Merge tag 'for-4.15/dm-changes-2' of git://git.kernel.org/pub/scm/linux/kernel/git...
[muen/linux.git] / drivers / md / dm-table.c
1 /*
2  * Copyright (C) 2001 Sistina Software (UK) Limited.
3  * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4  *
5  * This file is released under the GPL.
6  */
7
8 #include "dm-core.h"
9
10 #include <linux/module.h>
11 #include <linux/vmalloc.h>
12 #include <linux/blkdev.h>
13 #include <linux/namei.h>
14 #include <linux/ctype.h>
15 #include <linux/string.h>
16 #include <linux/slab.h>
17 #include <linux/interrupt.h>
18 #include <linux/mutex.h>
19 #include <linux/delay.h>
20 #include <linux/atomic.h>
21 #include <linux/blk-mq.h>
22 #include <linux/mount.h>
23 #include <linux/dax.h>
24
25 #define DM_MSG_PREFIX "table"
26
27 #define MAX_DEPTH 16
28 #define NODE_SIZE L1_CACHE_BYTES
29 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
30 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
31
32 struct dm_table {
33         struct mapped_device *md;
34         enum dm_queue_mode type;
35
36         /* btree table */
37         unsigned int depth;
38         unsigned int counts[MAX_DEPTH]; /* in nodes */
39         sector_t *index[MAX_DEPTH];
40
41         unsigned int num_targets;
42         unsigned int num_allocated;
43         sector_t *highs;
44         struct dm_target *targets;
45
46         struct target_type *immutable_target_type;
47
48         bool integrity_supported:1;
49         bool singleton:1;
50         bool all_blk_mq:1;
51         unsigned integrity_added:1;
52
53         /*
54          * Indicates the rw permissions for the new logical
55          * device.  This should be a combination of FMODE_READ
56          * and FMODE_WRITE.
57          */
58         fmode_t mode;
59
60         /* a list of devices used by this table */
61         struct list_head devices;
62
63         /* events get handed up using this callback */
64         void (*event_fn)(void *);
65         void *event_context;
66
67         struct dm_md_mempools *mempools;
68
69         struct list_head target_callbacks;
70 };
71
72 /*
73  * Similar to ceiling(log_size(n))
74  */
75 static unsigned int int_log(unsigned int n, unsigned int base)
76 {
77         int result = 0;
78
79         while (n > 1) {
80                 n = dm_div_up(n, base);
81                 result++;
82         }
83
84         return result;
85 }
86
87 /*
88  * Calculate the index of the child node of the n'th node k'th key.
89  */
90 static inline unsigned int get_child(unsigned int n, unsigned int k)
91 {
92         return (n * CHILDREN_PER_NODE) + k;
93 }
94
95 /*
96  * Return the n'th node of level l from table t.
97  */
98 static inline sector_t *get_node(struct dm_table *t,
99                                  unsigned int l, unsigned int n)
100 {
101         return t->index[l] + (n * KEYS_PER_NODE);
102 }
103
104 /*
105  * Return the highest key that you could lookup from the n'th
106  * node on level l of the btree.
107  */
108 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
109 {
110         for (; l < t->depth - 1; l++)
111                 n = get_child(n, CHILDREN_PER_NODE - 1);
112
113         if (n >= t->counts[l])
114                 return (sector_t) - 1;
115
116         return get_node(t, l, n)[KEYS_PER_NODE - 1];
117 }
118
119 /*
120  * Fills in a level of the btree based on the highs of the level
121  * below it.
122  */
123 static int setup_btree_index(unsigned int l, struct dm_table *t)
124 {
125         unsigned int n, k;
126         sector_t *node;
127
128         for (n = 0U; n < t->counts[l]; n++) {
129                 node = get_node(t, l, n);
130
131                 for (k = 0U; k < KEYS_PER_NODE; k++)
132                         node[k] = high(t, l + 1, get_child(n, k));
133         }
134
135         return 0;
136 }
137
138 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
139 {
140         unsigned long size;
141         void *addr;
142
143         /*
144          * Check that we're not going to overflow.
145          */
146         if (nmemb > (ULONG_MAX / elem_size))
147                 return NULL;
148
149         size = nmemb * elem_size;
150         addr = vzalloc(size);
151
152         return addr;
153 }
154 EXPORT_SYMBOL(dm_vcalloc);
155
156 /*
157  * highs, and targets are managed as dynamic arrays during a
158  * table load.
159  */
160 static int alloc_targets(struct dm_table *t, unsigned int num)
161 {
162         sector_t *n_highs;
163         struct dm_target *n_targets;
164
165         /*
166          * Allocate both the target array and offset array at once.
167          * Append an empty entry to catch sectors beyond the end of
168          * the device.
169          */
170         n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
171                                           sizeof(sector_t));
172         if (!n_highs)
173                 return -ENOMEM;
174
175         n_targets = (struct dm_target *) (n_highs + num);
176
177         memset(n_highs, -1, sizeof(*n_highs) * num);
178         vfree(t->highs);
179
180         t->num_allocated = num;
181         t->highs = n_highs;
182         t->targets = n_targets;
183
184         return 0;
185 }
186
187 int dm_table_create(struct dm_table **result, fmode_t mode,
188                     unsigned num_targets, struct mapped_device *md)
189 {
190         struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
191
192         if (!t)
193                 return -ENOMEM;
194
195         INIT_LIST_HEAD(&t->devices);
196         INIT_LIST_HEAD(&t->target_callbacks);
197
198         if (!num_targets)
199                 num_targets = KEYS_PER_NODE;
200
201         num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
202
203         if (!num_targets) {
204                 kfree(t);
205                 return -ENOMEM;
206         }
207
208         if (alloc_targets(t, num_targets)) {
209                 kfree(t);
210                 return -ENOMEM;
211         }
212
213         t->type = DM_TYPE_NONE;
214         t->mode = mode;
215         t->md = md;
216         *result = t;
217         return 0;
218 }
219
220 static void free_devices(struct list_head *devices, struct mapped_device *md)
221 {
222         struct list_head *tmp, *next;
223
224         list_for_each_safe(tmp, next, devices) {
225                 struct dm_dev_internal *dd =
226                     list_entry(tmp, struct dm_dev_internal, list);
227                 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
228                        dm_device_name(md), dd->dm_dev->name);
229                 dm_put_table_device(md, dd->dm_dev);
230                 kfree(dd);
231         }
232 }
233
234 void dm_table_destroy(struct dm_table *t)
235 {
236         unsigned int i;
237
238         if (!t)
239                 return;
240
241         /* free the indexes */
242         if (t->depth >= 2)
243                 vfree(t->index[t->depth - 2]);
244
245         /* free the targets */
246         for (i = 0; i < t->num_targets; i++) {
247                 struct dm_target *tgt = t->targets + i;
248
249                 if (tgt->type->dtr)
250                         tgt->type->dtr(tgt);
251
252                 dm_put_target_type(tgt->type);
253         }
254
255         vfree(t->highs);
256
257         /* free the device list */
258         free_devices(&t->devices, t->md);
259
260         dm_free_md_mempools(t->mempools);
261
262         kfree(t);
263 }
264
265 /*
266  * See if we've already got a device in the list.
267  */
268 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
269 {
270         struct dm_dev_internal *dd;
271
272         list_for_each_entry (dd, l, list)
273                 if (dd->dm_dev->bdev->bd_dev == dev)
274                         return dd;
275
276         return NULL;
277 }
278
279 /*
280  * If possible, this checks an area of a destination device is invalid.
281  */
282 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
283                                   sector_t start, sector_t len, void *data)
284 {
285         struct request_queue *q;
286         struct queue_limits *limits = data;
287         struct block_device *bdev = dev->bdev;
288         sector_t dev_size =
289                 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
290         unsigned short logical_block_size_sectors =
291                 limits->logical_block_size >> SECTOR_SHIFT;
292         char b[BDEVNAME_SIZE];
293
294         /*
295          * Some devices exist without request functions,
296          * such as loop devices not yet bound to backing files.
297          * Forbid the use of such devices.
298          */
299         q = bdev_get_queue(bdev);
300         if (!q || !q->make_request_fn) {
301                 DMWARN("%s: %s is not yet initialised: "
302                        "start=%llu, len=%llu, dev_size=%llu",
303                        dm_device_name(ti->table->md), bdevname(bdev, b),
304                        (unsigned long long)start,
305                        (unsigned long long)len,
306                        (unsigned long long)dev_size);
307                 return 1;
308         }
309
310         if (!dev_size)
311                 return 0;
312
313         if ((start >= dev_size) || (start + len > dev_size)) {
314                 DMWARN("%s: %s too small for target: "
315                        "start=%llu, len=%llu, dev_size=%llu",
316                        dm_device_name(ti->table->md), bdevname(bdev, b),
317                        (unsigned long long)start,
318                        (unsigned long long)len,
319                        (unsigned long long)dev_size);
320                 return 1;
321         }
322
323         /*
324          * If the target is mapped to zoned block device(s), check
325          * that the zones are not partially mapped.
326          */
327         if (bdev_zoned_model(bdev) != BLK_ZONED_NONE) {
328                 unsigned int zone_sectors = bdev_zone_sectors(bdev);
329
330                 if (start & (zone_sectors - 1)) {
331                         DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s",
332                                dm_device_name(ti->table->md),
333                                (unsigned long long)start,
334                                zone_sectors, bdevname(bdev, b));
335                         return 1;
336                 }
337
338                 /*
339                  * Note: The last zone of a zoned block device may be smaller
340                  * than other zones. So for a target mapping the end of a
341                  * zoned block device with such a zone, len would not be zone
342                  * aligned. We do not allow such last smaller zone to be part
343                  * of the mapping here to ensure that mappings with multiple
344                  * devices do not end up with a smaller zone in the middle of
345                  * the sector range.
346                  */
347                 if (len & (zone_sectors - 1)) {
348                         DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s",
349                                dm_device_name(ti->table->md),
350                                (unsigned long long)len,
351                                zone_sectors, bdevname(bdev, b));
352                         return 1;
353                 }
354         }
355
356         if (logical_block_size_sectors <= 1)
357                 return 0;
358
359         if (start & (logical_block_size_sectors - 1)) {
360                 DMWARN("%s: start=%llu not aligned to h/w "
361                        "logical block size %u of %s",
362                        dm_device_name(ti->table->md),
363                        (unsigned long long)start,
364                        limits->logical_block_size, bdevname(bdev, b));
365                 return 1;
366         }
367
368         if (len & (logical_block_size_sectors - 1)) {
369                 DMWARN("%s: len=%llu not aligned to h/w "
370                        "logical block size %u of %s",
371                        dm_device_name(ti->table->md),
372                        (unsigned long long)len,
373                        limits->logical_block_size, bdevname(bdev, b));
374                 return 1;
375         }
376
377         return 0;
378 }
379
380 /*
381  * This upgrades the mode on an already open dm_dev, being
382  * careful to leave things as they were if we fail to reopen the
383  * device and not to touch the existing bdev field in case
384  * it is accessed concurrently inside dm_table_any_congested().
385  */
386 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
387                         struct mapped_device *md)
388 {
389         int r;
390         struct dm_dev *old_dev, *new_dev;
391
392         old_dev = dd->dm_dev;
393
394         r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
395                                 dd->dm_dev->mode | new_mode, &new_dev);
396         if (r)
397                 return r;
398
399         dd->dm_dev = new_dev;
400         dm_put_table_device(md, old_dev);
401
402         return 0;
403 }
404
405 /*
406  * Convert the path to a device
407  */
408 dev_t dm_get_dev_t(const char *path)
409 {
410         dev_t dev;
411         struct block_device *bdev;
412
413         bdev = lookup_bdev(path);
414         if (IS_ERR(bdev))
415                 dev = name_to_dev_t(path);
416         else {
417                 dev = bdev->bd_dev;
418                 bdput(bdev);
419         }
420
421         return dev;
422 }
423 EXPORT_SYMBOL_GPL(dm_get_dev_t);
424
425 /*
426  * Add a device to the list, or just increment the usage count if
427  * it's already present.
428  */
429 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
430                   struct dm_dev **result)
431 {
432         int r;
433         dev_t dev;
434         struct dm_dev_internal *dd;
435         struct dm_table *t = ti->table;
436
437         BUG_ON(!t);
438
439         dev = dm_get_dev_t(path);
440         if (!dev)
441                 return -ENODEV;
442
443         dd = find_device(&t->devices, dev);
444         if (!dd) {
445                 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
446                 if (!dd)
447                         return -ENOMEM;
448
449                 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
450                         kfree(dd);
451                         return r;
452                 }
453
454                 refcount_set(&dd->count, 1);
455                 list_add(&dd->list, &t->devices);
456
457         } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
458                 r = upgrade_mode(dd, mode, t->md);
459                 if (r)
460                         return r;
461                 refcount_inc(&dd->count);
462         }
463
464         *result = dd->dm_dev;
465         return 0;
466 }
467 EXPORT_SYMBOL(dm_get_device);
468
469 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
470                                 sector_t start, sector_t len, void *data)
471 {
472         struct queue_limits *limits = data;
473         struct block_device *bdev = dev->bdev;
474         struct request_queue *q = bdev_get_queue(bdev);
475         char b[BDEVNAME_SIZE];
476
477         if (unlikely(!q)) {
478                 DMWARN("%s: Cannot set limits for nonexistent device %s",
479                        dm_device_name(ti->table->md), bdevname(bdev, b));
480                 return 0;
481         }
482
483         if (bdev_stack_limits(limits, bdev, start) < 0)
484                 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
485                        "physical_block_size=%u, logical_block_size=%u, "
486                        "alignment_offset=%u, start=%llu",
487                        dm_device_name(ti->table->md), bdevname(bdev, b),
488                        q->limits.physical_block_size,
489                        q->limits.logical_block_size,
490                        q->limits.alignment_offset,
491                        (unsigned long long) start << SECTOR_SHIFT);
492
493         limits->zoned = blk_queue_zoned_model(q);
494
495         return 0;
496 }
497
498 /*
499  * Decrement a device's use count and remove it if necessary.
500  */
501 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
502 {
503         int found = 0;
504         struct list_head *devices = &ti->table->devices;
505         struct dm_dev_internal *dd;
506
507         list_for_each_entry(dd, devices, list) {
508                 if (dd->dm_dev == d) {
509                         found = 1;
510                         break;
511                 }
512         }
513         if (!found) {
514                 DMWARN("%s: device %s not in table devices list",
515                        dm_device_name(ti->table->md), d->name);
516                 return;
517         }
518         if (refcount_dec_and_test(&dd->count)) {
519                 dm_put_table_device(ti->table->md, d);
520                 list_del(&dd->list);
521                 kfree(dd);
522         }
523 }
524 EXPORT_SYMBOL(dm_put_device);
525
526 /*
527  * Checks to see if the target joins onto the end of the table.
528  */
529 static int adjoin(struct dm_table *table, struct dm_target *ti)
530 {
531         struct dm_target *prev;
532
533         if (!table->num_targets)
534                 return !ti->begin;
535
536         prev = &table->targets[table->num_targets - 1];
537         return (ti->begin == (prev->begin + prev->len));
538 }
539
540 /*
541  * Used to dynamically allocate the arg array.
542  *
543  * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
544  * process messages even if some device is suspended. These messages have a
545  * small fixed number of arguments.
546  *
547  * On the other hand, dm-switch needs to process bulk data using messages and
548  * excessive use of GFP_NOIO could cause trouble.
549  */
550 static char **realloc_argv(unsigned *array_size, char **old_argv)
551 {
552         char **argv;
553         unsigned new_size;
554         gfp_t gfp;
555
556         if (*array_size) {
557                 new_size = *array_size * 2;
558                 gfp = GFP_KERNEL;
559         } else {
560                 new_size = 8;
561                 gfp = GFP_NOIO;
562         }
563         argv = kmalloc(new_size * sizeof(*argv), gfp);
564         if (argv) {
565                 memcpy(argv, old_argv, *array_size * sizeof(*argv));
566                 *array_size = new_size;
567         }
568
569         kfree(old_argv);
570         return argv;
571 }
572
573 /*
574  * Destructively splits up the argument list to pass to ctr.
575  */
576 int dm_split_args(int *argc, char ***argvp, char *input)
577 {
578         char *start, *end = input, *out, **argv = NULL;
579         unsigned array_size = 0;
580
581         *argc = 0;
582
583         if (!input) {
584                 *argvp = NULL;
585                 return 0;
586         }
587
588         argv = realloc_argv(&array_size, argv);
589         if (!argv)
590                 return -ENOMEM;
591
592         while (1) {
593                 /* Skip whitespace */
594                 start = skip_spaces(end);
595
596                 if (!*start)
597                         break;  /* success, we hit the end */
598
599                 /* 'out' is used to remove any back-quotes */
600                 end = out = start;
601                 while (*end) {
602                         /* Everything apart from '\0' can be quoted */
603                         if (*end == '\\' && *(end + 1)) {
604                                 *out++ = *(end + 1);
605                                 end += 2;
606                                 continue;
607                         }
608
609                         if (isspace(*end))
610                                 break;  /* end of token */
611
612                         *out++ = *end++;
613                 }
614
615                 /* have we already filled the array ? */
616                 if ((*argc + 1) > array_size) {
617                         argv = realloc_argv(&array_size, argv);
618                         if (!argv)
619                                 return -ENOMEM;
620                 }
621
622                 /* we know this is whitespace */
623                 if (*end)
624                         end++;
625
626                 /* terminate the string and put it in the array */
627                 *out = '\0';
628                 argv[*argc] = start;
629                 (*argc)++;
630         }
631
632         *argvp = argv;
633         return 0;
634 }
635
636 /*
637  * Impose necessary and sufficient conditions on a devices's table such
638  * that any incoming bio which respects its logical_block_size can be
639  * processed successfully.  If it falls across the boundary between
640  * two or more targets, the size of each piece it gets split into must
641  * be compatible with the logical_block_size of the target processing it.
642  */
643 static int validate_hardware_logical_block_alignment(struct dm_table *table,
644                                                  struct queue_limits *limits)
645 {
646         /*
647          * This function uses arithmetic modulo the logical_block_size
648          * (in units of 512-byte sectors).
649          */
650         unsigned short device_logical_block_size_sects =
651                 limits->logical_block_size >> SECTOR_SHIFT;
652
653         /*
654          * Offset of the start of the next table entry, mod logical_block_size.
655          */
656         unsigned short next_target_start = 0;
657
658         /*
659          * Given an aligned bio that extends beyond the end of a
660          * target, how many sectors must the next target handle?
661          */
662         unsigned short remaining = 0;
663
664         struct dm_target *uninitialized_var(ti);
665         struct queue_limits ti_limits;
666         unsigned i;
667
668         /*
669          * Check each entry in the table in turn.
670          */
671         for (i = 0; i < dm_table_get_num_targets(table); i++) {
672                 ti = dm_table_get_target(table, i);
673
674                 blk_set_stacking_limits(&ti_limits);
675
676                 /* combine all target devices' limits */
677                 if (ti->type->iterate_devices)
678                         ti->type->iterate_devices(ti, dm_set_device_limits,
679                                                   &ti_limits);
680
681                 /*
682                  * If the remaining sectors fall entirely within this
683                  * table entry are they compatible with its logical_block_size?
684                  */
685                 if (remaining < ti->len &&
686                     remaining & ((ti_limits.logical_block_size >>
687                                   SECTOR_SHIFT) - 1))
688                         break;  /* Error */
689
690                 next_target_start =
691                     (unsigned short) ((next_target_start + ti->len) &
692                                       (device_logical_block_size_sects - 1));
693                 remaining = next_target_start ?
694                     device_logical_block_size_sects - next_target_start : 0;
695         }
696
697         if (remaining) {
698                 DMWARN("%s: table line %u (start sect %llu len %llu) "
699                        "not aligned to h/w logical block size %u",
700                        dm_device_name(table->md), i,
701                        (unsigned long long) ti->begin,
702                        (unsigned long long) ti->len,
703                        limits->logical_block_size);
704                 return -EINVAL;
705         }
706
707         return 0;
708 }
709
710 int dm_table_add_target(struct dm_table *t, const char *type,
711                         sector_t start, sector_t len, char *params)
712 {
713         int r = -EINVAL, argc;
714         char **argv;
715         struct dm_target *tgt;
716
717         if (t->singleton) {
718                 DMERR("%s: target type %s must appear alone in table",
719                       dm_device_name(t->md), t->targets->type->name);
720                 return -EINVAL;
721         }
722
723         BUG_ON(t->num_targets >= t->num_allocated);
724
725         tgt = t->targets + t->num_targets;
726         memset(tgt, 0, sizeof(*tgt));
727
728         if (!len) {
729                 DMERR("%s: zero-length target", dm_device_name(t->md));
730                 return -EINVAL;
731         }
732
733         tgt->type = dm_get_target_type(type);
734         if (!tgt->type) {
735                 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
736                 return -EINVAL;
737         }
738
739         if (dm_target_needs_singleton(tgt->type)) {
740                 if (t->num_targets) {
741                         tgt->error = "singleton target type must appear alone in table";
742                         goto bad;
743                 }
744                 t->singleton = true;
745         }
746
747         if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
748                 tgt->error = "target type may not be included in a read-only table";
749                 goto bad;
750         }
751
752         if (t->immutable_target_type) {
753                 if (t->immutable_target_type != tgt->type) {
754                         tgt->error = "immutable target type cannot be mixed with other target types";
755                         goto bad;
756                 }
757         } else if (dm_target_is_immutable(tgt->type)) {
758                 if (t->num_targets) {
759                         tgt->error = "immutable target type cannot be mixed with other target types";
760                         goto bad;
761                 }
762                 t->immutable_target_type = tgt->type;
763         }
764
765         if (dm_target_has_integrity(tgt->type))
766                 t->integrity_added = 1;
767
768         tgt->table = t;
769         tgt->begin = start;
770         tgt->len = len;
771         tgt->error = "Unknown error";
772
773         /*
774          * Does this target adjoin the previous one ?
775          */
776         if (!adjoin(t, tgt)) {
777                 tgt->error = "Gap in table";
778                 goto bad;
779         }
780
781         r = dm_split_args(&argc, &argv, params);
782         if (r) {
783                 tgt->error = "couldn't split parameters (insufficient memory)";
784                 goto bad;
785         }
786
787         r = tgt->type->ctr(tgt, argc, argv);
788         kfree(argv);
789         if (r)
790                 goto bad;
791
792         t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
793
794         if (!tgt->num_discard_bios && tgt->discards_supported)
795                 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
796                        dm_device_name(t->md), type);
797
798         return 0;
799
800  bad:
801         DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
802         dm_put_target_type(tgt->type);
803         return r;
804 }
805
806 /*
807  * Target argument parsing helpers.
808  */
809 static int validate_next_arg(const struct dm_arg *arg,
810                              struct dm_arg_set *arg_set,
811                              unsigned *value, char **error, unsigned grouped)
812 {
813         const char *arg_str = dm_shift_arg(arg_set);
814         char dummy;
815
816         if (!arg_str ||
817             (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
818             (*value < arg->min) ||
819             (*value > arg->max) ||
820             (grouped && arg_set->argc < *value)) {
821                 *error = arg->error;
822                 return -EINVAL;
823         }
824
825         return 0;
826 }
827
828 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
829                 unsigned *value, char **error)
830 {
831         return validate_next_arg(arg, arg_set, value, error, 0);
832 }
833 EXPORT_SYMBOL(dm_read_arg);
834
835 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
836                       unsigned *value, char **error)
837 {
838         return validate_next_arg(arg, arg_set, value, error, 1);
839 }
840 EXPORT_SYMBOL(dm_read_arg_group);
841
842 const char *dm_shift_arg(struct dm_arg_set *as)
843 {
844         char *r;
845
846         if (as->argc) {
847                 as->argc--;
848                 r = *as->argv;
849                 as->argv++;
850                 return r;
851         }
852
853         return NULL;
854 }
855 EXPORT_SYMBOL(dm_shift_arg);
856
857 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
858 {
859         BUG_ON(as->argc < num_args);
860         as->argc -= num_args;
861         as->argv += num_args;
862 }
863 EXPORT_SYMBOL(dm_consume_args);
864
865 static bool __table_type_bio_based(enum dm_queue_mode table_type)
866 {
867         return (table_type == DM_TYPE_BIO_BASED ||
868                 table_type == DM_TYPE_DAX_BIO_BASED);
869 }
870
871 static bool __table_type_request_based(enum dm_queue_mode table_type)
872 {
873         return (table_type == DM_TYPE_REQUEST_BASED ||
874                 table_type == DM_TYPE_MQ_REQUEST_BASED);
875 }
876
877 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
878 {
879         t->type = type;
880 }
881 EXPORT_SYMBOL_GPL(dm_table_set_type);
882
883 static int device_supports_dax(struct dm_target *ti, struct dm_dev *dev,
884                                sector_t start, sector_t len, void *data)
885 {
886         struct request_queue *q = bdev_get_queue(dev->bdev);
887
888         return q && blk_queue_dax(q);
889 }
890
891 static bool dm_table_supports_dax(struct dm_table *t)
892 {
893         struct dm_target *ti;
894         unsigned i;
895
896         /* Ensure that all targets support DAX. */
897         for (i = 0; i < dm_table_get_num_targets(t); i++) {
898                 ti = dm_table_get_target(t, i);
899
900                 if (!ti->type->direct_access)
901                         return false;
902
903                 if (!ti->type->iterate_devices ||
904                     !ti->type->iterate_devices(ti, device_supports_dax, NULL))
905                         return false;
906         }
907
908         return true;
909 }
910
911 static int dm_table_determine_type(struct dm_table *t)
912 {
913         unsigned i;
914         unsigned bio_based = 0, request_based = 0, hybrid = 0;
915         unsigned sq_count = 0, mq_count = 0;
916         struct dm_target *tgt;
917         struct dm_dev_internal *dd;
918         struct list_head *devices = dm_table_get_devices(t);
919         enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
920
921         if (t->type != DM_TYPE_NONE) {
922                 /* target already set the table's type */
923                 if (t->type == DM_TYPE_BIO_BASED)
924                         return 0;
925                 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
926                 goto verify_rq_based;
927         }
928
929         for (i = 0; i < t->num_targets; i++) {
930                 tgt = t->targets + i;
931                 if (dm_target_hybrid(tgt))
932                         hybrid = 1;
933                 else if (dm_target_request_based(tgt))
934                         request_based = 1;
935                 else
936                         bio_based = 1;
937
938                 if (bio_based && request_based) {
939                         DMWARN("Inconsistent table: different target types"
940                                " can't be mixed up");
941                         return -EINVAL;
942                 }
943         }
944
945         if (hybrid && !bio_based && !request_based) {
946                 /*
947                  * The targets can work either way.
948                  * Determine the type from the live device.
949                  * Default to bio-based if device is new.
950                  */
951                 if (__table_type_request_based(live_md_type))
952                         request_based = 1;
953                 else
954                         bio_based = 1;
955         }
956
957         if (bio_based) {
958                 /* We must use this table as bio-based */
959                 t->type = DM_TYPE_BIO_BASED;
960                 if (dm_table_supports_dax(t) ||
961                     (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED))
962                         t->type = DM_TYPE_DAX_BIO_BASED;
963                 return 0;
964         }
965
966         BUG_ON(!request_based); /* No targets in this table */
967
968         /*
969          * The only way to establish DM_TYPE_MQ_REQUEST_BASED is by
970          * having a compatible target use dm_table_set_type.
971          */
972         t->type = DM_TYPE_REQUEST_BASED;
973
974 verify_rq_based:
975         /*
976          * Request-based dm supports only tables that have a single target now.
977          * To support multiple targets, request splitting support is needed,
978          * and that needs lots of changes in the block-layer.
979          * (e.g. request completion process for partial completion.)
980          */
981         if (t->num_targets > 1) {
982                 DMWARN("Request-based dm doesn't support multiple targets yet");
983                 return -EINVAL;
984         }
985
986         if (list_empty(devices)) {
987                 int srcu_idx;
988                 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
989
990                 /* inherit live table's type and all_blk_mq */
991                 if (live_table) {
992                         t->type = live_table->type;
993                         t->all_blk_mq = live_table->all_blk_mq;
994                 }
995                 dm_put_live_table(t->md, srcu_idx);
996                 return 0;
997         }
998
999         /* Non-request-stackable devices can't be used for request-based dm */
1000         list_for_each_entry(dd, devices, list) {
1001                 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1002
1003                 if (!queue_is_rq_based(q)) {
1004                         DMERR("table load rejected: including"
1005                               " non-request-stackable devices");
1006                         return -EINVAL;
1007                 }
1008
1009                 if (q->mq_ops)
1010                         mq_count++;
1011                 else
1012                         sq_count++;
1013         }
1014         if (sq_count && mq_count) {
1015                 DMERR("table load rejected: not all devices are blk-mq request-stackable");
1016                 return -EINVAL;
1017         }
1018         t->all_blk_mq = mq_count > 0;
1019
1020         if (t->type == DM_TYPE_MQ_REQUEST_BASED && !t->all_blk_mq) {
1021                 DMERR("table load rejected: all devices are not blk-mq request-stackable");
1022                 return -EINVAL;
1023         }
1024
1025         return 0;
1026 }
1027
1028 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
1029 {
1030         return t->type;
1031 }
1032
1033 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
1034 {
1035         return t->immutable_target_type;
1036 }
1037
1038 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
1039 {
1040         /* Immutable target is implicitly a singleton */
1041         if (t->num_targets > 1 ||
1042             !dm_target_is_immutable(t->targets[0].type))
1043                 return NULL;
1044
1045         return t->targets;
1046 }
1047
1048 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1049 {
1050         struct dm_target *ti;
1051         unsigned i;
1052
1053         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1054                 ti = dm_table_get_target(t, i);
1055                 if (dm_target_is_wildcard(ti->type))
1056                         return ti;
1057         }
1058
1059         return NULL;
1060 }
1061
1062 bool dm_table_bio_based(struct dm_table *t)
1063 {
1064         return __table_type_bio_based(dm_table_get_type(t));
1065 }
1066
1067 bool dm_table_request_based(struct dm_table *t)
1068 {
1069         return __table_type_request_based(dm_table_get_type(t));
1070 }
1071
1072 bool dm_table_all_blk_mq_devices(struct dm_table *t)
1073 {
1074         return t->all_blk_mq;
1075 }
1076
1077 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1078 {
1079         enum dm_queue_mode type = dm_table_get_type(t);
1080         unsigned per_io_data_size = 0;
1081         struct dm_target *tgt;
1082         unsigned i;
1083
1084         if (unlikely(type == DM_TYPE_NONE)) {
1085                 DMWARN("no table type is set, can't allocate mempools");
1086                 return -EINVAL;
1087         }
1088
1089         if (__table_type_bio_based(type))
1090                 for (i = 0; i < t->num_targets; i++) {
1091                         tgt = t->targets + i;
1092                         per_io_data_size = max(per_io_data_size, tgt->per_io_data_size);
1093                 }
1094
1095         t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported, per_io_data_size);
1096         if (!t->mempools)
1097                 return -ENOMEM;
1098
1099         return 0;
1100 }
1101
1102 void dm_table_free_md_mempools(struct dm_table *t)
1103 {
1104         dm_free_md_mempools(t->mempools);
1105         t->mempools = NULL;
1106 }
1107
1108 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1109 {
1110         return t->mempools;
1111 }
1112
1113 static int setup_indexes(struct dm_table *t)
1114 {
1115         int i;
1116         unsigned int total = 0;
1117         sector_t *indexes;
1118
1119         /* allocate the space for *all* the indexes */
1120         for (i = t->depth - 2; i >= 0; i--) {
1121                 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1122                 total += t->counts[i];
1123         }
1124
1125         indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1126         if (!indexes)
1127                 return -ENOMEM;
1128
1129         /* set up internal nodes, bottom-up */
1130         for (i = t->depth - 2; i >= 0; i--) {
1131                 t->index[i] = indexes;
1132                 indexes += (KEYS_PER_NODE * t->counts[i]);
1133                 setup_btree_index(i, t);
1134         }
1135
1136         return 0;
1137 }
1138
1139 /*
1140  * Builds the btree to index the map.
1141  */
1142 static int dm_table_build_index(struct dm_table *t)
1143 {
1144         int r = 0;
1145         unsigned int leaf_nodes;
1146
1147         /* how many indexes will the btree have ? */
1148         leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1149         t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1150
1151         /* leaf layer has already been set up */
1152         t->counts[t->depth - 1] = leaf_nodes;
1153         t->index[t->depth - 1] = t->highs;
1154
1155         if (t->depth >= 2)
1156                 r = setup_indexes(t);
1157
1158         return r;
1159 }
1160
1161 static bool integrity_profile_exists(struct gendisk *disk)
1162 {
1163         return !!blk_get_integrity(disk);
1164 }
1165
1166 /*
1167  * Get a disk whose integrity profile reflects the table's profile.
1168  * Returns NULL if integrity support was inconsistent or unavailable.
1169  */
1170 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1171 {
1172         struct list_head *devices = dm_table_get_devices(t);
1173         struct dm_dev_internal *dd = NULL;
1174         struct gendisk *prev_disk = NULL, *template_disk = NULL;
1175         unsigned i;
1176
1177         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1178                 struct dm_target *ti = dm_table_get_target(t, i);
1179                 if (!dm_target_passes_integrity(ti->type))
1180                         goto no_integrity;
1181         }
1182
1183         list_for_each_entry(dd, devices, list) {
1184                 template_disk = dd->dm_dev->bdev->bd_disk;
1185                 if (!integrity_profile_exists(template_disk))
1186                         goto no_integrity;
1187                 else if (prev_disk &&
1188                          blk_integrity_compare(prev_disk, template_disk) < 0)
1189                         goto no_integrity;
1190                 prev_disk = template_disk;
1191         }
1192
1193         return template_disk;
1194
1195 no_integrity:
1196         if (prev_disk)
1197                 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1198                        dm_device_name(t->md),
1199                        prev_disk->disk_name,
1200                        template_disk->disk_name);
1201         return NULL;
1202 }
1203
1204 /*
1205  * Register the mapped device for blk_integrity support if the
1206  * underlying devices have an integrity profile.  But all devices may
1207  * not have matching profiles (checking all devices isn't reliable
1208  * during table load because this table may use other DM device(s) which
1209  * must be resumed before they will have an initialized integity
1210  * profile).  Consequently, stacked DM devices force a 2 stage integrity
1211  * profile validation: First pass during table load, final pass during
1212  * resume.
1213  */
1214 static int dm_table_register_integrity(struct dm_table *t)
1215 {
1216         struct mapped_device *md = t->md;
1217         struct gendisk *template_disk = NULL;
1218
1219         /* If target handles integrity itself do not register it here. */
1220         if (t->integrity_added)
1221                 return 0;
1222
1223         template_disk = dm_table_get_integrity_disk(t);
1224         if (!template_disk)
1225                 return 0;
1226
1227         if (!integrity_profile_exists(dm_disk(md))) {
1228                 t->integrity_supported = true;
1229                 /*
1230                  * Register integrity profile during table load; we can do
1231                  * this because the final profile must match during resume.
1232                  */
1233                 blk_integrity_register(dm_disk(md),
1234                                        blk_get_integrity(template_disk));
1235                 return 0;
1236         }
1237
1238         /*
1239          * If DM device already has an initialized integrity
1240          * profile the new profile should not conflict.
1241          */
1242         if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1243                 DMWARN("%s: conflict with existing integrity profile: "
1244                        "%s profile mismatch",
1245                        dm_device_name(t->md),
1246                        template_disk->disk_name);
1247                 return 1;
1248         }
1249
1250         /* Preserve existing integrity profile */
1251         t->integrity_supported = true;
1252         return 0;
1253 }
1254
1255 /*
1256  * Prepares the table for use by building the indices,
1257  * setting the type, and allocating mempools.
1258  */
1259 int dm_table_complete(struct dm_table *t)
1260 {
1261         int r;
1262
1263         r = dm_table_determine_type(t);
1264         if (r) {
1265                 DMERR("unable to determine table type");
1266                 return r;
1267         }
1268
1269         r = dm_table_build_index(t);
1270         if (r) {
1271                 DMERR("unable to build btrees");
1272                 return r;
1273         }
1274
1275         r = dm_table_register_integrity(t);
1276         if (r) {
1277                 DMERR("could not register integrity profile.");
1278                 return r;
1279         }
1280
1281         r = dm_table_alloc_md_mempools(t, t->md);
1282         if (r)
1283                 DMERR("unable to allocate mempools");
1284
1285         return r;
1286 }
1287
1288 static DEFINE_MUTEX(_event_lock);
1289 void dm_table_event_callback(struct dm_table *t,
1290                              void (*fn)(void *), void *context)
1291 {
1292         mutex_lock(&_event_lock);
1293         t->event_fn = fn;
1294         t->event_context = context;
1295         mutex_unlock(&_event_lock);
1296 }
1297
1298 void dm_table_event(struct dm_table *t)
1299 {
1300         /*
1301          * You can no longer call dm_table_event() from interrupt
1302          * context, use a bottom half instead.
1303          */
1304         BUG_ON(in_interrupt());
1305
1306         mutex_lock(&_event_lock);
1307         if (t->event_fn)
1308                 t->event_fn(t->event_context);
1309         mutex_unlock(&_event_lock);
1310 }
1311 EXPORT_SYMBOL(dm_table_event);
1312
1313 sector_t dm_table_get_size(struct dm_table *t)
1314 {
1315         return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1316 }
1317 EXPORT_SYMBOL(dm_table_get_size);
1318
1319 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1320 {
1321         if (index >= t->num_targets)
1322                 return NULL;
1323
1324         return t->targets + index;
1325 }
1326
1327 /*
1328  * Search the btree for the correct target.
1329  *
1330  * Caller should check returned pointer with dm_target_is_valid()
1331  * to trap I/O beyond end of device.
1332  */
1333 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1334 {
1335         unsigned int l, n = 0, k = 0;
1336         sector_t *node;
1337
1338         for (l = 0; l < t->depth; l++) {
1339                 n = get_child(n, k);
1340                 node = get_node(t, l, n);
1341
1342                 for (k = 0; k < KEYS_PER_NODE; k++)
1343                         if (node[k] >= sector)
1344                                 break;
1345         }
1346
1347         return &t->targets[(KEYS_PER_NODE * n) + k];
1348 }
1349
1350 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1351                         sector_t start, sector_t len, void *data)
1352 {
1353         unsigned *num_devices = data;
1354
1355         (*num_devices)++;
1356
1357         return 0;
1358 }
1359
1360 /*
1361  * Check whether a table has no data devices attached using each
1362  * target's iterate_devices method.
1363  * Returns false if the result is unknown because a target doesn't
1364  * support iterate_devices.
1365  */
1366 bool dm_table_has_no_data_devices(struct dm_table *table)
1367 {
1368         struct dm_target *ti;
1369         unsigned i, num_devices;
1370
1371         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1372                 ti = dm_table_get_target(table, i);
1373
1374                 if (!ti->type->iterate_devices)
1375                         return false;
1376
1377                 num_devices = 0;
1378                 ti->type->iterate_devices(ti, count_device, &num_devices);
1379                 if (num_devices)
1380                         return false;
1381         }
1382
1383         return true;
1384 }
1385
1386 static int device_is_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1387                                  sector_t start, sector_t len, void *data)
1388 {
1389         struct request_queue *q = bdev_get_queue(dev->bdev);
1390         enum blk_zoned_model *zoned_model = data;
1391
1392         return q && blk_queue_zoned_model(q) == *zoned_model;
1393 }
1394
1395 static bool dm_table_supports_zoned_model(struct dm_table *t,
1396                                           enum blk_zoned_model zoned_model)
1397 {
1398         struct dm_target *ti;
1399         unsigned i;
1400
1401         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1402                 ti = dm_table_get_target(t, i);
1403
1404                 if (zoned_model == BLK_ZONED_HM &&
1405                     !dm_target_supports_zoned_hm(ti->type))
1406                         return false;
1407
1408                 if (!ti->type->iterate_devices ||
1409                     !ti->type->iterate_devices(ti, device_is_zoned_model, &zoned_model))
1410                         return false;
1411         }
1412
1413         return true;
1414 }
1415
1416 static int device_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1417                                        sector_t start, sector_t len, void *data)
1418 {
1419         struct request_queue *q = bdev_get_queue(dev->bdev);
1420         unsigned int *zone_sectors = data;
1421
1422         return q && blk_queue_zone_sectors(q) == *zone_sectors;
1423 }
1424
1425 static bool dm_table_matches_zone_sectors(struct dm_table *t,
1426                                           unsigned int zone_sectors)
1427 {
1428         struct dm_target *ti;
1429         unsigned i;
1430
1431         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1432                 ti = dm_table_get_target(t, i);
1433
1434                 if (!ti->type->iterate_devices ||
1435                     !ti->type->iterate_devices(ti, device_matches_zone_sectors, &zone_sectors))
1436                         return false;
1437         }
1438
1439         return true;
1440 }
1441
1442 static int validate_hardware_zoned_model(struct dm_table *table,
1443                                          enum blk_zoned_model zoned_model,
1444                                          unsigned int zone_sectors)
1445 {
1446         if (zoned_model == BLK_ZONED_NONE)
1447                 return 0;
1448
1449         if (!dm_table_supports_zoned_model(table, zoned_model)) {
1450                 DMERR("%s: zoned model is not consistent across all devices",
1451                       dm_device_name(table->md));
1452                 return -EINVAL;
1453         }
1454
1455         /* Check zone size validity and compatibility */
1456         if (!zone_sectors || !is_power_of_2(zone_sectors))
1457                 return -EINVAL;
1458
1459         if (!dm_table_matches_zone_sectors(table, zone_sectors)) {
1460                 DMERR("%s: zone sectors is not consistent across all devices",
1461                       dm_device_name(table->md));
1462                 return -EINVAL;
1463         }
1464
1465         return 0;
1466 }
1467
1468 /*
1469  * Establish the new table's queue_limits and validate them.
1470  */
1471 int dm_calculate_queue_limits(struct dm_table *table,
1472                               struct queue_limits *limits)
1473 {
1474         struct dm_target *ti;
1475         struct queue_limits ti_limits;
1476         unsigned i;
1477         enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1478         unsigned int zone_sectors = 0;
1479
1480         blk_set_stacking_limits(limits);
1481
1482         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1483                 blk_set_stacking_limits(&ti_limits);
1484
1485                 ti = dm_table_get_target(table, i);
1486
1487                 if (!ti->type->iterate_devices)
1488                         goto combine_limits;
1489
1490                 /*
1491                  * Combine queue limits of all the devices this target uses.
1492                  */
1493                 ti->type->iterate_devices(ti, dm_set_device_limits,
1494                                           &ti_limits);
1495
1496                 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1497                         /*
1498                          * After stacking all limits, validate all devices
1499                          * in table support this zoned model and zone sectors.
1500                          */
1501                         zoned_model = ti_limits.zoned;
1502                         zone_sectors = ti_limits.chunk_sectors;
1503                 }
1504
1505                 /* Set I/O hints portion of queue limits */
1506                 if (ti->type->io_hints)
1507                         ti->type->io_hints(ti, &ti_limits);
1508
1509                 /*
1510                  * Check each device area is consistent with the target's
1511                  * overall queue limits.
1512                  */
1513                 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1514                                               &ti_limits))
1515                         return -EINVAL;
1516
1517 combine_limits:
1518                 /*
1519                  * Merge this target's queue limits into the overall limits
1520                  * for the table.
1521                  */
1522                 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1523                         DMWARN("%s: adding target device "
1524                                "(start sect %llu len %llu) "
1525                                "caused an alignment inconsistency",
1526                                dm_device_name(table->md),
1527                                (unsigned long long) ti->begin,
1528                                (unsigned long long) ti->len);
1529
1530                 /*
1531                  * FIXME: this should likely be moved to blk_stack_limits(), would
1532                  * also eliminate limits->zoned stacking hack in dm_set_device_limits()
1533                  */
1534                 if (limits->zoned == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1535                         /*
1536                          * By default, the stacked limits zoned model is set to
1537                          * BLK_ZONED_NONE in blk_set_stacking_limits(). Update
1538                          * this model using the first target model reported
1539                          * that is not BLK_ZONED_NONE. This will be either the
1540                          * first target device zoned model or the model reported
1541                          * by the target .io_hints.
1542                          */
1543                         limits->zoned = ti_limits.zoned;
1544                 }
1545         }
1546
1547         /*
1548          * Verify that the zoned model and zone sectors, as determined before
1549          * any .io_hints override, are the same across all devices in the table.
1550          * - this is especially relevant if .io_hints is emulating a disk-managed
1551          *   zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1552          * BUT...
1553          */
1554         if (limits->zoned != BLK_ZONED_NONE) {
1555                 /*
1556                  * ...IF the above limits stacking determined a zoned model
1557                  * validate that all of the table's devices conform to it.
1558                  */
1559                 zoned_model = limits->zoned;
1560                 zone_sectors = limits->chunk_sectors;
1561         }
1562         if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1563                 return -EINVAL;
1564
1565         return validate_hardware_logical_block_alignment(table, limits);
1566 }
1567
1568 /*
1569  * Verify that all devices have an integrity profile that matches the
1570  * DM device's registered integrity profile.  If the profiles don't
1571  * match then unregister the DM device's integrity profile.
1572  */
1573 static void dm_table_verify_integrity(struct dm_table *t)
1574 {
1575         struct gendisk *template_disk = NULL;
1576
1577         if (t->integrity_added)
1578                 return;
1579
1580         if (t->integrity_supported) {
1581                 /*
1582                  * Verify that the original integrity profile
1583                  * matches all the devices in this table.
1584                  */
1585                 template_disk = dm_table_get_integrity_disk(t);
1586                 if (template_disk &&
1587                     blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1588                         return;
1589         }
1590
1591         if (integrity_profile_exists(dm_disk(t->md))) {
1592                 DMWARN("%s: unable to establish an integrity profile",
1593                        dm_device_name(t->md));
1594                 blk_integrity_unregister(dm_disk(t->md));
1595         }
1596 }
1597
1598 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1599                                 sector_t start, sector_t len, void *data)
1600 {
1601         unsigned long flush = (unsigned long) data;
1602         struct request_queue *q = bdev_get_queue(dev->bdev);
1603
1604         return q && (q->queue_flags & flush);
1605 }
1606
1607 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1608 {
1609         struct dm_target *ti;
1610         unsigned i;
1611
1612         /*
1613          * Require at least one underlying device to support flushes.
1614          * t->devices includes internal dm devices such as mirror logs
1615          * so we need to use iterate_devices here, which targets
1616          * supporting flushes must provide.
1617          */
1618         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1619                 ti = dm_table_get_target(t, i);
1620
1621                 if (!ti->num_flush_bios)
1622                         continue;
1623
1624                 if (ti->flush_supported)
1625                         return true;
1626
1627                 if (ti->type->iterate_devices &&
1628                     ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1629                         return true;
1630         }
1631
1632         return false;
1633 }
1634
1635 static int device_dax_write_cache_enabled(struct dm_target *ti,
1636                                           struct dm_dev *dev, sector_t start,
1637                                           sector_t len, void *data)
1638 {
1639         struct dax_device *dax_dev = dev->dax_dev;
1640
1641         if (!dax_dev)
1642                 return false;
1643
1644         if (dax_write_cache_enabled(dax_dev))
1645                 return true;
1646         return false;
1647 }
1648
1649 static int dm_table_supports_dax_write_cache(struct dm_table *t)
1650 {
1651         struct dm_target *ti;
1652         unsigned i;
1653
1654         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1655                 ti = dm_table_get_target(t, i);
1656
1657                 if (ti->type->iterate_devices &&
1658                     ti->type->iterate_devices(ti,
1659                                 device_dax_write_cache_enabled, NULL))
1660                         return true;
1661         }
1662
1663         return false;
1664 }
1665
1666 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1667                             sector_t start, sector_t len, void *data)
1668 {
1669         struct request_queue *q = bdev_get_queue(dev->bdev);
1670
1671         return q && blk_queue_nonrot(q);
1672 }
1673
1674 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1675                              sector_t start, sector_t len, void *data)
1676 {
1677         struct request_queue *q = bdev_get_queue(dev->bdev);
1678
1679         return q && !blk_queue_add_random(q);
1680 }
1681
1682 static int queue_supports_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1683                                    sector_t start, sector_t len, void *data)
1684 {
1685         struct request_queue *q = bdev_get_queue(dev->bdev);
1686
1687         return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1688 }
1689
1690 static bool dm_table_all_devices_attribute(struct dm_table *t,
1691                                            iterate_devices_callout_fn func)
1692 {
1693         struct dm_target *ti;
1694         unsigned i;
1695
1696         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1697                 ti = dm_table_get_target(t, i);
1698
1699                 if (!ti->type->iterate_devices ||
1700                     !ti->type->iterate_devices(ti, func, NULL))
1701                         return false;
1702         }
1703
1704         return true;
1705 }
1706
1707 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1708                                          sector_t start, sector_t len, void *data)
1709 {
1710         struct request_queue *q = bdev_get_queue(dev->bdev);
1711
1712         return q && !q->limits.max_write_same_sectors;
1713 }
1714
1715 static bool dm_table_supports_write_same(struct dm_table *t)
1716 {
1717         struct dm_target *ti;
1718         unsigned i;
1719
1720         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1721                 ti = dm_table_get_target(t, i);
1722
1723                 if (!ti->num_write_same_bios)
1724                         return false;
1725
1726                 if (!ti->type->iterate_devices ||
1727                     ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1728                         return false;
1729         }
1730
1731         return true;
1732 }
1733
1734 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1735                                            sector_t start, sector_t len, void *data)
1736 {
1737         struct request_queue *q = bdev_get_queue(dev->bdev);
1738
1739         return q && !q->limits.max_write_zeroes_sectors;
1740 }
1741
1742 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1743 {
1744         struct dm_target *ti;
1745         unsigned i = 0;
1746
1747         while (i < dm_table_get_num_targets(t)) {
1748                 ti = dm_table_get_target(t, i++);
1749
1750                 if (!ti->num_write_zeroes_bios)
1751                         return false;
1752
1753                 if (!ti->type->iterate_devices ||
1754                     ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1755                         return false;
1756         }
1757
1758         return true;
1759 }
1760
1761 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1762                                       sector_t start, sector_t len, void *data)
1763 {
1764         struct request_queue *q = bdev_get_queue(dev->bdev);
1765
1766         return q && !blk_queue_discard(q);
1767 }
1768
1769 static bool dm_table_supports_discards(struct dm_table *t)
1770 {
1771         struct dm_target *ti;
1772         unsigned i;
1773
1774         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1775                 ti = dm_table_get_target(t, i);
1776
1777                 if (!ti->num_discard_bios)
1778                         return false;
1779
1780                 /*
1781                  * Either the target provides discard support (as implied by setting
1782                  * 'discards_supported') or it relies on _all_ data devices having
1783                  * discard support.
1784                  */
1785                 if (!ti->discards_supported &&
1786                     (!ti->type->iterate_devices ||
1787                      ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1788                         return false;
1789         }
1790
1791         return true;
1792 }
1793
1794 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1795                                struct queue_limits *limits)
1796 {
1797         bool wc = false, fua = false;
1798
1799         /*
1800          * Copy table's limits to the DM device's request_queue
1801          */
1802         q->limits = *limits;
1803
1804         if (!dm_table_supports_discards(t)) {
1805                 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1806                 /* Must also clear discard limits... */
1807                 q->limits.max_discard_sectors = 0;
1808                 q->limits.max_hw_discard_sectors = 0;
1809                 q->limits.discard_granularity = 0;
1810                 q->limits.discard_alignment = 0;
1811                 q->limits.discard_misaligned = 0;
1812         } else
1813                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1814
1815         if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1816                 wc = true;
1817                 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1818                         fua = true;
1819         }
1820         blk_queue_write_cache(q, wc, fua);
1821
1822         if (dm_table_supports_dax_write_cache(t))
1823                 dax_write_cache(t->md->dax_dev, true);
1824
1825         /* Ensure that all underlying devices are non-rotational. */
1826         if (dm_table_all_devices_attribute(t, device_is_nonrot))
1827                 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1828         else
1829                 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1830
1831         if (!dm_table_supports_write_same(t))
1832                 q->limits.max_write_same_sectors = 0;
1833         if (!dm_table_supports_write_zeroes(t))
1834                 q->limits.max_write_zeroes_sectors = 0;
1835
1836         if (dm_table_all_devices_attribute(t, queue_supports_sg_merge))
1837                 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1838         else
1839                 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1840
1841         dm_table_verify_integrity(t);
1842
1843         /*
1844          * Determine whether or not this queue's I/O timings contribute
1845          * to the entropy pool, Only request-based targets use this.
1846          * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1847          * have it set.
1848          */
1849         if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1850                 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1851 }
1852
1853 unsigned int dm_table_get_num_targets(struct dm_table *t)
1854 {
1855         return t->num_targets;
1856 }
1857
1858 struct list_head *dm_table_get_devices(struct dm_table *t)
1859 {
1860         return &t->devices;
1861 }
1862
1863 fmode_t dm_table_get_mode(struct dm_table *t)
1864 {
1865         return t->mode;
1866 }
1867 EXPORT_SYMBOL(dm_table_get_mode);
1868
1869 enum suspend_mode {
1870         PRESUSPEND,
1871         PRESUSPEND_UNDO,
1872         POSTSUSPEND,
1873 };
1874
1875 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1876 {
1877         int i = t->num_targets;
1878         struct dm_target *ti = t->targets;
1879
1880         lockdep_assert_held(&t->md->suspend_lock);
1881
1882         while (i--) {
1883                 switch (mode) {
1884                 case PRESUSPEND:
1885                         if (ti->type->presuspend)
1886                                 ti->type->presuspend(ti);
1887                         break;
1888                 case PRESUSPEND_UNDO:
1889                         if (ti->type->presuspend_undo)
1890                                 ti->type->presuspend_undo(ti);
1891                         break;
1892                 case POSTSUSPEND:
1893                         if (ti->type->postsuspend)
1894                                 ti->type->postsuspend(ti);
1895                         break;
1896                 }
1897                 ti++;
1898         }
1899 }
1900
1901 void dm_table_presuspend_targets(struct dm_table *t)
1902 {
1903         if (!t)
1904                 return;
1905
1906         suspend_targets(t, PRESUSPEND);
1907 }
1908
1909 void dm_table_presuspend_undo_targets(struct dm_table *t)
1910 {
1911         if (!t)
1912                 return;
1913
1914         suspend_targets(t, PRESUSPEND_UNDO);
1915 }
1916
1917 void dm_table_postsuspend_targets(struct dm_table *t)
1918 {
1919         if (!t)
1920                 return;
1921
1922         suspend_targets(t, POSTSUSPEND);
1923 }
1924
1925 int dm_table_resume_targets(struct dm_table *t)
1926 {
1927         int i, r = 0;
1928
1929         lockdep_assert_held(&t->md->suspend_lock);
1930
1931         for (i = 0; i < t->num_targets; i++) {
1932                 struct dm_target *ti = t->targets + i;
1933
1934                 if (!ti->type->preresume)
1935                         continue;
1936
1937                 r = ti->type->preresume(ti);
1938                 if (r) {
1939                         DMERR("%s: %s: preresume failed, error = %d",
1940                               dm_device_name(t->md), ti->type->name, r);
1941                         return r;
1942                 }
1943         }
1944
1945         for (i = 0; i < t->num_targets; i++) {
1946                 struct dm_target *ti = t->targets + i;
1947
1948                 if (ti->type->resume)
1949                         ti->type->resume(ti);
1950         }
1951
1952         return 0;
1953 }
1954
1955 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1956 {
1957         list_add(&cb->list, &t->target_callbacks);
1958 }
1959 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1960
1961 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1962 {
1963         struct dm_dev_internal *dd;
1964         struct list_head *devices = dm_table_get_devices(t);
1965         struct dm_target_callbacks *cb;
1966         int r = 0;
1967
1968         list_for_each_entry(dd, devices, list) {
1969                 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1970                 char b[BDEVNAME_SIZE];
1971
1972                 if (likely(q))
1973                         r |= bdi_congested(q->backing_dev_info, bdi_bits);
1974                 else
1975                         DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1976                                      dm_device_name(t->md),
1977                                      bdevname(dd->dm_dev->bdev, b));
1978         }
1979
1980         list_for_each_entry(cb, &t->target_callbacks, list)
1981                 if (cb->congested_fn)
1982                         r |= cb->congested_fn(cb, bdi_bits);
1983
1984         return r;
1985 }
1986
1987 struct mapped_device *dm_table_get_md(struct dm_table *t)
1988 {
1989         return t->md;
1990 }
1991 EXPORT_SYMBOL(dm_table_get_md);
1992
1993 void dm_table_run_md_queue_async(struct dm_table *t)
1994 {
1995         struct mapped_device *md;
1996         struct request_queue *queue;
1997         unsigned long flags;
1998
1999         if (!dm_table_request_based(t))
2000                 return;
2001
2002         md = dm_table_get_md(t);
2003         queue = dm_get_md_queue(md);
2004         if (queue) {
2005                 if (queue->mq_ops)
2006                         blk_mq_run_hw_queues(queue, true);
2007                 else {
2008                         spin_lock_irqsave(queue->queue_lock, flags);
2009                         blk_run_queue_async(queue);
2010                         spin_unlock_irqrestore(queue->queue_lock, flags);
2011                 }
2012         }
2013 }
2014 EXPORT_SYMBOL(dm_table_run_md_queue_async);
2015