct scsi_device *sdev)
761 struct scsi_dh_data *scsi_dh_data;
762 struct alua_dh_data *h;
764 int err = SCSI_DH_OK;
766 scsi_dh_data = kzalloc(sizeof(*scsi_dh_data)
767 + sizeof(*h) , GFP_KERNEL);
769 sdev_printk(KERN_ERR, sdev, "%s: Attach failed\n",
774 scsi_dh_data->scsi_dh = &alua_dh;
775 h = (struct alua_dh_data *) scsi_dh_data->buf;
776 h->tpgs = TPGS_MODE_UNINITIALIZED;
777 h->state = TPGS_STATE_OPTIMIZED;
781 h->bufflen = ALUA_INQUIRY_SIZE;
784 err = alua_initialize(sdev, h);
785 if (err != SCSI_DH_OK)
788 if (!try_module_get(THIS_MODULE))
791 spin_lock_irqsave(sdev->request_queue->queue_lock, flags);
792 sdev->scsi_dh_data = scsi_dh_data;
793 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
799 sdev_printk(KERN_ERR, sdev, "%s: not attached\n", ALUA_DH_NAME);
804 * alua_bus_detach - Detach device handler
805 * @sdev: device to be detached from
807 static void alua_bus_detach(struct scsi_device *sdev)
809 struct scsi_dh_data *scsi_dh_data;
810 struct alua_dh_data *h;
813 spin_lock_irqsave(sdev->request_queue->queue_lock, flags);
814 scsi_dh_data = sdev->scsi_dh_data;
815 sdev->scsi_dh_data = NULL;
816 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
818 h = (struct alua_dh_data *) scsi_dh_data->buf;
819 if (h->buff && h->inq != h->buff)
822 module_put(THIS_MODULE);
823 sdev_printk(KERN_NOTICE, sdev, "%s: Detached\n", ALUA_DH_NAME);
826 static int __init alua_init(void)
830 r = scsi_register_device_handler(&alua_dh);
832 printk(KERN_ERR "%s: Failed to register scsi device handler",
837 static void __exit alua_exit(void)
839 scsi_unregister_device_handler(&alua_dh);
842 module_init(alua_init);
843 module_exit(alua_exit);
845 MODULE_DESCRIPTION("DM Multipath ALUA support");
846 MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
847 MODULE_LICENSE("GPL");
848 MODULE_VERSION(ALUA_DH_VER);