Merge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
[sfrench/cifs-2.6.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
44 /* device handler flags */
45 #define ALUA_OPTIMIZE_STPG              0x01
46 #define ALUA_RTPG_EXT_HDR_UNSUPP        0x02
47 /* State machine flags */
48 #define ALUA_PG_RUN_RTPG                0x10
49 #define ALUA_PG_RUN_STPG                0x20
50 #define ALUA_PG_RUNNING                 0x40
51
52 static uint optimize_stpg;
53 module_param(optimize_stpg, uint, S_IRUGO|S_IWUSR);
54 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.");
55
56 static LIST_HEAD(port_group_list);
57 static DEFINE_SPINLOCK(port_group_lock);
58 static struct workqueue_struct *kaluad_wq;
59
60 struct alua_port_group {
61         struct kref             kref;
62         struct rcu_head         rcu;
63         struct list_head        node;
64         struct list_head        dh_list;
65         unsigned char           device_id_str[256];
66         int                     device_id_len;
67         int                     group_id;
68         int                     tpgs;
69         int                     state;
70         int                     pref;
71         int                     valid_states;
72         unsigned                flags; /* used for optimizing STPG */
73         unsigned char           transition_tmo;
74         unsigned long           expiry;
75         unsigned long           interval;
76         struct delayed_work     rtpg_work;
77         spinlock_t              lock;
78         struct list_head        rtpg_list;
79         struct scsi_device      *rtpg_sdev;
80 };
81
82 struct alua_dh_data {
83         struct list_head        node;
84         struct alua_port_group __rcu *pg;
85         int                     group_id;
86         spinlock_t              pg_lock;
87         struct scsi_device      *sdev;
88         int                     init_error;
89         struct mutex            init_mutex;
90 };
91
92 struct alua_queue_data {
93         struct list_head        entry;
94         activate_complete       callback_fn;
95         void                    *callback_data;
96 };
97
98 #define ALUA_POLICY_SWITCH_CURRENT      0
99 #define ALUA_POLICY_SWITCH_ALL          1
100
101 static void alua_rtpg_work(struct work_struct *work);
102 static bool alua_rtpg_queue(struct alua_port_group *pg,
103                             struct scsi_device *sdev,
104                             struct alua_queue_data *qdata, bool force);
105 static void alua_check(struct scsi_device *sdev, bool force);
106
107 static void release_port_group(struct kref *kref)
108 {
109         struct alua_port_group *pg;
110
111         pg = container_of(kref, struct alua_port_group, kref);
112         if (pg->rtpg_sdev)
113                 flush_delayed_work(&pg->rtpg_work);
114         spin_lock(&port_group_lock);
115         list_del(&pg->node);
116         spin_unlock(&port_group_lock);
117         kfree_rcu(pg, rcu);
118 }
119
120 /*
121  * submit_rtpg - Issue a REPORT TARGET GROUP STATES command
122  * @sdev: sdev the command should be sent to
123  */
124 static int submit_rtpg(struct scsi_device *sdev, unsigned char *buff,
125                        int bufflen, struct scsi_sense_hdr *sshdr, int flags)
126 {
127         u8 cdb[MAX_COMMAND_SIZE];
128         int req_flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
129                 REQ_FAILFAST_DRIVER;
130
131         /* Prepare the command. */
132         memset(cdb, 0x0, MAX_COMMAND_SIZE);
133         cdb[0] = MAINTENANCE_IN;
134         if (!(flags & ALUA_RTPG_EXT_HDR_UNSUPP))
135                 cdb[1] = MI_REPORT_TARGET_PGS | MI_EXT_HDR_PARAM_FMT;
136         else
137                 cdb[1] = MI_REPORT_TARGET_PGS;
138         put_unaligned_be32(bufflen, &cdb[6]);
139
140         return scsi_execute(sdev, cdb, DMA_FROM_DEVICE, buff, bufflen, NULL,
141                         sshdr, ALUA_FAILOVER_TIMEOUT * HZ,
142                         ALUA_FAILOVER_RETRIES, req_flags, 0, NULL);
143 }
144
145 /*
146  * submit_stpg - Issue a SET TARGET PORT GROUP command
147  *
148  * Currently we're only setting the current target port group state
149  * to 'active/optimized' and let the array firmware figure out
150  * the states of the remaining groups.
151  */
152 static int submit_stpg(struct scsi_device *sdev, int group_id,
153                        struct scsi_sense_hdr *sshdr)
154 {
155         u8 cdb[MAX_COMMAND_SIZE];
156         unsigned char stpg_data[8];
157         int stpg_len = 8;
158         int req_flags = REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
159                 REQ_FAILFAST_DRIVER;
160
161         /* Prepare the data buffer */
162         memset(stpg_data, 0, stpg_len);
163         stpg_data[4] = SCSI_ACCESS_STATE_OPTIMAL;
164         put_unaligned_be16(group_id, &stpg_data[6]);
165
166         /* Prepare the command. */
167         memset(cdb, 0x0, MAX_COMMAND_SIZE);
168         cdb[0] = MAINTENANCE_OUT;
169         cdb[1] = MO_SET_TARGET_PGS;
170         put_unaligned_be32(stpg_len, &cdb[6]);
171
172         return scsi_execute(sdev, cdb, DMA_TO_DEVICE, stpg_data, stpg_len, NULL,
173                         sshdr, ALUA_FAILOVER_TIMEOUT * HZ,
174                         ALUA_FAILOVER_RETRIES, req_flags, 0, NULL);
175 }
176
177 static struct alua_port_group *alua_find_get_pg(char *id_str, size_t id_size,
178                                                 int group_id)
179 {
180         struct alua_port_group *pg;
181
182         if (!id_str || !id_size || !strlen(id_str))
183                 return NULL;
184
185         list_for_each_entry(pg, &port_group_list, node) {
186                 if (pg->group_id != group_id)
187                         continue;
188                 if (!pg->device_id_len || pg->device_id_len != id_size)
189                         continue;
190                 if (strncmp(pg->device_id_str, id_str, id_size))
191                         continue;
192                 if (!kref_get_unless_zero(&pg->kref))
193                         continue;
194                 return pg;
195         }
196
197         return NULL;
198 }
199
200 /*
201  * alua_alloc_pg - Allocate a new port_group structure
202  * @sdev: scsi device
203  * @group_id: port group id
204  * @tpgs: target port group settings
205  *
206  * Allocate a new port_group structure for a given
207  * device.
208  */
209 static struct alua_port_group *alua_alloc_pg(struct scsi_device *sdev,
210                                              int group_id, int tpgs)
211 {
212         struct alua_port_group *pg, *tmp_pg;
213
214         pg = kzalloc(sizeof(struct alua_port_group), GFP_KERNEL);
215         if (!pg)
216                 return ERR_PTR(-ENOMEM);
217
218         pg->device_id_len = scsi_vpd_lun_id(sdev, pg->device_id_str,
219                                             sizeof(pg->device_id_str));
220         if (pg->device_id_len <= 0) {
221                 /*
222                  * TPGS supported but no device identification found.
223                  * Generate private device identification.
224                  */
225                 sdev_printk(KERN_INFO, sdev,
226                             "%s: No device descriptors found\n",
227                             ALUA_DH_NAME);
228                 pg->device_id_str[0] = '\0';
229                 pg->device_id_len = 0;
230         }
231         pg->group_id = group_id;
232         pg->tpgs = tpgs;
233         pg->state = SCSI_ACCESS_STATE_OPTIMAL;
234         pg->valid_states = TPGS_SUPPORT_ALL;
235         if (optimize_stpg)
236                 pg->flags |= ALUA_OPTIMIZE_STPG;
237         kref_init(&pg->kref);
238         INIT_DELAYED_WORK(&pg->rtpg_work, alua_rtpg_work);
239         INIT_LIST_HEAD(&pg->rtpg_list);
240         INIT_LIST_HEAD(&pg->node);
241         INIT_LIST_HEAD(&pg->dh_list);
242         spin_lock_init(&pg->lock);
243
244         spin_lock(&port_group_lock);
245         tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len,
246                                   group_id);
247         if (tmp_pg) {
248                 spin_unlock(&port_group_lock);
249                 kfree(pg);
250                 return tmp_pg;
251         }
252
253         list_add(&pg->node, &port_group_list);
254         spin_unlock(&port_group_lock);
255
256         return pg;
257 }
258
259 /*
260  * alua_check_tpgs - Evaluate TPGS setting
261  * @sdev: device to be checked
262  *
263  * Examine the TPGS setting of the sdev to find out if ALUA
264  * is supported.
265  */
266 static int alua_check_tpgs(struct scsi_device *sdev)
267 {
268         int tpgs = TPGS_MODE_NONE;
269
270         /*
271          * ALUA support for non-disk devices is fraught with
272          * difficulties, so disable it for now.
273          */
274         if (sdev->type != TYPE_DISK) {
275                 sdev_printk(KERN_INFO, sdev,
276                             "%s: disable for non-disk devices\n",
277                             ALUA_DH_NAME);
278                 return tpgs;
279         }
280
281         tpgs = scsi_device_tpgs(sdev);
282         switch (tpgs) {
283         case TPGS_MODE_EXPLICIT|TPGS_MODE_IMPLICIT:
284                 sdev_printk(KERN_INFO, sdev,
285                             "%s: supports implicit and explicit TPGS\n",
286                             ALUA_DH_NAME);
287                 break;
288         case TPGS_MODE_EXPLICIT:
289                 sdev_printk(KERN_INFO, sdev, "%s: supports explicit TPGS\n",
290                             ALUA_DH_NAME);
291                 break;
292         case TPGS_MODE_IMPLICIT:
293                 sdev_printk(KERN_INFO, sdev, "%s: supports implicit TPGS\n",
294                             ALUA_DH_NAME);
295                 break;
296         case TPGS_MODE_NONE:
297                 sdev_printk(KERN_INFO, sdev, "%s: not supported\n",
298                             ALUA_DH_NAME);
299                 break;
300         default:
301                 sdev_printk(KERN_INFO, sdev,
302                             "%s: unsupported TPGS setting %d\n",
303                             ALUA_DH_NAME, tpgs);
304                 tpgs = TPGS_MODE_NONE;
305                 break;
306         }
307
308         return tpgs;
309 }
310
311 /*
312  * alua_check_vpd - Evaluate INQUIRY vpd page 0x83
313  * @sdev: device to be checked
314  *
315  * Extract the relative target port and the target port group
316  * descriptor from the list of identificators.
317  */
318 static int alua_check_vpd(struct scsi_device *sdev, struct alua_dh_data *h,
319                           int tpgs)
320 {
321         int rel_port = -1, group_id;
322         struct alua_port_group *pg, *old_pg = NULL;
323         bool pg_updated = false;
324         unsigned long flags;
325
326         group_id = scsi_vpd_tpg_id(sdev, &rel_port);
327         if (group_id < 0) {
328                 /*
329                  * Internal error; TPGS supported but required
330                  * VPD identification descriptors not present.
331                  * Disable ALUA support
332                  */
333                 sdev_printk(KERN_INFO, sdev,
334                             "%s: No target port descriptors found\n",
335                             ALUA_DH_NAME);
336                 return SCSI_DH_DEV_UNSUPP;
337         }
338
339         pg = alua_alloc_pg(sdev, group_id, tpgs);
340         if (IS_ERR(pg)) {
341                 if (PTR_ERR(pg) == -ENOMEM)
342                         return SCSI_DH_NOMEM;
343                 return SCSI_DH_DEV_UNSUPP;
344         }
345         if (pg->device_id_len)
346                 sdev_printk(KERN_INFO, sdev,
347                             "%s: device %s port group %x rel port %x\n",
348                             ALUA_DH_NAME, pg->device_id_str,
349                             group_id, rel_port);
350         else
351                 sdev_printk(KERN_INFO, sdev,
352                             "%s: port group %x rel port %x\n",
353                             ALUA_DH_NAME, group_id, rel_port);
354
355         /* Check for existing port group references */
356         spin_lock(&h->pg_lock);
357         old_pg = rcu_dereference_protected(h->pg, lockdep_is_held(&h->pg_lock));
358         if (old_pg != pg) {
359                 /* port group has changed. Update to new port group */
360                 if (h->pg) {
361                         spin_lock_irqsave(&old_pg->lock, flags);
362                         list_del_rcu(&h->node);
363                         spin_unlock_irqrestore(&old_pg->lock, flags);
364                 }
365                 rcu_assign_pointer(h->pg, pg);
366                 pg_updated = true;
367         }
368
369         spin_lock_irqsave(&pg->lock, flags);
370         if (pg_updated)
371                 list_add_rcu(&h->node, &pg->dh_list);
372         spin_unlock_irqrestore(&pg->lock, flags);
373
374         alua_rtpg_queue(rcu_dereference_protected(h->pg,
375                                                   lockdep_is_held(&h->pg_lock)),
376                         sdev, NULL, true);
377         spin_unlock(&h->pg_lock);
378
379         if (old_pg)
380                 kref_put(&old_pg->kref, release_port_group);
381
382         return SCSI_DH_OK;
383 }
384
385 static char print_alua_state(unsigned char state)
386 {
387         switch (state) {
388         case SCSI_ACCESS_STATE_OPTIMAL:
389                 return 'A';
390         case SCSI_ACCESS_STATE_ACTIVE:
391                 return 'N';
392         case SCSI_ACCESS_STATE_STANDBY:
393                 return 'S';
394         case SCSI_ACCESS_STATE_UNAVAILABLE:
395                 return 'U';
396         case SCSI_ACCESS_STATE_LBA:
397                 return 'L';
398         case SCSI_ACCESS_STATE_OFFLINE:
399                 return 'O';
400         case SCSI_ACCESS_STATE_TRANSITIONING:
401                 return 'T';
402         default:
403                 return 'X';
404         }
405 }
406
407 static int alua_check_sense(struct scsi_device *sdev,
408                             struct scsi_sense_hdr *sense_hdr)
409 {
410         switch (sense_hdr->sense_key) {
411         case NOT_READY:
412                 if (sense_hdr->asc == 0x04 && sense_hdr->ascq == 0x0a) {
413                         /*
414                          * LUN Not Accessible - ALUA state transition
415                          */
416                         alua_check(sdev, false);
417                         return NEEDS_RETRY;
418                 }
419                 break;
420         case UNIT_ATTENTION:
421                 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x00) {
422                         /*
423                          * Power On, Reset, or Bus Device Reset.
424                          * Might have obscured a state transition,
425                          * so schedule a recheck.
426                          */
427                         alua_check(sdev, true);
428                         return ADD_TO_MLQUEUE;
429                 }
430                 if (sense_hdr->asc == 0x29 && sense_hdr->ascq == 0x04)
431                         /*
432                          * Device internal reset
433                          */
434                         return ADD_TO_MLQUEUE;
435                 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x01)
436                         /*
437                          * Mode Parameters Changed
438                          */
439                         return ADD_TO_MLQUEUE;
440                 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x06) {
441                         /*
442                          * ALUA state changed
443                          */
444                         alua_check(sdev, true);
445                         return ADD_TO_MLQUEUE;
446                 }
447                 if (sense_hdr->asc == 0x2a && sense_hdr->ascq == 0x07) {
448                         /*
449                          * Implicit ALUA state transition failed
450                          */
451                         alua_check(sdev, true);
452                         return ADD_TO_MLQUEUE;
453                 }
454                 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x03)
455                         /*
456                          * Inquiry data has changed
457                          */
458                         return ADD_TO_MLQUEUE;
459                 if (sense_hdr->asc == 0x3f && sense_hdr->ascq == 0x0e)
460                         /*
461                          * REPORTED_LUNS_DATA_HAS_CHANGED is reported
462                          * when switching controllers on targets like
463                          * Intel Multi-Flex. We can just retry.
464                          */
465                         return ADD_TO_MLQUEUE;
466                 break;
467         }
468
469         return SCSI_RETURN_NOT_HANDLED;
470 }
471
472 /*
473  * alua_tur - Send a TEST UNIT READY
474  * @sdev: device to which the TEST UNIT READY command should be send
475  *
476  * Send a TEST UNIT READY to @sdev to figure out the device state
477  * Returns SCSI_DH_RETRY if the sense code is NOT READY/ALUA TRANSITIONING,
478  * SCSI_DH_OK if no error occurred, and SCSI_DH_IO otherwise.
479  */
480 static int alua_tur(struct scsi_device *sdev)
481 {
482         struct scsi_sense_hdr sense_hdr;
483         int retval;
484
485         retval = scsi_test_unit_ready(sdev, ALUA_FAILOVER_TIMEOUT * HZ,
486                                       ALUA_FAILOVER_RETRIES, &sense_hdr);
487         if (sense_hdr.sense_key == NOT_READY &&
488             sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a)
489                 return SCSI_DH_RETRY;
490         else if (retval)
491                 return SCSI_DH_IO;
492         else
493                 return SCSI_DH_OK;
494 }
495
496 /*
497  * alua_rtpg - Evaluate REPORT TARGET GROUP STATES
498  * @sdev: the device to be evaluated.
499  *
500  * Evaluate the Target Port Group State.
501  * Returns SCSI_DH_DEV_OFFLINED if the path is
502  * found to be unusable.
503  */
504 static int alua_rtpg(struct scsi_device *sdev, struct alua_port_group *pg)
505 {
506         struct scsi_sense_hdr sense_hdr;
507         struct alua_port_group *tmp_pg;
508         int len, k, off, bufflen = ALUA_RTPG_SIZE;
509         unsigned char *desc, *buff;
510         unsigned err, retval;
511         unsigned int tpg_desc_tbl_off;
512         unsigned char orig_transition_tmo;
513         unsigned long flags;
514
515         if (!pg->expiry) {
516                 unsigned long transition_tmo = ALUA_FAILOVER_TIMEOUT * HZ;
517
518                 if (pg->transition_tmo)
519                         transition_tmo = pg->transition_tmo * HZ;
520
521                 pg->expiry = round_jiffies_up(jiffies + transition_tmo);
522         }
523
524         buff = kzalloc(bufflen, GFP_KERNEL);
525         if (!buff)
526                 return SCSI_DH_DEV_TEMP_BUSY;
527
528  retry:
529         err = 0;
530         retval = submit_rtpg(sdev, buff, bufflen, &sense_hdr, pg->flags);
531
532         if (retval) {
533                 /*
534                  * Some (broken) implementations have a habit of returning
535                  * an error during things like firmware update etc.
536                  * But if the target only supports active/optimized there's
537                  * not much we can do; it's not that we can switch paths
538                  * or anything.
539                  * So ignore any errors to avoid spurious failures during
540                  * path failover.
541                  */
542                 if ((pg->valid_states & ~TPGS_SUPPORT_OPTIMIZED) == 0) {
543                         sdev_printk(KERN_INFO, sdev,
544                                     "%s: ignoring rtpg result %d\n",
545                                     ALUA_DH_NAME, retval);
546                         kfree(buff);
547                         return SCSI_DH_OK;
548                 }
549                 if (!scsi_sense_valid(&sense_hdr)) {
550                         sdev_printk(KERN_INFO, sdev,
551                                     "%s: rtpg failed, result %d\n",
552                                     ALUA_DH_NAME, retval);
553                         kfree(buff);
554                         if (driver_byte(retval) == DRIVER_ERROR)
555                                 return SCSI_DH_DEV_TEMP_BUSY;
556                         return SCSI_DH_IO;
557                 }
558
559                 /*
560                  * submit_rtpg() has failed on existing arrays
561                  * when requesting extended header info, and
562                  * the array doesn't support extended headers,
563                  * even though it shouldn't according to T10.
564                  * The retry without rtpg_ext_hdr_req set
565                  * handles this.
566                  */
567                 if (!(pg->flags & ALUA_RTPG_EXT_HDR_UNSUPP) &&
568                     sense_hdr.sense_key == ILLEGAL_REQUEST &&
569                     sense_hdr.asc == 0x24 && sense_hdr.ascq == 0) {
570                         pg->flags |= ALUA_RTPG_EXT_HDR_UNSUPP;
571                         goto retry;
572                 }
573                 /*
574                  * Retry on ALUA state transition or if any
575                  * UNIT ATTENTION occurred.
576                  */
577                 if (sense_hdr.sense_key == NOT_READY &&
578                     sense_hdr.asc == 0x04 && sense_hdr.ascq == 0x0a)
579                         err = SCSI_DH_RETRY;
580                 else if (sense_hdr.sense_key == UNIT_ATTENTION)
581                         err = SCSI_DH_RETRY;
582                 if (err == SCSI_DH_RETRY &&
583                     pg->expiry != 0 && time_before(jiffies, pg->expiry)) {
584                         sdev_printk(KERN_ERR, sdev, "%s: rtpg retry\n",
585                                     ALUA_DH_NAME);
586                         scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
587                         kfree(buff);
588                         return err;
589                 }
590                 sdev_printk(KERN_ERR, sdev, "%s: rtpg failed\n",
591                             ALUA_DH_NAME);
592                 scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
593                 kfree(buff);
594                 pg->expiry = 0;
595                 return SCSI_DH_IO;
596         }
597
598         len = get_unaligned_be32(&buff[0]) + 4;
599
600         if (len > bufflen) {
601                 /* Resubmit with the correct length */
602                 kfree(buff);
603                 bufflen = len;
604                 buff = kmalloc(bufflen, GFP_KERNEL);
605                 if (!buff) {
606                         sdev_printk(KERN_WARNING, sdev,
607                                     "%s: kmalloc buffer failed\n",__func__);
608                         /* Temporary failure, bypass */
609                         pg->expiry = 0;
610                         return SCSI_DH_DEV_TEMP_BUSY;
611                 }
612                 goto retry;
613         }
614
615         orig_transition_tmo = pg->transition_tmo;
616         if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR && buff[5] != 0)
617                 pg->transition_tmo = buff[5];
618         else
619                 pg->transition_tmo = ALUA_FAILOVER_TIMEOUT;
620
621         if (orig_transition_tmo != pg->transition_tmo) {
622                 sdev_printk(KERN_INFO, sdev,
623                             "%s: transition timeout set to %d seconds\n",
624                             ALUA_DH_NAME, pg->transition_tmo);
625                 pg->expiry = jiffies + pg->transition_tmo * HZ;
626         }
627
628         if ((buff[4] & RTPG_FMT_MASK) == RTPG_FMT_EXT_HDR)
629                 tpg_desc_tbl_off = 8;
630         else
631                 tpg_desc_tbl_off = 4;
632
633         for (k = tpg_desc_tbl_off, desc = buff + tpg_desc_tbl_off;
634              k < len;
635              k += off, desc += off) {
636                 u16 group_id = get_unaligned_be16(&desc[2]);
637
638                 spin_lock_irqsave(&port_group_lock, flags);
639                 tmp_pg = alua_find_get_pg(pg->device_id_str, pg->device_id_len,
640                                           group_id);
641                 spin_unlock_irqrestore(&port_group_lock, flags);
642                 if (tmp_pg) {
643                         if (spin_trylock_irqsave(&tmp_pg->lock, flags)) {
644                                 if ((tmp_pg == pg) ||
645                                     !(tmp_pg->flags & ALUA_PG_RUNNING)) {
646                                         struct alua_dh_data *h;
647
648                                         tmp_pg->state = desc[0] & 0x0f;
649                                         tmp_pg->pref = desc[0] >> 7;
650                                         rcu_read_lock();
651                                         list_for_each_entry_rcu(h,
652                                                 &tmp_pg->dh_list, node) {
653                                                 /* h->sdev should always be valid */
654                                                 BUG_ON(!h->sdev);
655                                                 h->sdev->access_state = desc[0];
656                                         }
657                                         rcu_read_unlock();
658                                 }
659                                 if (tmp_pg == pg)
660                                         tmp_pg->valid_states = desc[1];
661                                 spin_unlock_irqrestore(&tmp_pg->lock, flags);
662                         }
663                         kref_put(&tmp_pg->kref, release_port_group);
664                 }
665                 off = 8 + (desc[7] * 4);
666         }
667
668         spin_lock_irqsave(&pg->lock, flags);
669         sdev_printk(KERN_INFO, sdev,
670                     "%s: port group %02x state %c %s supports %c%c%c%c%c%c%c\n",
671                     ALUA_DH_NAME, pg->group_id, print_alua_state(pg->state),
672                     pg->pref ? "preferred" : "non-preferred",
673                     pg->valid_states&TPGS_SUPPORT_TRANSITION?'T':'t',
674                     pg->valid_states&TPGS_SUPPORT_OFFLINE?'O':'o',
675                     pg->valid_states&TPGS_SUPPORT_LBA_DEPENDENT?'L':'l',
676                     pg->valid_states&TPGS_SUPPORT_UNAVAILABLE?'U':'u',
677                     pg->valid_states&TPGS_SUPPORT_STANDBY?'S':'s',
678                     pg->valid_states&TPGS_SUPPORT_NONOPTIMIZED?'N':'n',
679                     pg->valid_states&TPGS_SUPPORT_OPTIMIZED?'A':'a');
680
681         switch (pg->state) {
682         case SCSI_ACCESS_STATE_TRANSITIONING:
683                 if (time_before(jiffies, pg->expiry)) {
684                         /* State transition, retry */
685                         pg->interval = 2;
686                         err = SCSI_DH_RETRY;
687                 } else {
688                         struct alua_dh_data *h;
689
690                         /* Transitioning time exceeded, set port to standby */
691                         err = SCSI_DH_IO;
692                         pg->state = SCSI_ACCESS_STATE_STANDBY;
693                         pg->expiry = 0;
694                         rcu_read_lock();
695                         list_for_each_entry_rcu(h, &pg->dh_list, node) {
696                                 BUG_ON(!h->sdev);
697                                 h->sdev->access_state =
698                                         (pg->state & SCSI_ACCESS_STATE_MASK);
699                                 if (pg->pref)
700                                         h->sdev->access_state |=
701                                                 SCSI_ACCESS_STATE_PREFERRED;
702                         }
703                         rcu_read_unlock();
704                 }
705                 break;
706         case SCSI_ACCESS_STATE_OFFLINE:
707                 /* Path unusable */
708                 err = SCSI_DH_DEV_OFFLINED;
709                 pg->expiry = 0;
710                 break;
711         default:
712                 /* Useable path if active */
713                 err = SCSI_DH_OK;
714                 pg->expiry = 0;
715                 break;
716         }
717         spin_unlock_irqrestore(&pg->lock, flags);
718         kfree(buff);
719         return err;
720 }
721
722 /*
723  * alua_stpg - Issue a SET TARGET PORT GROUP command
724  *
725  * Issue a SET TARGET PORT GROUP command and evaluate the
726  * response. Returns SCSI_DH_RETRY per default to trigger
727  * a re-evaluation of the target group state or SCSI_DH_OK
728  * if no further action needs to be taken.
729  */
730 static unsigned alua_stpg(struct scsi_device *sdev, struct alua_port_group *pg)
731 {
732         int retval;
733         struct scsi_sense_hdr sense_hdr;
734
735         if (!(pg->tpgs & TPGS_MODE_EXPLICIT)) {
736                 /* Only implicit ALUA supported, retry */
737                 return SCSI_DH_RETRY;
738         }
739         switch (pg->state) {
740         case SCSI_ACCESS_STATE_OPTIMAL:
741                 return SCSI_DH_OK;
742         case SCSI_ACCESS_STATE_ACTIVE:
743                 if ((pg->flags & ALUA_OPTIMIZE_STPG) &&
744                     !pg->pref &&
745                     (pg->tpgs & TPGS_MODE_IMPLICIT))
746                         return SCSI_DH_OK;
747                 break;
748         case SCSI_ACCESS_STATE_STANDBY:
749         case SCSI_ACCESS_STATE_UNAVAILABLE:
750                 break;
751         case SCSI_ACCESS_STATE_OFFLINE:
752                 return SCSI_DH_IO;
753         case SCSI_ACCESS_STATE_TRANSITIONING:
754                 break;
755         default:
756                 sdev_printk(KERN_INFO, sdev,
757                             "%s: stpg failed, unhandled TPGS state %d",
758                             ALUA_DH_NAME, pg->state);
759                 return SCSI_DH_NOSYS;
760         }
761         retval = submit_stpg(sdev, pg->group_id, &sense_hdr);
762
763         if (retval) {
764                 if (!scsi_sense_valid(&sense_hdr)) {
765                         sdev_printk(KERN_INFO, sdev,
766                                     "%s: stpg failed, result %d",
767                                     ALUA_DH_NAME, retval);
768                         if (driver_byte(retval) == DRIVER_ERROR)
769                                 return SCSI_DH_DEV_TEMP_BUSY;
770                 } else {
771                         sdev_printk(KERN_INFO, sdev, "%s: stpg failed\n",
772                                     ALUA_DH_NAME);
773                         scsi_print_sense_hdr(sdev, ALUA_DH_NAME, &sense_hdr);
774                 }
775         }
776         /* Retry RTPG */
777         return SCSI_DH_RETRY;
778 }
779
780 static void alua_rtpg_work(struct work_struct *work)
781 {
782         struct alua_port_group *pg =
783                 container_of(work, struct alua_port_group, rtpg_work.work);
784         struct scsi_device *sdev;
785         LIST_HEAD(qdata_list);
786         int err = SCSI_DH_OK;
787         struct alua_queue_data *qdata, *tmp;
788         unsigned long flags;
789
790         spin_lock_irqsave(&pg->lock, flags);
791         sdev = pg->rtpg_sdev;
792         if (!sdev) {
793                 WARN_ON(pg->flags & ALUA_PG_RUN_RTPG);
794                 WARN_ON(pg->flags & ALUA_PG_RUN_STPG);
795                 spin_unlock_irqrestore(&pg->lock, flags);
796                 kref_put(&pg->kref, release_port_group);
797                 return;
798         }
799         pg->flags |= ALUA_PG_RUNNING;
800         if (pg->flags & ALUA_PG_RUN_RTPG) {
801                 int state = pg->state;
802
803                 pg->flags &= ~ALUA_PG_RUN_RTPG;
804                 spin_unlock_irqrestore(&pg->lock, flags);
805                 if (state == SCSI_ACCESS_STATE_TRANSITIONING) {
806                         if (alua_tur(sdev) == SCSI_DH_RETRY) {
807                                 spin_lock_irqsave(&pg->lock, flags);
808                                 pg->flags &= ~ALUA_PG_RUNNING;
809                                 pg->flags |= ALUA_PG_RUN_RTPG;
810                                 spin_unlock_irqrestore(&pg->lock, flags);
811                                 queue_delayed_work(kaluad_wq, &pg->rtpg_work,
812                                                    pg->interval * HZ);
813                                 return;
814                         }
815                         /* Send RTPG on failure or if TUR indicates SUCCESS */
816                 }
817                 err = alua_rtpg(sdev, pg);
818                 spin_lock_irqsave(&pg->lock, flags);
819                 if (err == SCSI_DH_RETRY || pg->flags & ALUA_PG_RUN_RTPG) {
820                         pg->flags &= ~ALUA_PG_RUNNING;
821                         pg->flags |= ALUA_PG_RUN_RTPG;
822                         spin_unlock_irqrestore(&pg->lock, flags);
823                         queue_delayed_work(kaluad_wq, &pg->rtpg_work,
824                                            pg->interval * HZ);
825                         return;
826                 }
827                 if (err != SCSI_DH_OK)
828                         pg->flags &= ~ALUA_PG_RUN_STPG;
829         }
830         if (pg->flags & ALUA_PG_RUN_STPG) {
831                 pg->flags &= ~ALUA_PG_RUN_STPG;
832                 spin_unlock_irqrestore(&pg->lock, flags);
833                 err = alua_stpg(sdev, pg);
834                 spin_lock_irqsave(&pg->lock, flags);
835                 if (err == SCSI_DH_RETRY || pg->flags & ALUA_PG_RUN_RTPG) {
836                         pg->flags |= ALUA_PG_RUN_RTPG;
837                         pg->interval = 0;
838                         pg->flags &= ~ALUA_PG_RUNNING;
839                         spin_unlock_irqrestore(&pg->lock, flags);
840                         queue_delayed_work(kaluad_wq, &pg->rtpg_work,
841                                            pg->interval * HZ);
842                         return;
843                 }
844         }
845
846         list_splice_init(&pg->rtpg_list, &qdata_list);
847         pg->rtpg_sdev = NULL;
848         spin_unlock_irqrestore(&pg->lock, flags);
849
850         list_for_each_entry_safe(qdata, tmp, &qdata_list, entry) {
851                 list_del(&qdata->entry);
852                 if (qdata->callback_fn)
853                         qdata->callback_fn(qdata->callback_data, err);
854                 kfree(qdata);
855         }
856         spin_lock_irqsave(&pg->lock, flags);
857         pg->flags &= ~ALUA_PG_RUNNING;
858         spin_unlock_irqrestore(&pg->lock, flags);
859         scsi_device_put(sdev);
860         kref_put(&pg->kref, release_port_group);
861 }
862
863 /**
864  * alua_rtpg_queue() - cause RTPG to be submitted asynchronously
865  * @pg: ALUA port group associated with @sdev.
866  * @sdev: SCSI device for which to submit an RTPG.
867  * @qdata: Information about the callback to invoke after the RTPG.
868  * @force: Whether or not to submit an RTPG if a work item that will submit an
869  *         RTPG already has been scheduled.
870  *
871  * Returns true if and only if alua_rtpg_work() will be called asynchronously.
872  * That function is responsible for calling @qdata->fn().
873  */
874 static bool alua_rtpg_queue(struct alua_port_group *pg,
875                             struct scsi_device *sdev,
876                             struct alua_queue_data *qdata, bool force)
877 {
878         int start_queue = 0;
879         unsigned long flags;
880         if (WARN_ON_ONCE(!pg) || scsi_device_get(sdev))
881                 return false;
882
883         spin_lock_irqsave(&pg->lock, flags);
884         if (qdata) {
885                 list_add_tail(&qdata->entry, &pg->rtpg_list);
886                 pg->flags |= ALUA_PG_RUN_STPG;
887                 force = true;
888         }
889         if (pg->rtpg_sdev == NULL) {
890                 pg->interval = 0;
891                 pg->flags |= ALUA_PG_RUN_RTPG;
892                 kref_get(&pg->kref);
893                 pg->rtpg_sdev = sdev;
894                 start_queue = 1;
895         } else if (!(pg->flags & ALUA_PG_RUN_RTPG) && force) {
896                 pg->flags |= ALUA_PG_RUN_RTPG;
897                 /* Do not queue if the worker is already running */
898                 if (!(pg->flags & ALUA_PG_RUNNING)) {
899                         kref_get(&pg->kref);
900                         start_queue = 1;
901                 }
902         }
903
904         spin_unlock_irqrestore(&pg->lock, flags);
905
906         if (start_queue) {
907                 if (queue_delayed_work(kaluad_wq, &pg->rtpg_work,
908                                 msecs_to_jiffies(ALUA_RTPG_DELAY_MSECS)))
909                         sdev = NULL;
910                 else
911                         kref_put(&pg->kref, release_port_group);
912         }
913         if (sdev)
914                 scsi_device_put(sdev);
915
916         return true;
917 }
918
919 /*
920  * alua_initialize - Initialize ALUA state
921  * @sdev: the device to be initialized
922  *
923  * For the prep_fn to work correctly we have
924  * to initialize the ALUA state for the device.
925  */
926 static int alua_initialize(struct scsi_device *sdev, struct alua_dh_data *h)
927 {
928         int err = SCSI_DH_DEV_UNSUPP, tpgs;
929
930         mutex_lock(&h->init_mutex);
931         tpgs = alua_check_tpgs(sdev);
932         if (tpgs != TPGS_MODE_NONE)
933                 err = alua_check_vpd(sdev, h, tpgs);
934         h->init_error = err;
935         mutex_unlock(&h->init_mutex);
936         return err;
937 }
938 /*
939  * alua_set_params - set/unset the optimize flag
940  * @sdev: device on the path to be activated
941  * params - parameters in the following format
942  *      "no_of_params\0param1\0param2\0param3\0...\0"
943  * For example, to set the flag pass the following parameters
944  * from multipath.conf
945  *     hardware_handler        "2 alua 1"
946  */
947 static int alua_set_params(struct scsi_device *sdev, const char *params)
948 {
949         struct alua_dh_data *h = sdev->handler_data;
950         struct alua_port_group *pg = NULL;
951         unsigned int optimize = 0, argc;
952         const char *p = params;
953         int result = SCSI_DH_OK;
954         unsigned long flags;
955
956         if ((sscanf(params, "%u", &argc) != 1) || (argc != 1))
957                 return -EINVAL;
958
959         while (*p++)
960                 ;
961         if ((sscanf(p, "%u", &optimize) != 1) || (optimize > 1))
962                 return -EINVAL;
963
964         rcu_read_lock();
965         pg = rcu_dereference(h->pg);
966         if (!pg) {
967                 rcu_read_unlock();
968                 return -ENXIO;
969         }
970         spin_lock_irqsave(&pg->lock, flags);
971         if (optimize)
972                 pg->flags |= ALUA_OPTIMIZE_STPG;
973         else
974                 pg->flags &= ~ALUA_OPTIMIZE_STPG;
975         spin_unlock_irqrestore(&pg->lock, flags);
976         rcu_read_unlock();
977
978         return result;
979 }
980
981 /*
982  * alua_activate - activate a path
983  * @sdev: device on the path to be activated
984  *
985  * We're currently switching the port group to be activated only and
986  * let the array figure out the rest.
987  * There may be other arrays which require us to switch all port groups
988  * based on a certain policy. But until we actually encounter them it
989  * should be okay.
990  */
991 static int alua_activate(struct scsi_device *sdev,
992                         activate_complete fn, void *data)
993 {
994         struct alua_dh_data *h = sdev->handler_data;
995         int err = SCSI_DH_OK;
996         struct alua_queue_data *qdata;
997         struct alua_port_group *pg;
998
999         qdata = kzalloc(sizeof(*qdata), GFP_KERNEL);
1000         if (!qdata) {
1001                 err = SCSI_DH_RES_TEMP_UNAVAIL;
1002                 goto out;
1003         }
1004         qdata->callback_fn = fn;
1005         qdata->callback_data = data;
1006
1007         mutex_lock(&h->init_mutex);
1008         rcu_read_lock();
1009         pg = rcu_dereference(h->pg);
1010         if (!pg || !kref_get_unless_zero(&pg->kref)) {
1011                 rcu_read_unlock();
1012                 kfree(qdata);
1013                 err = h->init_error;
1014                 mutex_unlock(&h->init_mutex);
1015                 goto out;
1016         }
1017         rcu_read_unlock();
1018         mutex_unlock(&h->init_mutex);
1019
1020         if (alua_rtpg_queue(pg, sdev, qdata, true))
1021                 fn = NULL;
1022         else
1023                 err = SCSI_DH_DEV_OFFLINED;
1024         kref_put(&pg->kref, release_port_group);
1025 out:
1026         if (fn)
1027                 fn(data, err);
1028         return 0;
1029 }
1030
1031 /*
1032  * alua_check - check path status
1033  * @sdev: device on the path to be checked
1034  *
1035  * Check the device status
1036  */
1037 static void alua_check(struct scsi_device *sdev, bool force)
1038 {
1039         struct alua_dh_data *h = sdev->handler_data;
1040         struct alua_port_group *pg;
1041
1042         rcu_read_lock();
1043         pg = rcu_dereference(h->pg);
1044         if (!pg || !kref_get_unless_zero(&pg->kref)) {
1045                 rcu_read_unlock();
1046                 return;
1047         }
1048         rcu_read_unlock();
1049
1050         alua_rtpg_queue(pg, sdev, NULL, force);
1051         kref_put(&pg->kref, release_port_group);
1052 }
1053
1054 /*
1055  * alua_prep_fn - request callback
1056  *
1057  * Fail I/O to all paths not in state
1058  * active/optimized or active/non-optimized.
1059  */
1060 static blk_status_t alua_prep_fn(struct scsi_device *sdev, struct request *req)
1061 {
1062         struct alua_dh_data *h = sdev->handler_data;
1063         struct alua_port_group *pg;
1064         unsigned char state = SCSI_ACCESS_STATE_OPTIMAL;
1065
1066         rcu_read_lock();
1067         pg = rcu_dereference(h->pg);
1068         if (pg)
1069                 state = pg->state;
1070         rcu_read_unlock();
1071
1072         switch (state) {
1073         case SCSI_ACCESS_STATE_OPTIMAL:
1074         case SCSI_ACCESS_STATE_ACTIVE:
1075         case SCSI_ACCESS_STATE_LBA:
1076                 return BLK_STS_OK;
1077         case SCSI_ACCESS_STATE_TRANSITIONING:
1078                 return BLK_STS_RESOURCE;
1079         default:
1080                 req->rq_flags |= RQF_QUIET;
1081                 return BLK_STS_IOERR;
1082         }
1083 }
1084
1085 static void alua_rescan(struct scsi_device *sdev)
1086 {
1087         struct alua_dh_data *h = sdev->handler_data;
1088
1089         alua_initialize(sdev, h);
1090 }
1091
1092 /*
1093  * alua_bus_attach - Attach device handler
1094  * @sdev: device to be attached to
1095  */
1096 static int alua_bus_attach(struct scsi_device *sdev)
1097 {
1098         struct alua_dh_data *h;
1099         int err;
1100
1101         h = kzalloc(sizeof(*h) , GFP_KERNEL);
1102         if (!h)
1103                 return SCSI_DH_NOMEM;
1104         spin_lock_init(&h->pg_lock);
1105         rcu_assign_pointer(h->pg, NULL);
1106         h->init_error = SCSI_DH_OK;
1107         h->sdev = sdev;
1108         INIT_LIST_HEAD(&h->node);
1109
1110         mutex_init(&h->init_mutex);
1111         err = alua_initialize(sdev, h);
1112         if (err != SCSI_DH_OK && err != SCSI_DH_DEV_OFFLINED)
1113                 goto failed;
1114
1115         sdev->handler_data = h;
1116         return SCSI_DH_OK;
1117 failed:
1118         kfree(h);
1119         return err;
1120 }
1121
1122 /*
1123  * alua_bus_detach - Detach device handler
1124  * @sdev: device to be detached from
1125  */
1126 static void alua_bus_detach(struct scsi_device *sdev)
1127 {
1128         struct alua_dh_data *h = sdev->handler_data;
1129         struct alua_port_group *pg;
1130
1131         spin_lock(&h->pg_lock);
1132         pg = rcu_dereference_protected(h->pg, lockdep_is_held(&h->pg_lock));
1133         rcu_assign_pointer(h->pg, NULL);
1134         h->sdev = NULL;
1135         spin_unlock(&h->pg_lock);
1136         if (pg) {
1137                 spin_lock_irq(&pg->lock);
1138                 list_del_rcu(&h->node);
1139                 spin_unlock_irq(&pg->lock);
1140                 kref_put(&pg->kref, release_port_group);
1141         }
1142         sdev->handler_data = NULL;
1143         kfree(h);
1144 }
1145
1146 static struct scsi_device_handler alua_dh = {
1147         .name = ALUA_DH_NAME,
1148         .module = THIS_MODULE,
1149         .attach = alua_bus_attach,
1150         .detach = alua_bus_detach,
1151         .prep_fn = alua_prep_fn,
1152         .check_sense = alua_check_sense,
1153         .activate = alua_activate,
1154         .rescan = alua_rescan,
1155         .set_params = alua_set_params,
1156 };
1157
1158 static int __init alua_init(void)
1159 {
1160         int r;
1161
1162         kaluad_wq = alloc_workqueue("kaluad", WQ_MEM_RECLAIM, 0);
1163         if (!kaluad_wq)
1164                 return -ENOMEM;
1165
1166         r = scsi_register_device_handler(&alua_dh);
1167         if (r != 0) {
1168                 printk(KERN_ERR "%s: Failed to register scsi device handler",
1169                         ALUA_DH_NAME);
1170                 destroy_workqueue(kaluad_wq);
1171         }
1172         return r;
1173 }
1174
1175 static void __exit alua_exit(void)
1176 {
1177         scsi_unregister_device_handler(&alua_dh);
1178         destroy_workqueue(kaluad_wq);
1179 }
1180
1181 module_init(alua_init);
1182 module_exit(alua_exit);
1183
1184 MODULE_DESCRIPTION("DM Multipath ALUA support");
1185 MODULE_AUTHOR("Hannes Reinecke <hare@suse.de>");
1186 MODULE_LICENSE("GPL");
1187 MODULE_VERSION(ALUA_DH_VER);