Linux 4.19.133
[linux/fpc-iii.git] / drivers / scsi / device_handler / scsi_dh_alua.c
blobc95c782b93a53b54e6fa5b292426f85cd64c4e30
1 /*
2 * Generic SCSI-3 ALUA SCSI Device Handler
4 * Copyright (C) 2007-2010 Hannes Reinecke, SUSE Linux Products GmbH.
5 * All rights reserved.
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
43 #define TPGS_SUPPORT_ALL 0xdf
45 #define RTPG_FMT_MASK 0x70
46 #define RTPG_FMT_EXT_HDR 0x10
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_RTPG_SIZE 128
54 #define ALUA_FAILOVER_TIMEOUT 60
55 #define ALUA_FAILOVER_RETRIES 5
56 #define ALUA_RTPG_DELAY_MSECS 5
57 #define ALUA_RTPG_RETRY_DELAY 2
59 /* device handler flags */
60 #define ALUA_OPTIMIZE_STPG 0x01
61 #define ALUA_RTPG_EXT_HDR_UNSUPP 0x02
62 /* State machine flags */
63 #define ALUA_PG_RUN_RTPG 0x10
64 #define ALUA_PG_RUN_STPG 0x20
65 #define ALUA_PG_RUNNING 0x40
67 static uint optimize_stpg;
68 module_param(optimize_stpg, uint, S_IRUGO|S_IWUSR);
69 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.");
71 static LIST_HEAD(port_group_list);
72 static DEFINE_SPINLOCK(port_group_lock);
73 static struct workqueue_struct *kaluad_wq;
75 struct alua_port_group {
76 struct kref kref;
77 struct rcu_head rcu;
78 struct list_head node;
79 struct list_head dh_list;
80 unsigned char device_id_str[256];
81 int device_id_len;
82 int group_id;
83 int tpgs;
84 int state;
85 int pref;
86 int valid_states;
87 unsigned flags; /* used for optimizing STPG */
88 unsigned char transition_tmo;
89 unsigned long expiry;
90 unsigned long interval;
91 struct delayed_work rtpg_work;
92 spinlock_t lock;
93 struct list_head rtpg_list;
94 struct scsi_device *rtpg_sdev;
97 struct alua_dh_data {
98 struct list_head node;
99 struct alua_port_group __rcu *pg;
100 int group_id;
101 spinlock_t pg_lock;
102 struct scsi_device *sdev;
103 int init_error;
104 struct mutex init_mutex;
107 struct alua_queue_data {
108 struct list_head entry;
109 activate_complete callback_fn;
110 void *callback_data;
113 #define ALUA_POLICY_SWITCH_CURRENT 0
114 #define ALUA_POLICY_SWITCH_ALL 1
116 static void alua_rtpg_work(struct work_struct *work);
117 static bool alua_rtpg_queue(struct alua_port_group *pg,
118 struct scsi_device *sdev,
119 struct alua_queue_data *qdata, bool force);
120 static void alua_check(struct scsi_device *sdev, bool force);
122 static void release_port_group(struct kref *kref)
124 struct alua_port_group *pg;
126 pg = container_of(kref, struct alua_port_group, kref);
127 if (pg->rtpg_sdev)
128 flush_delayed_work(&pg->rtpg_work);
129 spin_lock(&port_group_lock);
130 list_del(&pg->node);
131 spin_unlock(&port_group_lock);
132 kfree_rcu(pg, rcu);
136 * submit_rtpg - Issue a REPORT TARGET GROUP STATES command
137 * @sdev: sdev the command should be sent to
139 static int submit_rtpg(struct scsi_device *sdev, unsigned char *buff,
140 int bufflen, struct scsi_sense_hdr *sshdr, int flags)
142 u8 cdb[MAX_COMMAND_SIZE];
143 int req_flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
144 REQ_FAILFAST_DRIVER;
146 /* Prepare the command. */
147 memset(cdb, 0x0, MAX_COMMAND_SIZE);
148 cdb[0] = MAINTENANCE_IN;
149 if (!(flags & ALUA_RTPG_EXT_HDR_UNSUPP))
150 cdb[1] = MI_REPORT_TARGET_PGS | MI_EXT_HDR_PARAM_FMT;
151 else
152 cdb[1] = MI_REPORT_TARGET_PGS;
153 put_unaligned_be32(bufflen, &cdb[6]);
155 return scsi_execute(sdev, cdb, DMA_FROM_DEVICE, buff, bufflen, NULL,
156 sshdr, ALUA_FAILOVER_TIMEOUT * HZ,
157 ALUA_FAILOVER_RETRIES, req_flags, 0, NULL);
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[MAX_COMMAND_SIZE];
171 unsigned char stpg_data[8];
172 int stpg_len = 8;
173 int req_flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
174 REQ_FAILFAST_DRIVER;
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, MAX_COMMAND_SIZE);
183 cdb[0] = MAINTENANCE_OUT;
184 cdb[1] = MO_SET_TARGET_PGS;
185 put_unaligned_be32(stpg_len, &cdb[6]);
187 return scsi_execute(sdev, cdb, DMA_TO_DEVICE, stpg_data, stpg_len, NULL,
188 sshdr, ALUA_FAILOVER_TIMEOUT * HZ,
189 ALUA_FAILOVER_RETRIES, req_flags, 0, NULL);
192 static struct alua_port_group *alua_find_get_pg(char *id_str, size_t id_size,
193 int group_id)
195 struct alua_port_group *pg;
197 if (!id_str || !id_size || !strlen(id_str))
198 return NULL;
200 list_for_each_entry(pg, &port_group_list, node) {
201 if (pg->group_id != group_id)
202 continue;
203 if (!pg->device_id_len || pg->device_id_len != id_size)
204 continue;
205 if (strncmp(pg->device_id_str, id_str, id_size))
206 continue;
207 if (!kref_get_unless_zero(&pg->kref))
208 continue;
209 return pg;
212 return NULL;
216 * alua_alloc_pg - Allocate a new port_group structure
217 * @sdev: scsi device
218 * @group_id: port group id
219 * @tpgs: target port group settings
221 * Allocate a new port_group structure for a given
222 * device.
224 static struct alua_port_group *alua_alloc_pg(struct scsi_device *sdev,
225 int group_id, int tpgs)
227 struct alua_port_group *pg, *tmp_pg;
229 pg = kzalloc(sizeof(struct alua_port_group), GFP_KERNEL);
230 if (!pg)
231 return ERR_PTR(-ENOMEM);
233 pg->device_id_len = scsi_vpd_lun_id(sdev, pg->device_id_str,
234 sizeof(pg->device_id_str));
235 if (pg->device_id_len <= 0) {
237 * TPGS supported but no device identification found.
238 * Generate private device identification.
240 sdev_printk(KERN_INFO, sdev,
241 "%s: No device descriptors found\n",
242 ALUA_DH_NAME);
243 pg->device_id_str[0] = '\0';
244 pg->device_id_len = 0;
246 pg->group_id = group_id;
247 pg->tpgs = tpgs;
248 pg->state = SCSI_ACCESS_STATE_OPTIMAL;
249 pg->valid_states = TPGS_SUPPORT_ALL;
250 if (optimize_stpg)
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,
261 group_id);
262 if (tmp_pg) {
263 spin_unlock(&port_group_lock);
264 kfree(pg);
265 return tmp_pg;
268 list_add(&pg->node, &port_group_list);
269 spin_unlock(&port_group_lock);
271 return pg;
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
279 * is supported.
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",
292 ALUA_DH_NAME);
293 return tpgs;
296 tpgs = scsi_device_tpgs(sdev);
297 switch (tpgs) {
298 case TPGS_MODE_EXPLICIT|TPGS_MODE_IMPLICIT:
299 sdev_printk(KERN_INFO, sdev,
300 "%s: supports implicit and explicit TPGS\n",
301 ALUA_DH_NAME);
302 break;
303 case TPGS_MODE_EXPLICIT:
304 sdev_printk(KERN_INFO, sdev, "%s: supports explicit TPGS\n",
305 ALUA_DH_NAME);
306 break;
307 case TPGS_MODE_IMPLICIT:
308 sdev_printk(KERN_INFO, sdev, "%s: supports implicit TPGS\n",
309 ALUA_DH_NAME);
310 break;
311 case TPGS_MODE_NONE:
312 sdev_printk(KERN_INFO, sdev, "%s: not supported\n",
313 ALUA_DH_NAME);
314 break;
315 default:
316 sdev_printk(KERN_INFO, sdev,
317 "%s: unsupported TPGS setting %d\n",
318 ALUA_DH_NAME, tpgs);
319 tpgs = TPGS_MODE_NONE;
320 break;
323 return tpgs;
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,
334 int tpgs)
336 int rel_port = -1, group_id;
337 struct alua_port_group *pg, *old_pg = NULL;
338 bool pg_updated = false;
339 unsigned long flags;
341 group_id = scsi_vpd_tpg_id(sdev, &rel_port);
342 if (group_id < 0) {
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",
350 ALUA_DH_NAME);
351 return SCSI_DH_DEV_UNSUPP;
354 pg = alua_alloc_pg(sdev, group_id, tpgs);
355 if (IS_ERR(pg)) {
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,
364 group_id, rel_port);
365 else
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);
372 old_pg = rcu_dereference_protected(h->pg, lockdep_is_held(&h->pg_lock));
373 if (old_pg != pg) {
374 /* port group has changed. Update to new port group */
375 if (h->pg) {
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);
381 pg_updated = true;
384 spin_lock_irqsave(&pg->lock, flags);
385 if (pg_updated)
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)),
391 sdev, NULL, true);
392 spin_unlock(&h->pg_lock);
394 if (old_pg)
395 kref_put(&old_pg->kref, release_port_group);
397 return SCSI_DH_OK;
400 static char print_alua_state(unsigned char state)
402 switch (state) {
403 case SCSI_ACCESS_STATE_OPTIMAL:
404 return 'A';
405 case SCSI_ACCESS_STATE_ACTIVE:
406 return 'N';
407 case SCSI_ACCESS_STATE_STANDBY:
408 return 'S';
409 case SCSI_ACCESS_STATE_UNAVAILABLE:
410 return 'U';
411 case SCSI_ACCESS_STATE_LBA:
412 return 'L';
413 case SCSI_ACCESS_STATE_OFFLINE:
414 return 'O';
415 case SCSI_ACCESS_STATE_TRANSITIONING:
416 return 'T';
417 default:
418 return 'X';
422 static int alua_check_sense(struct scsi_device *sdev,
423 struct scsi_sense_hdr *sense_hdr)
425 switch (sense_hdr->sense_key) {
426 case NOT_READY:
427 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a) {
429 * LUN Not Accessible - ALUA state transition
431 alua_check(sdev, false);
432 return NEEDS_RETRY;
434 break;
435 case UNIT_ATTENTION:
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) {
457 * ALUA state changed
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;
481 break;
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;
498 int retval;
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;
505 else if (retval)
506 return SCSI_DH_IO;
507 else
508 return SCSI_DH_OK;
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, 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;
528 unsigned long flags;
529 bool transitioning_sense = false;
531 if (!pg->expiry) {
532 unsigned long transition_tmo = ALUA_FAILOVER_TIMEOUT * HZ;
534 if (pg->transition_tmo)
535 transition_tmo = pg->transition_tmo * HZ;
537 pg->expiry = round_jiffies_up(jiffies + transition_tmo);
540 buff = kzalloc(bufflen, GFP_KERNEL);
541 if (!buff)
542 return SCSI_DH_DEV_TEMP_BUSY;
544 retry:
545 err = 0;
546 retval = submit_rtpg(sdev, buff, bufflen, &sense_hdr, pg->flags);
548 if (retval) {
550 * Some (broken) implementations have a habit of returning
551 * an error during things like firmware update etc.
552 * But if the target only supports active/optimized there's
553 * not much we can do; it's not that we can switch paths
554 * or anything.
555 * So ignore any errors to avoid spurious failures during
556 * path failover.
558 if ((pg->valid_states & ~TPGS_SUPPORT_OPTIMIZED) == 0) {
559 sdev_printk(KERN_INFO, sdev,
560 "%s: ignoring rtpg result %d\n",
561 ALUA_DH_NAME, retval);
562 kfree(buff);
563 return SCSI_DH_OK;
565 if (!scsi_sense_valid(&sense_hdr)) {
566 sdev_printk(KERN_INFO, sdev,
567 "%s: rtpg failed, result %d\n",
568 ALUA_DH_NAME, retval);
569 kfree(buff);
570 if (driver_byte(retval) == DRIVER_ERROR)
571 return SCSI_DH_DEV_TEMP_BUSY;
572 return SCSI_DH_IO;
576 * submit_rtpg() has failed on existing arrays
577 * when requesting extended header info, and
578 * the array doesn't support extended headers,
579 * even though it shouldn't according to T10.
580 * The retry without rtpg_ext_hdr_req set
581 * handles this.
583 if (!(pg->flags & ALUA_RTPG_EXT_HDR_UNSUPP) &&
584 sense_hdr.sense_key == ILLEGAL_REQUEST &&
585 sense_hdr.asc == 0x24 && sense_hdr.ascq == 0) {
586 pg->flags |= ALUA_RTPG_EXT_HDR_UNSUPP;
587 goto retry;
590 * If the array returns with 'ALUA state transition'
591 * sense code here it cannot return RTPG data during
592 * transition. So set the state to 'transitioning' directly.
594 if (sense_hdr.sense_key == NOT_READY &&
595 sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a) {
596 transitioning_sense = true;
597 goto skip_rtpg;
600 * Retry on any other UNIT ATTENTION occurred.
602 if (sense_hdr.sense_key == UNIT_ATTENTION)
603 err = SCSI_DH_RETRY;
604 if (err == SCSI_DH_RETRY &&
605 pg->expiry != 0 && time_before(jiffies, pg->expiry)) {
606 sdev_printk(KERN_ERR, sdev, "%s: rtpg retry\n",
607 ALUA_DH_NAME);
608 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
609 kfree(buff);
610 return err;
612 sdev_printk(KERN_ERR, sdev, "%s: rtpg failed\n",
613 ALUA_DH_NAME);
614 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
615 kfree(buff);
616 pg->expiry = 0;
617 return SCSI_DH_IO;
620 len = get_unaligned_be32(&buff[0]) + 4;
622 if (len > bufflen) {
623 /* Resubmit with the correct length */
624 kfree(buff);
625 bufflen = len;
626 buff = kmalloc(bufflen, GFP_KERNEL);
627 if (!buff) {
628 sdev_printk(KERN_WARNING, sdev,
629 "%s: kmalloc buffer failed\n",__func__);
630 /* Temporary failure, bypass */
631 pg->expiry = 0;
632 return SCSI_DH_DEV_TEMP_BUSY;
634 goto retry;
637 orig_transition_tmo = pg->transition_tmo;
638 if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR && buff[5] != 0)
639 pg->transition_tmo = buff[5];
640 else
641 pg->transition_tmo = ALUA_FAILOVER_TIMEOUT;
643 if (orig_transition_tmo != pg->transition_tmo) {
644 sdev_printk(KERN_INFO, sdev,
645 "%s: transition timeout set to %d seconds\n",
646 ALUA_DH_NAME, pg->transition_tmo);
647 pg->expiry = jiffies + pg->transition_tmo * HZ;
650 if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR)
651 tpg_desc_tbl_off = 8;
652 else
653 tpg_desc_tbl_off = 4;
655 for (k = tpg_desc_tbl_off, desc = buff + tpg_desc_tbl_off;
656 k < len;
657 k += off, desc += off) {
658 u16 group_id = get_unaligned_be16(&desc[2]);
660 spin_lock_irqsave(&port_group_lock, flags);
661 tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len,
662 group_id);
663 spin_unlock_irqrestore(&port_group_lock, flags);
664 if (tmp_pg) {
665 if (spin_trylock_irqsave(&tmp_pg->lock, flags)) {
666 if ((tmp_pg == pg) ||
667 !(tmp_pg->flags & ALUA_PG_RUNNING)) {
668 struct alua_dh_data *h;
670 tmp_pg->state = desc[0] & 0x0f;
671 tmp_pg->pref = desc[0] >> 7;
672 rcu_read_lock();
673 list_for_each_entry_rcu(h,
674 &tmp_pg->dh_list, node) {
675 /* h->sdev should always be valid */
676 BUG_ON(!h->sdev);
677 h->sdev->access_state = desc[0];
679 rcu_read_unlock();
681 if (tmp_pg == pg)
682 tmp_pg->valid_states = desc[1];
683 spin_unlock_irqrestore(&tmp_pg->lock, flags);
685 kref_put(&tmp_pg->kref, release_port_group);
687 off = 8 + (desc[7] * 4);
690 skip_rtpg:
691 spin_lock_irqsave(&pg->lock, flags);
692 if (transitioning_sense)
693 pg->state = SCSI_ACCESS_STATE_TRANSITIONING;
695 sdev_printk(KERN_INFO, sdev,
696 "%s: port group %02x state %c %s supports %c%c%c%c%c%c%c\n",
697 ALUA_DH_NAME, pg->group_id, print_alua_state(pg->state),
698 pg->pref ? "preferred" : "non-preferred",
699 pg->valid_states&TPGS_SUPPORT_TRANSITION?'T':'t',
700 pg->valid_states&TPGS_SUPPORT_OFFLINE?'O':'o',
701 pg->valid_states&TPGS_SUPPORT_LBA_DEPENDENT?'L':'l',
702 pg->valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u',
703 pg->valid_states&TPGS_SUPPORT_STANDBY?'S':'s',
704 pg->valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n',
705 pg->valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a');
707 switch (pg->state) {
708 case SCSI_ACCESS_STATE_TRANSITIONING:
709 if (time_before(jiffies, pg->expiry)) {
710 /* State transition, retry */
711 pg->interval = ALUA_RTPG_RETRY_DELAY;
712 err = SCSI_DH_RETRY;
713 } else {
714 struct alua_dh_data *h;
716 /* Transitioning time exceeded, set port to standby */
717 err = SCSI_DH_IO;
718 pg->state = SCSI_ACCESS_STATE_STANDBY;
719 pg->expiry = 0;
720 rcu_read_lock();
721 list_for_each_entry_rcu(h, &pg->dh_list, node) {
722 BUG_ON(!h->sdev);
723 h->sdev->access_state =
724 (pg->state & SCSI_ACCESS_STATE_MASK);
725 if (pg->pref)
726 h->sdev->access_state |=
727 SCSI_ACCESS_STATE_PREFERRED;
729 rcu_read_unlock();
731 break;
732 case SCSI_ACCESS_STATE_OFFLINE:
733 /* Path unusable */
734 err = SCSI_DH_DEV_OFFLINED;
735 pg->expiry = 0;
736 break;
737 default:
738 /* Useable path if active */
739 err = SCSI_DH_OK;
740 pg->expiry = 0;
741 break;
743 spin_unlock_irqrestore(&pg->lock, flags);
744 kfree(buff);
745 return err;
749 * alua_stpg - Issue a SET TARGET PORT GROUP command
751 * Issue a SET TARGET PORT GROUP command and evaluate the
752 * response. Returns SCSI_DH_RETRY per default to trigger
753 * a re-evaluation of the target group state or SCSI_DH_OK
754 * if no further action needs to be taken.
756 static unsigned alua_stpg(struct scsi_device *sdev, struct alua_port_group *pg)
758 int retval;
759 struct scsi_sense_hdr sense_hdr;
761 if (!(pg->tpgs & TPGS_MODE_EXPLICIT)) {
762 /* Only implicit ALUA supported, retry */
763 return SCSI_DH_RETRY;
765 switch (pg->state) {
766 case SCSI_ACCESS_STATE_OPTIMAL:
767 return SCSI_DH_OK;
768 case SCSI_ACCESS_STATE_ACTIVE:
769 if ((pg->flags & ALUA_OPTIMIZE_STPG) &&
770 !pg->pref &&
771 (pg->tpgs & TPGS_MODE_IMPLICIT))
772 return SCSI_DH_OK;
773 break;
774 case SCSI_ACCESS_STATE_STANDBY:
775 case SCSI_ACCESS_STATE_UNAVAILABLE:
776 break;
777 case SCSI_ACCESS_STATE_OFFLINE:
778 return SCSI_DH_IO;
779 case SCSI_ACCESS_STATE_TRANSITIONING:
780 break;
781 default:
782 sdev_printk(KERN_INFO, sdev,
783 "%s: stpg failed, unhandled TPGS state %d",
784 ALUA_DH_NAME, pg->state);
785 return SCSI_DH_NOSYS;
787 retval = submit_stpg(sdev, pg->group_id, &sense_hdr);
789 if (retval) {
790 if (!scsi_sense_valid(&sense_hdr)) {
791 sdev_printk(KERN_INFO, sdev,
792 "%s: stpg failed, result %d",
793 ALUA_DH_NAME, retval);
794 if (driver_byte(retval) == DRIVER_ERROR)
795 return SCSI_DH_DEV_TEMP_BUSY;
796 } else {
797 sdev_printk(KERN_INFO, sdev, "%s: stpg failed\n",
798 ALUA_DH_NAME);
799 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
802 /* Retry RTPG */
803 return SCSI_DH_RETRY;
806 static void alua_rtpg_work(struct work_struct *work)
808 struct alua_port_group *pg =
809 container_of(work, struct alua_port_group, rtpg_work.work);
810 struct scsi_device *sdev;
811 LIST_HEAD(qdata_list);
812 int err = SCSI_DH_OK;
813 struct alua_queue_data *qdata, *tmp;
814 unsigned long flags;
816 spin_lock_irqsave(&pg->lock, flags);
817 sdev = pg->rtpg_sdev;
818 if (!sdev) {
819 WARN_ON(pg->flags & ALUA_PG_RUN_RTPG);
820 WARN_ON(pg->flags & ALUA_PG_RUN_STPG);
821 spin_unlock_irqrestore(&pg->lock, flags);
822 kref_put(&pg->kref, release_port_group);
823 return;
825 pg->flags |= ALUA_PG_RUNNING;
826 if (pg->flags & ALUA_PG_RUN_RTPG) {
827 int state = pg->state;
829 pg->flags &= ~ALUA_PG_RUN_RTPG;
830 spin_unlock_irqrestore(&pg->lock, flags);
831 if (state == SCSI_ACCESS_STATE_TRANSITIONING) {
832 if (alua_tur(sdev) == SCSI_DH_RETRY) {
833 spin_lock_irqsave(&pg->lock, flags);
834 pg->flags &= ~ALUA_PG_RUNNING;
835 pg->flags |= ALUA_PG_RUN_RTPG;
836 if (!pg->interval)
837 pg->interval = ALUA_RTPG_RETRY_DELAY;
838 spin_unlock_irqrestore(&pg->lock, flags);
839 queue_delayed_work(kaluad_wq, &pg->rtpg_work,
840 pg->interval * HZ);
841 return;
843 /* Send RTPG on failure or if TUR indicates SUCCESS */
845 err = alua_rtpg(sdev, pg);
846 spin_lock_irqsave(&pg->lock, flags);
847 if (err == SCSI_DH_RETRY || pg->flags & ALUA_PG_RUN_RTPG) {
848 pg->flags &= ~ALUA_PG_RUNNING;
849 if (!pg->interval && !(pg->flags & ALUA_PG_RUN_RTPG))
850 pg->interval = ALUA_RTPG_RETRY_DELAY;
851 pg->flags |= ALUA_PG_RUN_RTPG;
852 spin_unlock_irqrestore(&pg->lock, flags);
853 queue_delayed_work(kaluad_wq, &pg->rtpg_work,
854 pg->interval * HZ);
855 return;
857 if (err != SCSI_DH_OK)
858 pg->flags &= ~ALUA_PG_RUN_STPG;
860 if (pg->flags & ALUA_PG_RUN_STPG) {
861 pg->flags &= ~ALUA_PG_RUN_STPG;
862 spin_unlock_irqrestore(&pg->lock, flags);
863 err = alua_stpg(sdev, pg);
864 spin_lock_irqsave(&pg->lock, flags);
865 if (err == SCSI_DH_RETRY || pg->flags & ALUA_PG_RUN_RTPG) {
866 pg->flags |= ALUA_PG_RUN_RTPG;
867 pg->interval = 0;
868 pg->flags &= ~ALUA_PG_RUNNING;
869 spin_unlock_irqrestore(&pg->lock, flags);
870 queue_delayed_work(kaluad_wq, &pg->rtpg_work,
871 pg->interval * HZ);
872 return;
876 list_splice_init(&pg->rtpg_list, &qdata_list);
877 pg->rtpg_sdev = NULL;
878 spin_unlock_irqrestore(&pg->lock, flags);
880 list_for_each_entry_safe(qdata, tmp, &qdata_list, entry) {
881 list_del(&qdata->entry);
882 if (qdata->callback_fn)
883 qdata->callback_fn(qdata->callback_data, err);
884 kfree(qdata);
886 spin_lock_irqsave(&pg->lock, flags);
887 pg->flags &= ~ALUA_PG_RUNNING;
888 spin_unlock_irqrestore(&pg->lock, flags);
889 scsi_device_put(sdev);
890 kref_put(&pg->kref, release_port_group);
894 * alua_rtpg_queue() - cause RTPG to be submitted asynchronously
895 * @pg: ALUA port group associated with @sdev.
896 * @sdev: SCSI device for which to submit an RTPG.
897 * @qdata: Information about the callback to invoke after the RTPG.
898 * @force: Whether or not to submit an RTPG if a work item that will submit an
899 * RTPG already has been scheduled.
901 * Returns true if and only if alua_rtpg_work() will be called asynchronously.
902 * That function is responsible for calling @qdata->fn().
904 static bool alua_rtpg_queue(struct alua_port_group *pg,
905 struct scsi_device *sdev,
906 struct alua_queue_data *qdata, bool force)
908 int start_queue = 0;
909 unsigned long flags;
910 if (WARN_ON_ONCE(!pg) || scsi_device_get(sdev))
911 return false;
913 spin_lock_irqsave(&pg->lock, flags);
914 if (qdata) {
915 list_add_tail(&qdata->entry, &pg->rtpg_list);
916 pg->flags |= ALUA_PG_RUN_STPG;
917 force = true;
919 if (pg->rtpg_sdev == NULL) {
920 pg->interval = 0;
921 pg->flags |= ALUA_PG_RUN_RTPG;
922 kref_get(&pg->kref);
923 pg->rtpg_sdev = sdev;
924 start_queue = 1;
925 } else if (!(pg->flags & ALUA_PG_RUN_RTPG) && force) {
926 pg->flags |= ALUA_PG_RUN_RTPG;
927 /* Do not queue if the worker is already running */
928 if (!(pg->flags & ALUA_PG_RUNNING)) {
929 kref_get(&pg->kref);
930 start_queue = 1;
934 spin_unlock_irqrestore(&pg->lock, flags);
936 if (start_queue) {
937 if (queue_delayed_work(kaluad_wq, &pg->rtpg_work,
938 msecs_to_jiffies(ALUA_RTPG_DELAY_MSECS)))
939 sdev = NULL;
940 else
941 kref_put(&pg->kref, release_port_group);
943 if (sdev)
944 scsi_device_put(sdev);
946 return true;
950 * alua_initialize - Initialize ALUA state
951 * @sdev: the device to be initialized
953 * For the prep_fn to work correctly we have
954 * to initialize the ALUA state for the device.
956 static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h)
958 int err = SCSI_DH_DEV_UNSUPP, tpgs;
960 mutex_lock(&h->init_mutex);
961 tpgs = alua_check_tpgs(sdev);
962 if (tpgs != TPGS_MODE_NONE)
963 err = alua_check_vpd(sdev, h, tpgs);
964 h->init_error = err;
965 mutex_unlock(&h->init_mutex);
966 return err;
969 * alua_set_params - set/unset the optimize flag
970 * @sdev: device on the path to be activated
971 * params - parameters in the following format
972 * "no_of_params\0param1\0param2\0param3\0...\0"
973 * For example, to set the flag pass the following parameters
974 * from multipath.conf
975 * hardware_handler "2 alua 1"
977 static int alua_set_params(struct scsi_device *sdev, const char *params)
979 struct alua_dh_data *h = sdev->handler_data;
980 struct alua_port_group *pg = NULL;
981 unsigned int optimize = 0, argc;
982 const char *p = params;
983 int result = SCSI_DH_OK;
984 unsigned long flags;
986 if ((sscanf(params, "%u", &argc) != 1) || (argc != 1))
987 return -EINVAL;
989 while (*p++)
991 if ((sscanf(p, "%u", &optimize) != 1) || (optimize > 1))
992 return -EINVAL;
994 rcu_read_lock();
995 pg = rcu_dereference(h->pg);
996 if (!pg) {
997 rcu_read_unlock();
998 return -ENXIO;
1000 spin_lock_irqsave(&pg->lock, flags);
1001 if (optimize)
1002 pg->flags |= ALUA_OPTIMIZE_STPG;
1003 else
1004 pg->flags &= ~ALUA_OPTIMIZE_STPG;
1005 spin_unlock_irqrestore(&pg->lock, flags);
1006 rcu_read_unlock();
1008 return result;
1012 * alua_activate - activate a path
1013 * @sdev: device on the path to be activated
1015 * We're currently switching the port group to be activated only and
1016 * let the array figure out the rest.
1017 * There may be other arrays which require us to switch all port groups
1018 * based on a certain policy. But until we actually encounter them it
1019 * should be okay.
1021 static int alua_activate(struct scsi_device *sdev,
1022 activate_complete fn, void *data)
1024 struct alua_dh_data *h = sdev->handler_data;
1025 int err = SCSI_DH_OK;
1026 struct alua_queue_data *qdata;
1027 struct alua_port_group *pg;
1029 qdata = kzalloc(sizeof(*qdata), GFP_KERNEL);
1030 if (!qdata) {
1031 err = SCSI_DH_RES_TEMP_UNAVAIL;
1032 goto out;
1034 qdata->callback_fn = fn;
1035 qdata->callback_data = data;
1037 mutex_lock(&h->init_mutex);
1038 rcu_read_lock();
1039 pg = rcu_dereference(h->pg);
1040 if (!pg || !kref_get_unless_zero(&pg->kref)) {
1041 rcu_read_unlock();
1042 kfree(qdata);
1043 err = h->init_error;
1044 mutex_unlock(&h->init_mutex);
1045 goto out;
1047 rcu_read_unlock();
1048 mutex_unlock(&h->init_mutex);
1050 if (alua_rtpg_queue(pg, sdev, qdata, true))
1051 fn = NULL;
1052 else
1053 err = SCSI_DH_DEV_OFFLINED;
1054 kref_put(&pg->kref, release_port_group);
1055 out:
1056 if (fn)
1057 fn(data, err);
1058 return 0;
1062 * alua_check - check path status
1063 * @sdev: device on the path to be checked
1065 * Check the device status
1067 static void alua_check(struct scsi_device *sdev, bool force)
1069 struct alua_dh_data *h = sdev->handler_data;
1070 struct alua_port_group *pg;
1072 rcu_read_lock();
1073 pg = rcu_dereference(h->pg);
1074 if (!pg || !kref_get_unless_zero(&pg->kref)) {
1075 rcu_read_unlock();
1076 return;
1078 rcu_read_unlock();
1080 alua_rtpg_queue(pg, sdev, NULL, force);
1081 kref_put(&pg->kref, release_port_group);
1085 * alua_prep_fn - request callback
1087 * Fail I/O to all paths not in state
1088 * active/optimized or active/non-optimized.
1090 static int alua_prep_fn(struct scsi_device *sdev, struct request *req)
1092 struct alua_dh_data *h = sdev->handler_data;
1093 struct alua_port_group *pg;
1094 unsigned char state = SCSI_ACCESS_STATE_OPTIMAL;
1095 int ret = BLKPREP_OK;
1097 rcu_read_lock();
1098 pg = rcu_dereference(h->pg);
1099 if (pg)
1100 state = pg->state;
1101 rcu_read_unlock();
1102 if (state == SCSI_ACCESS_STATE_TRANSITIONING)
1103 ret = BLKPREP_DEFER;
1104 else if (state != SCSI_ACCESS_STATE_OPTIMAL &&
1105 state != SCSI_ACCESS_STATE_ACTIVE &&
1106 state != SCSI_ACCESS_STATE_LBA) {
1107 ret = BLKPREP_KILL;
1108 req->rq_flags |= RQF_QUIET;
1110 return ret;
1114 static void alua_rescan(struct scsi_device *sdev)
1116 struct alua_dh_data *h = sdev->handler_data;
1118 alua_initialize(sdev, h);
1122 * alua_bus_attach - Attach device handler
1123 * @sdev: device to be attached to
1125 static int alua_bus_attach(struct scsi_device *sdev)
1127 struct alua_dh_data *h;
1128 int err;
1130 h = kzalloc(sizeof(*h) , GFP_KERNEL);
1131 if (!h)
1132 return SCSI_DH_NOMEM;
1133 spin_lock_init(&h->pg_lock);
1134 rcu_assign_pointer(h->pg, NULL);
1135 h->init_error = SCSI_DH_OK;
1136 h->sdev = sdev;
1137 INIT_LIST_HEAD(&h->node);
1139 mutex_init(&h->init_mutex);
1140 err = alua_initialize(sdev, h);
1141 if (err != SCSI_DH_OK && err != SCSI_DH_DEV_OFFLINED)
1142 goto failed;
1144 sdev->handler_data = h;
1145 return SCSI_DH_OK;
1146 failed:
1147 kfree(h);
1148 return err;
1152 * alua_bus_detach - Detach device handler
1153 * @sdev: device to be detached from
1155 static void alua_bus_detach(struct scsi_device *sdev)
1157 struct alua_dh_data *h = sdev->handler_data;
1158 struct alua_port_group *pg;
1160 spin_lock(&h->pg_lock);
1161 pg = rcu_dereference_protected(h->pg, lockdep_is_held(&h->pg_lock));
1162 rcu_assign_pointer(h->pg, NULL);
1163 h->sdev = NULL;
1164 spin_unlock(&h->pg_lock);
1165 if (pg) {
1166 spin_lock_irq(&pg->lock);
1167 list_del_rcu(&h->node);
1168 spin_unlock_irq(&pg->lock);
1169 kref_put(&pg->kref, release_port_group);
1171 sdev->handler_data = NULL;
1172 kfree(h);
1175 static struct scsi_device_handler alua_dh = {
1176 .name = ALUA_DH_NAME,
1177 .module = THIS_MODULE,
1178 .attach = alua_bus_attach,
1179 .detach = alua_bus_detach,
1180 .prep_fn = alua_prep_fn,
1181 .check_sense = alua_check_sense,
1182 .activate = alua_activate,
1183 .rescan = alua_rescan,
1184 .set_params = alua_set_params,
1187 static int __init alua_init(void)
1189 int r;
1191 kaluad_wq = alloc_workqueue("kaluad", WQ_MEM_RECLAIM, 0);
1192 if (!kaluad_wq)
1193 return -ENOMEM;
1195 r = scsi_register_device_handler(&alua_dh);
1196 if (r != 0) {
1197 printk(KERN_ERR "%s: Failed to register scsi device handler",
1198 ALUA_DH_NAME);
1199 destroy_workqueue(kaluad_wq);
1201 return r;
1204 static void __exit alua_exit(void)
1206 scsi_unregister_device_handler(&alua_dh);
1207 destroy_workqueue(kaluad_wq);
1210 module_init(alua_init);
1211 module_exit(alua_exit);
1213 MODULE_DESCRIPTION("DM Multipath ALUA support");
1214 MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
1215 MODULE_LICENSE("GPL");
1216 MODULE_VERSION(ALUA_DH_VER);