GNU Linux-libre 5.10.153-gnu1
[releases.git] / drivers / scsi / device_handler / scsi_dh_alua.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Generic SCSI-3 ALUA SCSI Device Handler
4  *
5  * Copyright (C) 2007-2010 Hannes Reinecke, SUSE Linux Products GmbH.
6  * All rights reserved.
7  */
8 #include <linux/slab.h>
9 #include <linux/delay.h>
10 #include <linux/module.h>
11 #include <asm/unaligned.h>
12 #include <scsi/scsi.h>
13 #include <scsi/scsi_proto.h>
14 #include <scsi/scsi_dbg.h>
15 #include <scsi/scsi_eh.h>
16 #include <scsi/scsi_dh.h>
17
18 #define ALUA_DH_NAME "alua"
19 #define ALUA_DH_VER "2.0"
20
21 #define TPGS_SUPPORT_NONE               0x00
22 #define TPGS_SUPPORT_OPTIMIZED          0x01
23 #define TPGS_SUPPORT_NONOPTIMIZED       0x02
24 #define TPGS_SUPPORT_STANDBY            0x04
25 #define TPGS_SUPPORT_UNAVAILABLE        0x08
26 #define TPGS_SUPPORT_LBA_DEPENDENT      0x10
27 #define TPGS_SUPPORT_OFFLINE            0x40
28 #define TPGS_SUPPORT_TRANSITION         0x80
29 #define TPGS_SUPPORT_ALL                0xdf
30
31 #define RTPG_FMT_MASK                   0x70
32 #define RTPG_FMT_EXT_HDR                0x10
33
34 #define TPGS_MODE_UNINITIALIZED          -1
35 #define TPGS_MODE_NONE                  0x0
36 #define TPGS_MODE_IMPLICIT              0x1
37 #define TPGS_MODE_EXPLICIT              0x2
38
39 #define ALUA_RTPG_SIZE                  128
40 #define ALUA_FAILOVER_TIMEOUT           60
41 #define ALUA_FAILOVER_RETRIES           5
42 #define ALUA_RTPG_DELAY_MSECS           5
43 #define ALUA_RTPG_RETRY_DELAY           2
44
45 /* device handler flags */
46 #define ALUA_OPTIMIZE_STPG              0x01
47 #define ALUA_RTPG_EXT_HDR_UNSUPP        0x02
48 /* State machine flags */
49 #define ALUA_PG_RUN_RTPG                0x10
50 #define ALUA_PG_RUN_STPG                0x20
51 #define ALUA_PG_RUNNING                 0x40
52
53 static uint optimize_stpg;
54 module_param(optimize_stpg, uint, S_IRUGO|S_IWUSR);
55 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.");
56
57 static LIST_HEAD(port_group_list);
58 static DEFINE_SPINLOCK(port_group_lock);
59 static struct workqueue_struct *kaluad_wq;
60
61 struct alua_port_group {
62         struct kref             kref;
63         struct rcu_head         rcu;
64         struct list_head        node;
65         struct list_head        dh_list;
66         unsigned char           device_id_str[256];
67         int                     device_id_len;
68         int                     group_id;
69         int                     tpgs;
70         int                     state;
71         int                     pref;
72         int                     valid_states;
73         unsigned                flags; /* used for optimizing STPG */
74         unsigned char           transition_tmo;
75         unsigned long           expiry;
76         unsigned long           interval;
77         struct delayed_work     rtpg_work;
78         spinlock_t              lock;
79         struct list_head        rtpg_list;
80         struct scsi_device      *rtpg_sdev;
81 };
82
83 struct alua_dh_data {
84         struct list_head        node;
85         struct alua_port_group __rcu *pg;
86         int                     group_id;
87         spinlock_t              pg_lock;
88         struct scsi_device      *sdev;
89         int                     init_error;
90         struct mutex            init_mutex;
91 };
92
93 struct alua_queue_data {
94         struct list_head        entry;
95         activate_complete       callback_fn;
96         void                    *callback_data;
97 };
98
99 #define ALUA_POLICY_SWITCH_CURRENT      0
100 #define ALUA_POLICY_SWITCH_ALL          1
101
102 static void alua_rtpg_work(struct work_struct *work);
103 static bool alua_rtpg_queue(struct alua_port_group *pg,
104                             struct scsi_device *sdev,
105                             struct alua_queue_data *qdata, bool force);
106 static void alua_check(struct scsi_device *sdev, bool force);
107
108 static void release_port_group(struct kref *kref)
109 {
110         struct alua_port_group *pg;
111
112         pg = container_of(kref, struct alua_port_group, kref);
113         if (pg->rtpg_sdev)
114                 flush_delayed_work(&pg->rtpg_work);
115         spin_lock(&port_group_lock);
116         list_del(&pg->node);
117         spin_unlock(&port_group_lock);
118         kfree_rcu(pg, rcu);
119 }
120
121 /*
122  * submit_rtpg - Issue a REPORT TARGET GROUP STATES command
123  * @sdev: sdev the command should be sent to
124  */
125 static int submit_rtpg(struct scsi_device *sdev, unsigned char *buff,
126                        int bufflen, struct scsi_sense_hdr *sshdr, int flags)
127 {
128         u8 cdb[MAX_COMMAND_SIZE];
129         int req_flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
130                 REQ_FAILFAST_DRIVER;
131
132         /* Prepare the command. */
133         memset(cdb, 0x0, MAX_COMMAND_SIZE);
134         cdb[0] = MAINTENANCE_IN;
135         if (!(flags & ALUA_RTPG_EXT_HDR_UNSUPP))
136                 cdb[1] = MI_REPORT_TARGET_PGS | MI_EXT_HDR_PARAM_FMT;
137         else
138                 cdb[1] = MI_REPORT_TARGET_PGS;
139         put_unaligned_be32(bufflen, &cdb[6]);
140
141         return scsi_execute(sdev, cdb, DMA_FROM_DEVICE, buff, bufflen, NULL,
142                         sshdr, ALUA_FAILOVER_TIMEOUT * HZ,
143                         ALUA_FAILOVER_RETRIES, req_flags, 0, NULL);
144 }
145
146 /*
147  * submit_stpg - Issue a SET TARGET PORT GROUP command
148  *
149  * Currently we're only setting the current target port group state
150  * to 'active/optimized' and let the array firmware figure out
151  * the states of the remaining groups.
152  */
153 static int submit_stpg(struct scsi_device *sdev, int group_id,
154                        struct scsi_sense_hdr *sshdr)
155 {
156         u8 cdb[MAX_COMMAND_SIZE];
157         unsigned char stpg_data[8];
158         int stpg_len = 8;
159         int req_flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
160                 REQ_FAILFAST_DRIVER;
161
162         /* Prepare the data buffer */
163         memset(stpg_data, 0, stpg_len);
164         stpg_data[4] = SCSI_ACCESS_STATE_OPTIMAL;
165         put_unaligned_be16(group_id, &stpg_data[6]);
166
167         /* Prepare the command. */
168         memset(cdb, 0x0, MAX_COMMAND_SIZE);
169         cdb[0] = MAINTENANCE_OUT;
170         cdb[1] = MO_SET_TARGET_PGS;
171         put_unaligned_be32(stpg_len, &cdb[6]);
172
173         return scsi_execute(sdev, cdb, DMA_TO_DEVICE, stpg_data, stpg_len, NULL,
174                         sshdr, ALUA_FAILOVER_TIMEOUT * HZ,
175                         ALUA_FAILOVER_RETRIES, req_flags, 0, NULL);
176 }
177
178 static struct alua_port_group *alua_find_get_pg(char *id_str, size_t id_size,
179                                                 int group_id)
180 {
181         struct alua_port_group *pg;
182
183         if (!id_str || !id_size || !strlen(id_str))
184                 return NULL;
185
186         list_for_each_entry(pg, &port_group_list, node) {
187                 if (pg->group_id != group_id)
188                         continue;
189                 if (!pg->device_id_len || pg->device_id_len != id_size)
190                         continue;
191                 if (strncmp(pg->device_id_str, id_str, id_size))
192                         continue;
193                 if (!kref_get_unless_zero(&pg->kref))
194                         continue;
195                 return pg;
196         }
197
198         return NULL;
199 }
200
201 /*
202  * alua_alloc_pg - Allocate a new port_group structure
203  * @sdev: scsi device
204  * @group_id: port group id
205  * @tpgs: target port group settings
206  *
207  * Allocate a new port_group structure for a given
208  * device.
209  */
210 static struct alua_port_group *alua_alloc_pg(struct scsi_device *sdev,
211                                              int group_id, int tpgs)
212 {
213         struct alua_port_group *pg, *tmp_pg;
214
215         pg = kzalloc(sizeof(struct alua_port_group), GFP_KERNEL);
216         if (!pg)
217                 return ERR_PTR(-ENOMEM);
218
219         pg->device_id_len = scsi_vpd_lun_id(sdev, pg->device_id_str,
220                                             sizeof(pg->device_id_str));
221         if (pg->device_id_len <= 0) {
222                 /*
223                  * TPGS supported but no device identification found.
224                  * Generate private device identification.
225                  */
226                 sdev_printk(KERN_INFO, sdev,
227                             "%s: No device descriptors found\n",
228                             ALUA_DH_NAME);
229                 pg->device_id_str[0] = '\0';
230                 pg->device_id_len = 0;
231         }
232         pg->group_id = group_id;
233         pg->tpgs = tpgs;
234         pg->state = SCSI_ACCESS_STATE_OPTIMAL;
235         pg->valid_states = TPGS_SUPPORT_ALL;
236         if (optimize_stpg)
237                 pg->flags |= ALUA_OPTIMIZE_STPG;
238         kref_init(&pg->kref);
239         INIT_DELAYED_WORK(&pg->rtpg_work, alua_rtpg_work);
240         INIT_LIST_HEAD(&pg->rtpg_list);
241         INIT_LIST_HEAD(&pg->node);
242         INIT_LIST_HEAD(&pg->dh_list);
243         spin_lock_init(&pg->lock);
244
245         spin_lock(&port_group_lock);
246         tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len,
247                                   group_id);
248         if (tmp_pg) {
249                 spin_unlock(&port_group_lock);
250                 kfree(pg);
251                 return tmp_pg;
252         }
253
254         list_add(&pg->node, &port_group_list);
255         spin_unlock(&port_group_lock);
256
257         return pg;
258 }
259
260 /*
261  * alua_check_tpgs - Evaluate TPGS setting
262  * @sdev: device to be checked
263  *
264  * Examine the TPGS setting of the sdev to find out if ALUA
265  * is supported.
266  */
267 static int alua_check_tpgs(struct scsi_device *sdev)
268 {
269         int tpgs = TPGS_MODE_NONE;
270
271         /*
272          * ALUA support for non-disk devices is fraught with
273          * difficulties, so disable it for now.
274          */
275         if (sdev->type != TYPE_DISK) {
276                 sdev_printk(KERN_INFO, sdev,
277                             "%s: disable for non-disk devices\n",
278                             ALUA_DH_NAME);
279                 return tpgs;
280         }
281
282         tpgs = scsi_device_tpgs(sdev);
283         switch (tpgs) {
284         case TPGS_MODE_EXPLICIT|TPGS_MODE_IMPLICIT:
285                 sdev_printk(KERN_INFO, sdev,
286                             "%s: supports implicit and explicit TPGS\n",
287                             ALUA_DH_NAME);
288                 break;
289         case TPGS_MODE_EXPLICIT:
290                 sdev_printk(KERN_INFO, sdev, "%s: supports explicit TPGS\n",
291                             ALUA_DH_NAME);
292                 break;
293         case TPGS_MODE_IMPLICIT:
294                 sdev_printk(KERN_INFO, sdev, "%s: supports implicit TPGS\n",
295                             ALUA_DH_NAME);
296                 break;
297         case TPGS_MODE_NONE:
298                 sdev_printk(KERN_INFO, sdev, "%s: not supported\n",
299                             ALUA_DH_NAME);
300                 break;
301         default:
302                 sdev_printk(KERN_INFO, sdev,
303                             "%s: unsupported TPGS setting %d\n",
304                             ALUA_DH_NAME, tpgs);
305                 tpgs = TPGS_MODE_NONE;
306                 break;
307         }
308
309         return tpgs;
310 }
311
312 /*
313  * alua_check_vpd - Evaluate INQUIRY vpd page 0x83
314  * @sdev: device to be checked
315  *
316  * Extract the relative target port and the target port group
317  * descriptor from the list of identificators.
318  */
319 static int alua_check_vpd(struct scsi_device *sdev, struct alua_dh_data *h,
320                           int tpgs)
321 {
322         int rel_port = -1, group_id;
323         struct alua_port_group *pg, *old_pg = NULL;
324         bool pg_updated = false;
325         unsigned long flags;
326
327         group_id = scsi_vpd_tpg_id(sdev, &rel_port);
328         if (group_id < 0) {
329                 /*
330                  * Internal error; TPGS supported but required
331                  * VPD identification descriptors not present.
332                  * Disable ALUA support
333                  */
334                 sdev_printk(KERN_INFO, sdev,
335                             "%s: No target port descriptors found\n",
336                             ALUA_DH_NAME);
337                 return SCSI_DH_DEV_UNSUPP;
338         }
339
340         pg = alua_alloc_pg(sdev, group_id, tpgs);
341         if (IS_ERR(pg)) {
342                 if (PTR_ERR(pg) == -ENOMEM)
343                         return SCSI_DH_NOMEM;
344                 return SCSI_DH_DEV_UNSUPP;
345         }
346         if (pg->device_id_len)
347                 sdev_printk(KERN_INFO, sdev,
348                             "%s: device %s port group %x rel port %x\n",
349                             ALUA_DH_NAME, pg->device_id_str,
350                             group_id, rel_port);
351         else
352                 sdev_printk(KERN_INFO, sdev,
353                             "%s: port group %x rel port %x\n",
354                             ALUA_DH_NAME, group_id, rel_port);
355
356         /* Check for existing port group references */
357         spin_lock(&h->pg_lock);
358         old_pg = rcu_dereference_protected(h->pg, lockdep_is_held(&h->pg_lock));
359         if (old_pg != pg) {
360                 /* port group has changed. Update to new port group */
361                 if (h->pg) {
362                         spin_lock_irqsave(&old_pg->lock, flags);
363                         list_del_rcu(&h->node);
364                         spin_unlock_irqrestore(&old_pg->lock, flags);
365                 }
366                 rcu_assign_pointer(h->pg, pg);
367                 pg_updated = true;
368         }
369
370         spin_lock_irqsave(&pg->lock, flags);
371         if (pg_updated)
372                 list_add_rcu(&h->node, &pg->dh_list);
373         spin_unlock_irqrestore(&pg->lock, flags);
374
375         alua_rtpg_queue(rcu_dereference_protected(h->pg,
376                                                   lockdep_is_held(&h->pg_lock)),
377                         sdev, NULL, true);
378         spin_unlock(&h->pg_lock);
379
380         if (old_pg)
381                 kref_put(&old_pg->kref, release_port_group);
382
383         return SCSI_DH_OK;
384 }
385
386 static char print_alua_state(unsigned char state)
387 {
388         switch (state) {
389         case SCSI_ACCESS_STATE_OPTIMAL:
390                 return 'A';
391         case SCSI_ACCESS_STATE_ACTIVE:
392                 return 'N';
393         case SCSI_ACCESS_STATE_STANDBY:
394                 return 'S';
395         case SCSI_ACCESS_STATE_UNAVAILABLE:
396                 return 'U';
397         case SCSI_ACCESS_STATE_LBA:
398                 return 'L';
399         case SCSI_ACCESS_STATE_OFFLINE:
400                 return 'O';
401         case SCSI_ACCESS_STATE_TRANSITIONING:
402                 return 'T';
403         default:
404                 return 'X';
405         }
406 }
407
408 static int alua_check_sense(struct scsi_device *sdev,
409                             struct scsi_sense_hdr *sense_hdr)
410 {
411         switch (sense_hdr->sense_key) {
412         case NOT_READY:
413                 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a) {
414                         /*
415                          * LUN Not Accessible - ALUA state transition
416                          */
417                         alua_check(sdev, false);
418                         return NEEDS_RETRY;
419                 }
420                 break;
421         case UNIT_ATTENTION:
422                 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x00) {
423                         /*
424                          * Power On, Reset, or Bus Device Reset.
425                          * Might have obscured a state transition,
426                          * so schedule a recheck.
427                          */
428                         alua_check(sdev, true);
429                         return ADD_TO_MLQUEUE;
430                 }
431                 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x04)
432                         /*
433                          * Device internal reset
434                          */
435                         return ADD_TO_MLQUEUE;
436                 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x01)
437                         /*
438                          * Mode Parameters Changed
439                          */
440                         return ADD_TO_MLQUEUE;
441                 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x06) {
442                         /*
443                          * ALUA state changed
444                          */
445                         alua_check(sdev, true);
446                         return ADD_TO_MLQUEUE;
447                 }
448                 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x07) {
449                         /*
450                          * Implicit ALUA state transition failed
451                          */
452                         alua_check(sdev, true);
453                         return ADD_TO_MLQUEUE;
454                 }
455                 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x03)
456                         /*
457                          * Inquiry data has changed
458                          */
459                         return ADD_TO_MLQUEUE;
460                 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x0e)
461                         /*
462                          * REPORTED_LUNS_DATA_HAS_CHANGED is reported
463                          * when switching controllers on targets like
464                          * Intel Multi-Flex. We can just retry.
465                          */
466                         return ADD_TO_MLQUEUE;
467                 break;
468         }
469
470         return SCSI_RETURN_NOT_HANDLED;
471 }
472
473 /*
474  * alua_tur - Send a TEST UNIT READY
475  * @sdev: device to which the TEST UNIT READY command should be send
476  *
477  * Send a TEST UNIT READY to @sdev to figure out the device state
478  * Returns SCSI_DH_RETRY if the sense code is NOT READY/ALUA TRANSITIONING,
479  * SCSI_DH_OK if no error occurred, and SCSI_DH_IO otherwise.
480  */
481 static int alua_tur(struct scsi_device *sdev)
482 {
483         struct scsi_sense_hdr sense_hdr;
484         int retval;
485
486         retval = scsi_test_unit_ready(sdev, ALUA_FAILOVER_TIMEOUT * HZ,
487                                       ALUA_FAILOVER_RETRIES, &sense_hdr);
488         if (sense_hdr.sense_key == NOT_READY &&
489             sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a)
490                 return SCSI_DH_RETRY;
491         else if (retval)
492                 return SCSI_DH_IO;
493         else
494                 return SCSI_DH_OK;
495 }
496
497 /*
498  * alua_rtpg - Evaluate REPORT TARGET GROUP STATES
499  * @sdev: the device to be evaluated.
500  *
501  * Evaluate the Target Port Group State.
502  * Returns SCSI_DH_DEV_OFFLINED if the path is
503  * found to be unusable.
504  */
505 static int alua_rtpg(struct scsi_device *sdev, struct alua_port_group *pg)
506 {
507         struct scsi_sense_hdr sense_hdr;
508         struct alua_port_group *tmp_pg;
509         int len, k, off, bufflen = ALUA_RTPG_SIZE;
510         unsigned char *desc, *buff;
511         unsigned err;
512         int retval;
513         unsigned int tpg_desc_tbl_off;
514         unsigned char orig_transition_tmo;
515         unsigned long flags;
516         bool transitioning_sense = false;
517
518         if (!pg->expiry) {
519                 unsigned long transition_tmo = ALUA_FAILOVER_TIMEOUT * HZ;
520
521                 if (pg->transition_tmo)
522                         transition_tmo = pg->transition_tmo * HZ;
523
524                 pg->expiry = round_jiffies_up(jiffies + transition_tmo);
525         }
526
527         buff = kzalloc(bufflen, GFP_KERNEL);
528         if (!buff)
529                 return SCSI_DH_DEV_TEMP_BUSY;
530
531  retry:
532         err = 0;
533         retval = submit_rtpg(sdev, buff, bufflen, &sense_hdr, pg->flags);
534
535         if (retval) {
536                 /*
537                  * Some (broken) implementations have a habit of returning
538                  * an error during things like firmware update etc.
539                  * But if the target only supports active/optimized there's
540                  * not much we can do; it's not that we can switch paths
541                  * or anything.
542                  * So ignore any errors to avoid spurious failures during
543                  * path failover.
544                  */
545                 if ((pg->valid_states & ~TPGS_SUPPORT_OPTIMIZED) == 0) {
546                         sdev_printk(KERN_INFO, sdev,
547                                     "%s: ignoring rtpg result %d\n",
548                                     ALUA_DH_NAME, retval);
549                         kfree(buff);
550                         return SCSI_DH_OK;
551                 }
552                 if (retval < 0 || !scsi_sense_valid(&sense_hdr)) {
553                         sdev_printk(KERN_INFO, sdev,
554                                     "%s: rtpg failed, result %d\n",
555                                     ALUA_DH_NAME, retval);
556                         kfree(buff);
557                         if (retval < 0)
558                                 return SCSI_DH_DEV_TEMP_BUSY;
559                         return SCSI_DH_IO;
560                 }
561
562                 /*
563                  * submit_rtpg() has failed on existing arrays
564                  * when requesting extended header info, and
565                  * the array doesn't support extended headers,
566                  * even though it shouldn't according to T10.
567                  * The retry without rtpg_ext_hdr_req set
568                  * handles this.
569                  * Note:  some arrays return a sense key of ILLEGAL_REQUEST
570                  * with ASC 00h if they don't support the extended header.
571                  */
572                 if (!(pg->flags & ALUA_RTPG_EXT_HDR_UNSUPP) &&
573                     sense_hdr.sense_key == ILLEGAL_REQUEST) {
574                         pg->flags |= ALUA_RTPG_EXT_HDR_UNSUPP;
575                         goto retry;
576                 }
577                 /*
578                  * If the array returns with 'ALUA state transition'
579                  * sense code here it cannot return RTPG data during
580                  * transition. So set the state to 'transitioning' directly.
581                  */
582                 if (sense_hdr.sense_key == NOT_READY &&
583                     sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a) {
584                         transitioning_sense = true;
585                         goto skip_rtpg;
586                 }
587                 /*
588                  * Retry on any other UNIT ATTENTION occurred.
589                  */
590                 if (sense_hdr.sense_key == UNIT_ATTENTION)
591                         err = SCSI_DH_RETRY;
592                 if (err == SCSI_DH_RETRY &&
593                     pg->expiry != 0 && time_before(jiffies, pg->expiry)) {
594                         sdev_printk(KERN_ERR, sdev, "%s: rtpg retry\n",
595                                     ALUA_DH_NAME);
596                         scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
597                         kfree(buff);
598                         return err;
599                 }
600                 sdev_printk(KERN_ERR, sdev, "%s: rtpg failed\n",
601                             ALUA_DH_NAME);
602                 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
603                 kfree(buff);
604                 pg->expiry = 0;
605                 return SCSI_DH_IO;
606         }
607
608         len = get_unaligned_be32(&buff[0]) + 4;
609
610         if (len > bufflen) {
611                 /* Resubmit with the correct length */
612                 kfree(buff);
613                 bufflen = len;
614                 buff = kmalloc(bufflen, GFP_KERNEL);
615                 if (!buff) {
616                         sdev_printk(KERN_WARNING, sdev,
617                                     "%s: kmalloc buffer failed\n",__func__);
618                         /* Temporary failure, bypass */
619                         pg->expiry = 0;
620                         return SCSI_DH_DEV_TEMP_BUSY;
621                 }
622                 goto retry;
623         }
624
625         orig_transition_tmo = pg->transition_tmo;
626         if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR && buff[5] != 0)
627                 pg->transition_tmo = buff[5];
628         else
629                 pg->transition_tmo = ALUA_FAILOVER_TIMEOUT;
630
631         if (orig_transition_tmo != pg->transition_tmo) {
632                 sdev_printk(KERN_INFO, sdev,
633                             "%s: transition timeout set to %d seconds\n",
634                             ALUA_DH_NAME, pg->transition_tmo);
635                 pg->expiry = jiffies + pg->transition_tmo * HZ;
636         }
637
638         if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR)
639                 tpg_desc_tbl_off = 8;
640         else
641                 tpg_desc_tbl_off = 4;
642
643         for (k = tpg_desc_tbl_off, desc = buff + tpg_desc_tbl_off;
644              k < len;
645              k += off, desc += off) {
646                 u16 group_id = get_unaligned_be16(&desc[2]);
647
648                 spin_lock_irqsave(&port_group_lock, flags);
649                 tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len,
650                                           group_id);
651                 spin_unlock_irqrestore(&port_group_lock, flags);
652                 if (tmp_pg) {
653                         if (spin_trylock_irqsave(&tmp_pg->lock, flags)) {
654                                 if ((tmp_pg == pg) ||
655                                     !(tmp_pg->flags & ALUA_PG_RUNNING)) {
656                                         struct alua_dh_data *h;
657
658                                         tmp_pg->state = desc[0] & 0x0f;
659                                         tmp_pg->pref = desc[0] >> 7;
660                                         rcu_read_lock();
661                                         list_for_each_entry_rcu(h,
662                                                 &tmp_pg->dh_list, node) {
663                                                 if (!h->sdev)
664                                                         continue;
665                                                 h->sdev->access_state = desc[0];
666                                         }
667                                         rcu_read_unlock();
668                                 }
669                                 if (tmp_pg == pg)
670                                         tmp_pg->valid_states = desc[1];
671                                 spin_unlock_irqrestore(&tmp_pg->lock, flags);
672                         }
673                         kref_put(&tmp_pg->kref, release_port_group);
674                 }
675                 off = 8 + (desc[7] * 4);
676         }
677
678  skip_rtpg:
679         spin_lock_irqsave(&pg->lock, flags);
680         if (transitioning_sense)
681                 pg->state = SCSI_ACCESS_STATE_TRANSITIONING;
682
683         sdev_printk(KERN_INFO, sdev,
684                     "%s: port group %02x state %c %s supports %c%c%c%c%c%c%c\n",
685                     ALUA_DH_NAME, pg->group_id, print_alua_state(pg->state),
686                     pg->pref ? "preferred" : "non-preferred",
687                     pg->valid_states&TPGS_SUPPORT_TRANSITION?'T':'t',
688                     pg->valid_states&TPGS_SUPPORT_OFFLINE?'O':'o',
689                     pg->valid_states&TPGS_SUPPORT_LBA_DEPENDENT?'L':'l',
690                     pg->valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u',
691                     pg->valid_states&TPGS_SUPPORT_STANDBY?'S':'s',
692                     pg->valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n',
693                     pg->valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a');
694
695         switch (pg->state) {
696         case SCSI_ACCESS_STATE_TRANSITIONING:
697                 if (time_before(jiffies, pg->expiry)) {
698                         /* State transition, retry */
699                         pg->interval = ALUA_RTPG_RETRY_DELAY;
700                         err = SCSI_DH_RETRY;
701                 } else {
702                         struct alua_dh_data *h;
703
704                         /* Transitioning time exceeded, set port to standby */
705                         err = SCSI_DH_IO;
706                         pg->state = SCSI_ACCESS_STATE_STANDBY;
707                         pg->expiry = 0;
708                         rcu_read_lock();
709                         list_for_each_entry_rcu(h, &pg->dh_list, node) {
710                                 if (!h->sdev)
711                                         continue;
712                                 h->sdev->access_state =
713                                         (pg->state & SCSI_ACCESS_STATE_MASK);
714                                 if (pg->pref)
715                                         h->sdev->access_state |=
716                                                 SCSI_ACCESS_STATE_PREFERRED;
717                         }
718                         rcu_read_unlock();
719                 }
720                 break;
721         case SCSI_ACCESS_STATE_OFFLINE:
722                 /* Path unusable */
723                 err = SCSI_DH_DEV_OFFLINED;
724                 pg->expiry = 0;
725                 break;
726         default:
727                 /* Useable path if active */
728                 err = SCSI_DH_OK;
729                 pg->expiry = 0;
730                 break;
731         }
732         spin_unlock_irqrestore(&pg->lock, flags);
733         kfree(buff);
734         return err;
735 }
736
737 /*
738  * alua_stpg - Issue a SET TARGET PORT GROUP command
739  *
740  * Issue a SET TARGET PORT GROUP command and evaluate the
741  * response. Returns SCSI_DH_RETRY per default to trigger
742  * a re-evaluation of the target group state or SCSI_DH_OK
743  * if no further action needs to be taken.
744  */
745 static unsigned alua_stpg(struct scsi_device *sdev, struct alua_port_group *pg)
746 {
747         int retval;
748         struct scsi_sense_hdr sense_hdr;
749
750         if (!(pg->tpgs & TPGS_MODE_EXPLICIT)) {
751                 /* Only implicit ALUA supported, retry */
752                 return SCSI_DH_RETRY;
753         }
754         switch (pg->state) {
755         case SCSI_ACCESS_STATE_OPTIMAL:
756                 return SCSI_DH_OK;
757         case SCSI_ACCESS_STATE_ACTIVE:
758                 if ((pg->flags & ALUA_OPTIMIZE_STPG) &&
759                     !pg->pref &&
760                     (pg->tpgs & TPGS_MODE_IMPLICIT))
761                         return SCSI_DH_OK;
762                 break;
763         case SCSI_ACCESS_STATE_STANDBY:
764         case SCSI_ACCESS_STATE_UNAVAILABLE:
765                 break;
766         case SCSI_ACCESS_STATE_OFFLINE:
767                 return SCSI_DH_IO;
768         case SCSI_ACCESS_STATE_TRANSITIONING:
769                 break;
770         default:
771                 sdev_printk(KERN_INFO, sdev,
772                             "%s: stpg failed, unhandled TPGS state %d",
773                             ALUA_DH_NAME, pg->state);
774                 return SCSI_DH_NOSYS;
775         }
776         retval = submit_stpg(sdev, pg->group_id, &sense_hdr);
777
778         if (retval) {
779                 if (retval < 0 || !scsi_sense_valid(&sense_hdr)) {
780                         sdev_printk(KERN_INFO, sdev,
781                                     "%s: stpg failed, result %d",
782                                     ALUA_DH_NAME, retval);
783                         if (retval < 0)
784                                 return SCSI_DH_DEV_TEMP_BUSY;
785                 } else {
786                         sdev_printk(KERN_INFO, sdev, "%s: stpg failed\n",
787                                     ALUA_DH_NAME);
788                         scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
789                 }
790         }
791         /* Retry RTPG */
792         return SCSI_DH_RETRY;
793 }
794
795 static void alua_rtpg_work(struct work_struct *work)
796 {
797         struct alua_port_group *pg =
798                 container_of(work, struct alua_port_group, rtpg_work.work);
799         struct scsi_device *sdev;
800         LIST_HEAD(qdata_list);
801         int err = SCSI_DH_OK;
802         struct alua_queue_data *qdata, *tmp;
803         unsigned long flags;
804
805         spin_lock_irqsave(&pg->lock, flags);
806         sdev = pg->rtpg_sdev;
807         if (!sdev) {
808                 WARN_ON(pg->flags & ALUA_PG_RUN_RTPG);
809                 WARN_ON(pg->flags & ALUA_PG_RUN_STPG);
810                 spin_unlock_irqrestore(&pg->lock, flags);
811                 kref_put(&pg->kref, release_port_group);
812                 return;
813         }
814         pg->flags |= ALUA_PG_RUNNING;
815         if (pg->flags & ALUA_PG_RUN_RTPG) {
816                 int state = pg->state;
817
818                 pg->flags &= ~ALUA_PG_RUN_RTPG;
819                 spin_unlock_irqrestore(&pg->lock, flags);
820                 if (state == SCSI_ACCESS_STATE_TRANSITIONING) {
821                         if (alua_tur(sdev) == SCSI_DH_RETRY) {
822                                 spin_lock_irqsave(&pg->lock, flags);
823                                 pg->flags &= ~ALUA_PG_RUNNING;
824                                 pg->flags |= ALUA_PG_RUN_RTPG;
825                                 if (!pg->interval)
826                                         pg->interval = ALUA_RTPG_RETRY_DELAY;
827                                 spin_unlock_irqrestore(&pg->lock, flags);
828                                 queue_delayed_work(kaluad_wq, &pg->rtpg_work,
829                                                    pg->interval * HZ);
830                                 return;
831                         }
832                         /* Send RTPG on failure or if TUR indicates SUCCESS */
833                 }
834                 err = alua_rtpg(sdev, pg);
835                 spin_lock_irqsave(&pg->lock, flags);
836                 if (err == SCSI_DH_RETRY || pg->flags & ALUA_PG_RUN_RTPG) {
837                         pg->flags &= ~ALUA_PG_RUNNING;
838                         if (!pg->interval && !(pg->flags & ALUA_PG_RUN_RTPG))
839                                 pg->interval = ALUA_RTPG_RETRY_DELAY;
840                         pg->flags |= ALUA_PG_RUN_RTPG;
841                         spin_unlock_irqrestore(&pg->lock, flags);
842                         queue_delayed_work(kaluad_wq, &pg->rtpg_work,
843                                            pg->interval * HZ);
844                         return;
845                 }
846                 if (err != SCSI_DH_OK)
847                         pg->flags &= ~ALUA_PG_RUN_STPG;
848         }
849         if (pg->flags & ALUA_PG_RUN_STPG) {
850                 pg->flags &= ~ALUA_PG_RUN_STPG;
851                 spin_unlock_irqrestore(&pg->lock, flags);
852                 err = alua_stpg(sdev, pg);
853                 spin_lock_irqsave(&pg->lock, flags);
854                 if (err == SCSI_DH_RETRY || pg->flags & ALUA_PG_RUN_RTPG) {
855                         pg->flags |= ALUA_PG_RUN_RTPG;
856                         pg->interval = 0;
857                         pg->flags &= ~ALUA_PG_RUNNING;
858                         spin_unlock_irqrestore(&pg->lock, flags);
859                         queue_delayed_work(kaluad_wq, &pg->rtpg_work,
860                                            pg->interval * HZ);
861                         return;
862                 }
863         }
864
865         list_splice_init(&pg->rtpg_list, &qdata_list);
866         pg->rtpg_sdev = NULL;
867         spin_unlock_irqrestore(&pg->lock, flags);
868
869         list_for_each_entry_safe(qdata, tmp, &qdata_list, entry) {
870                 list_del(&qdata->entry);
871                 if (qdata->callback_fn)
872                         qdata->callback_fn(qdata->callback_data, err);
873                 kfree(qdata);
874         }
875         spin_lock_irqsave(&pg->lock, flags);
876         pg->flags &= ~ALUA_PG_RUNNING;
877         spin_unlock_irqrestore(&pg->lock, flags);
878         scsi_device_put(sdev);
879         kref_put(&pg->kref, release_port_group);
880 }
881
882 /**
883  * alua_rtpg_queue() - cause RTPG to be submitted asynchronously
884  * @pg: ALUA port group associated with @sdev.
885  * @sdev: SCSI device for which to submit an RTPG.
886  * @qdata: Information about the callback to invoke after the RTPG.
887  * @force: Whether or not to submit an RTPG if a work item that will submit an
888  *         RTPG already has been scheduled.
889  *
890  * Returns true if and only if alua_rtpg_work() will be called asynchronously.
891  * That function is responsible for calling @qdata->fn().
892  */
893 static bool alua_rtpg_queue(struct alua_port_group *pg,
894                             struct scsi_device *sdev,
895                             struct alua_queue_data *qdata, bool force)
896 {
897         int start_queue = 0;
898         unsigned long flags;
899         if (WARN_ON_ONCE(!pg) || scsi_device_get(sdev))
900                 return false;
901
902         spin_lock_irqsave(&pg->lock, flags);
903         if (qdata) {
904                 list_add_tail(&qdata->entry, &pg->rtpg_list);
905                 pg->flags |= ALUA_PG_RUN_STPG;
906                 force = true;
907         }
908         if (pg->rtpg_sdev == NULL) {
909                 pg->interval = 0;
910                 pg->flags |= ALUA_PG_RUN_RTPG;
911                 kref_get(&pg->kref);
912                 pg->rtpg_sdev = sdev;
913                 start_queue = 1;
914         } else if (!(pg->flags & ALUA_PG_RUN_RTPG) && force) {
915                 pg->flags |= ALUA_PG_RUN_RTPG;
916                 /* Do not queue if the worker is already running */
917                 if (!(pg->flags & ALUA_PG_RUNNING)) {
918                         kref_get(&pg->kref);
919                         start_queue = 1;
920                 }
921         }
922
923         spin_unlock_irqrestore(&pg->lock, flags);
924
925         if (start_queue) {
926                 if (queue_delayed_work(kaluad_wq, &pg->rtpg_work,
927                                 msecs_to_jiffies(ALUA_RTPG_DELAY_MSECS)))
928                         sdev = NULL;
929                 else
930                         kref_put(&pg->kref, release_port_group);
931         }
932         if (sdev)
933                 scsi_device_put(sdev);
934
935         return true;
936 }
937
938 /*
939  * alua_initialize - Initialize ALUA state
940  * @sdev: the device to be initialized
941  *
942  * For the prep_fn to work correctly we have
943  * to initialize the ALUA state for the device.
944  */
945 static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h)
946 {
947         int err = SCSI_DH_DEV_UNSUPP, tpgs;
948
949         mutex_lock(&h->init_mutex);
950         tpgs = alua_check_tpgs(sdev);
951         if (tpgs != TPGS_MODE_NONE)
952                 err = alua_check_vpd(sdev, h, tpgs);
953         h->init_error = err;
954         mutex_unlock(&h->init_mutex);
955         return err;
956 }
957 /*
958  * alua_set_params - set/unset the optimize flag
959  * @sdev: device on the path to be activated
960  * params - parameters in the following format
961  *      "no_of_params\0param1\0param2\0param3\0...\0"
962  * For example, to set the flag pass the following parameters
963  * from multipath.conf
964  *     hardware_handler        "2 alua 1"
965  */
966 static int alua_set_params(struct scsi_device *sdev, const char *params)
967 {
968         struct alua_dh_data *h = sdev->handler_data;
969         struct alua_port_group *pg = NULL;
970         unsigned int optimize = 0, argc;
971         const char *p = params;
972         int result = SCSI_DH_OK;
973         unsigned long flags;
974
975         if ((sscanf(params, "%u", &argc) != 1) || (argc != 1))
976                 return -EINVAL;
977
978         while (*p++)
979                 ;
980         if ((sscanf(p, "%u", &optimize) != 1) || (optimize > 1))
981                 return -EINVAL;
982
983         rcu_read_lock();
984         pg = rcu_dereference(h->pg);
985         if (!pg) {
986                 rcu_read_unlock();
987                 return -ENXIO;
988         }
989         spin_lock_irqsave(&pg->lock, flags);
990         if (optimize)
991                 pg->flags |= ALUA_OPTIMIZE_STPG;
992         else
993                 pg->flags &= ~ALUA_OPTIMIZE_STPG;
994         spin_unlock_irqrestore(&pg->lock, flags);
995         rcu_read_unlock();
996
997         return result;
998 }
999
1000 /*
1001  * alua_activate - activate a path
1002  * @sdev: device on the path to be activated
1003  *
1004  * We're currently switching the port group to be activated only and
1005  * let the array figure out the rest.
1006  * There may be other arrays which require us to switch all port groups
1007  * based on a certain policy. But until we actually encounter them it
1008  * should be okay.
1009  */
1010 static int alua_activate(struct scsi_device *sdev,
1011                         activate_complete fn, void *data)
1012 {
1013         struct alua_dh_data *h = sdev->handler_data;
1014         int err = SCSI_DH_OK;
1015         struct alua_queue_data *qdata;
1016         struct alua_port_group *pg;
1017
1018         qdata = kzalloc(sizeof(*qdata), GFP_KERNEL);
1019         if (!qdata) {
1020                 err = SCSI_DH_RES_TEMP_UNAVAIL;
1021                 goto out;
1022         }
1023         qdata->callback_fn = fn;
1024         qdata->callback_data = data;
1025
1026         mutex_lock(&h->init_mutex);
1027         rcu_read_lock();
1028         pg = rcu_dereference(h->pg);
1029         if (!pg || !kref_get_unless_zero(&pg->kref)) {
1030                 rcu_read_unlock();
1031                 kfree(qdata);
1032                 err = h->init_error;
1033                 mutex_unlock(&h->init_mutex);
1034                 goto out;
1035         }
1036         rcu_read_unlock();
1037         mutex_unlock(&h->init_mutex);
1038
1039         if (alua_rtpg_queue(pg, sdev, qdata, true))
1040                 fn = NULL;
1041         else
1042                 err = SCSI_DH_DEV_OFFLINED;
1043         kref_put(&pg->kref, release_port_group);
1044 out:
1045         if (fn)
1046                 fn(data, err);
1047         return 0;
1048 }
1049
1050 /*
1051  * alua_check - check path status
1052  * @sdev: device on the path to be checked
1053  *
1054  * Check the device status
1055  */
1056 static void alua_check(struct scsi_device *sdev, bool force)
1057 {
1058         struct alua_dh_data *h = sdev->handler_data;
1059         struct alua_port_group *pg;
1060
1061         rcu_read_lock();
1062         pg = rcu_dereference(h->pg);
1063         if (!pg || !kref_get_unless_zero(&pg->kref)) {
1064                 rcu_read_unlock();
1065                 return;
1066         }
1067         rcu_read_unlock();
1068
1069         alua_rtpg_queue(pg, sdev, NULL, force);
1070         kref_put(&pg->kref, release_port_group);
1071 }
1072
1073 /*
1074  * alua_prep_fn - request callback
1075  *
1076  * Fail I/O to all paths not in state
1077  * active/optimized or active/non-optimized.
1078  */
1079 static blk_status_t alua_prep_fn(struct scsi_device *sdev, struct request *req)
1080 {
1081         struct alua_dh_data *h = sdev->handler_data;
1082         struct alua_port_group *pg;
1083         unsigned char state = SCSI_ACCESS_STATE_OPTIMAL;
1084
1085         rcu_read_lock();
1086         pg = rcu_dereference(h->pg);
1087         if (pg)
1088                 state = pg->state;
1089         rcu_read_unlock();
1090
1091         switch (state) {
1092         case SCSI_ACCESS_STATE_OPTIMAL:
1093         case SCSI_ACCESS_STATE_ACTIVE:
1094         case SCSI_ACCESS_STATE_LBA:
1095                 return BLK_STS_OK;
1096         case SCSI_ACCESS_STATE_TRANSITIONING:
1097                 return BLK_STS_RESOURCE;
1098         default:
1099                 req->rq_flags |= RQF_QUIET;
1100                 return BLK_STS_IOERR;
1101         }
1102 }
1103
1104 static void alua_rescan(struct scsi_device *sdev)
1105 {
1106         struct alua_dh_data *h = sdev->handler_data;
1107
1108         alua_initialize(sdev, h);
1109 }
1110
1111 /*
1112  * alua_bus_attach - Attach device handler
1113  * @sdev: device to be attached to
1114  */
1115 static int alua_bus_attach(struct scsi_device *sdev)
1116 {
1117         struct alua_dh_data *h;
1118         int err;
1119
1120         h = kzalloc(sizeof(*h) , GFP_KERNEL);
1121         if (!h)
1122                 return SCSI_DH_NOMEM;
1123         spin_lock_init(&h->pg_lock);
1124         rcu_assign_pointer(h->pg, NULL);
1125         h->init_error = SCSI_DH_OK;
1126         h->sdev = sdev;
1127         INIT_LIST_HEAD(&h->node);
1128
1129         mutex_init(&h->init_mutex);
1130         err = alua_initialize(sdev, h);
1131         if (err != SCSI_DH_OK && err != SCSI_DH_DEV_OFFLINED)
1132                 goto failed;
1133
1134         sdev->handler_data = h;
1135         return SCSI_DH_OK;
1136 failed:
1137         kfree(h);
1138         return err;
1139 }
1140
1141 /*
1142  * alua_bus_detach - Detach device handler
1143  * @sdev: device to be detached from
1144  */
1145 static void alua_bus_detach(struct scsi_device *sdev)
1146 {
1147         struct alua_dh_data *h = sdev->handler_data;
1148         struct alua_port_group *pg;
1149
1150         spin_lock(&h->pg_lock);
1151         pg = rcu_dereference_protected(h->pg, lockdep_is_held(&h->pg_lock));
1152         rcu_assign_pointer(h->pg, NULL);
1153         spin_unlock(&h->pg_lock);
1154         if (pg) {
1155                 spin_lock_irq(&pg->lock);
1156                 list_del_rcu(&h->node);
1157                 spin_unlock_irq(&pg->lock);
1158                 kref_put(&pg->kref, release_port_group);
1159         }
1160         sdev->handler_data = NULL;
1161         synchronize_rcu();
1162         kfree(h);
1163 }
1164
1165 static struct scsi_device_handler alua_dh = {
1166         .name = ALUA_DH_NAME,
1167         .module = THIS_MODULE,
1168         .attach = alua_bus_attach,
1169         .detach = alua_bus_detach,
1170         .prep_fn = alua_prep_fn,
1171         .check_sense = alua_check_sense,
1172         .activate = alua_activate,
1173         .rescan = alua_rescan,
1174         .set_params = alua_set_params,
1175 };
1176
1177 static int __init alua_init(void)
1178 {
1179         int r;
1180
1181         kaluad_wq = alloc_workqueue("kaluad", WQ_MEM_RECLAIM, 0);
1182         if (!kaluad_wq)
1183                 return -ENOMEM;
1184
1185         r = scsi_register_device_handler(&alua_dh);
1186         if (r != 0) {
1187                 printk(KERN_ERR "%s: Failed to register scsi device handler",
1188                         ALUA_DH_NAME);
1189                 destroy_workqueue(kaluad_wq);
1190         }
1191         return r;
1192 }
1193
1194 static void __exit alua_exit(void)
1195 {
1196         scsi_unregister_device_handler(&alua_dh);
1197         destroy_workqueue(kaluad_wq);
1198 }
1199
1200 module_init(alua_init);
1201 module_exit(alua_exit);
1202
1203 MODULE_DESCRIPTION("DM Multipath ALUA support");
1204 MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
1205 MODULE_LICENSE("GPL");
1206 MODULE_VERSION(ALUA_DH_VER);