Merge tag 'scsi-misc' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
[platform/kernel/linux-starfive.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/blk-integrity.h>
14 #include <linux/namei.h>
15 #include <linux/ctype.h>
16 #include <linux/string.h>
17 #include <linux/slab.h>
18 #include <linux/interrupt.h>
19 #include <linux/mutex.h>
20 #include <linux/delay.h>
21 #include <linux/atomic.h>
22 #include <linux/blk-mq.h>
23 #include <linux/mount.h>
24 #include <linux/dax.h>
25
26 #define DM_MSG_PREFIX "table"
27
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 /*
33  * Similar to ceiling(log_size(n))
34  */
35 static unsigned int int_log(unsigned int n, unsigned int base)
36 {
37         int result = 0;
38
39         while (n > 1) {
40                 n = dm_div_up(n, base);
41                 result++;
42         }
43
44         return result;
45 }
46
47 /*
48  * Calculate the index of the child node of the n'th node k'th key.
49  */
50 static inline unsigned int get_child(unsigned int n, unsigned int k)
51 {
52         return (n * CHILDREN_PER_NODE) + k;
53 }
54
55 /*
56  * Return the n'th node of level l from table t.
57  */
58 static inline sector_t *get_node(struct dm_table *t,
59                                  unsigned int l, unsigned int n)
60 {
61         return t->index[l] + (n * KEYS_PER_NODE);
62 }
63
64 /*
65  * Return the highest key that you could lookup from the n'th
66  * node on level l of the btree.
67  */
68 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
69 {
70         for (; l < t->depth - 1; l++)
71                 n = get_child(n, CHILDREN_PER_NODE - 1);
72
73         if (n >= t->counts[l])
74                 return (sector_t) - 1;
75
76         return get_node(t, l, n)[KEYS_PER_NODE - 1];
77 }
78
79 /*
80  * Fills in a level of the btree based on the highs of the level
81  * below it.
82  */
83 static int setup_btree_index(unsigned int l, struct dm_table *t)
84 {
85         unsigned int n, k;
86         sector_t *node;
87
88         for (n = 0U; n < t->counts[l]; n++) {
89                 node = get_node(t, l, n);
90
91                 for (k = 0U; k < KEYS_PER_NODE; k++)
92                         node[k] = high(t, l + 1, get_child(n, k));
93         }
94
95         return 0;
96 }
97
98 /*
99  * highs, and targets are managed as dynamic arrays during a
100  * table load.
101  */
102 static int alloc_targets(struct dm_table *t, unsigned int num)
103 {
104         sector_t *n_highs;
105         struct dm_target *n_targets;
106
107         /*
108          * Allocate both the target array and offset array at once.
109          */
110         n_highs = kvcalloc(num, sizeof(struct dm_target) + sizeof(sector_t),
111                            GFP_KERNEL);
112         if (!n_highs)
113                 return -ENOMEM;
114
115         n_targets = (struct dm_target *) (n_highs + num);
116
117         memset(n_highs, -1, sizeof(*n_highs) * num);
118         kvfree(t->highs);
119
120         t->num_allocated = num;
121         t->highs = n_highs;
122         t->targets = n_targets;
123
124         return 0;
125 }
126
127 int dm_table_create(struct dm_table **result, fmode_t mode,
128                     unsigned num_targets, struct mapped_device *md)
129 {
130         struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
131
132         if (!t)
133                 return -ENOMEM;
134
135         INIT_LIST_HEAD(&t->devices);
136
137         if (!num_targets)
138                 num_targets = KEYS_PER_NODE;
139
140         num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
141
142         if (!num_targets) {
143                 kfree(t);
144                 return -ENOMEM;
145         }
146
147         if (alloc_targets(t, num_targets)) {
148                 kfree(t);
149                 return -ENOMEM;
150         }
151
152         t->type = DM_TYPE_NONE;
153         t->mode = mode;
154         t->md = md;
155         *result = t;
156         return 0;
157 }
158
159 static void free_devices(struct list_head *devices, struct mapped_device *md)
160 {
161         struct list_head *tmp, *next;
162
163         list_for_each_safe(tmp, next, devices) {
164                 struct dm_dev_internal *dd =
165                     list_entry(tmp, struct dm_dev_internal, list);
166                 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
167                        dm_device_name(md), dd->dm_dev->name);
168                 dm_put_table_device(md, dd->dm_dev);
169                 kfree(dd);
170         }
171 }
172
173 static void dm_table_destroy_crypto_profile(struct dm_table *t);
174
175 void dm_table_destroy(struct dm_table *t)
176 {
177         unsigned int i;
178
179         if (!t)
180                 return;
181
182         /* free the indexes */
183         if (t->depth >= 2)
184                 kvfree(t->index[t->depth - 2]);
185
186         /* free the targets */
187         for (i = 0; i < t->num_targets; i++) {
188                 struct dm_target *tgt = t->targets + i;
189
190                 if (tgt->type->dtr)
191                         tgt->type->dtr(tgt);
192
193                 dm_put_target_type(tgt->type);
194         }
195
196         kvfree(t->highs);
197
198         /* free the device list */
199         free_devices(&t->devices, t->md);
200
201         dm_free_md_mempools(t->mempools);
202
203         dm_table_destroy_crypto_profile(t);
204
205         kfree(t);
206 }
207
208 /*
209  * See if we've already got a device in the list.
210  */
211 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
212 {
213         struct dm_dev_internal *dd;
214
215         list_for_each_entry (dd, l, list)
216                 if (dd->dm_dev->bdev->bd_dev == dev)
217                         return dd;
218
219         return NULL;
220 }
221
222 /*
223  * If possible, this checks an area of a destination device is invalid.
224  */
225 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
226                                   sector_t start, sector_t len, void *data)
227 {
228         struct queue_limits *limits = data;
229         struct block_device *bdev = dev->bdev;
230         sector_t dev_size = bdev_nr_sectors(bdev);
231         unsigned short logical_block_size_sectors =
232                 limits->logical_block_size >> SECTOR_SHIFT;
233
234         if (!dev_size)
235                 return 0;
236
237         if ((start >= dev_size) || (start + len > dev_size)) {
238                 DMWARN("%s: %pg too small for target: "
239                        "start=%llu, len=%llu, dev_size=%llu",
240                        dm_device_name(ti->table->md), bdev,
241                        (unsigned long long)start,
242                        (unsigned long long)len,
243                        (unsigned long long)dev_size);
244                 return 1;
245         }
246
247         /*
248          * If the target is mapped to zoned block device(s), check
249          * that the zones are not partially mapped.
250          */
251         if (bdev_is_zoned(bdev)) {
252                 unsigned int zone_sectors = bdev_zone_sectors(bdev);
253
254                 if (start & (zone_sectors - 1)) {
255                         DMWARN("%s: start=%llu not aligned to h/w zone size %u of %pg",
256                                dm_device_name(ti->table->md),
257                                (unsigned long long)start,
258                                zone_sectors, bdev);
259                         return 1;
260                 }
261
262                 /*
263                  * Note: The last zone of a zoned block device may be smaller
264                  * than other zones. So for a target mapping the end of a
265                  * zoned block device with such a zone, len would not be zone
266                  * aligned. We do not allow such last smaller zone to be part
267                  * of the mapping here to ensure that mappings with multiple
268                  * devices do not end up with a smaller zone in the middle of
269                  * the sector range.
270                  */
271                 if (len & (zone_sectors - 1)) {
272                         DMWARN("%s: len=%llu not aligned to h/w zone size %u of %pg",
273                                dm_device_name(ti->table->md),
274                                (unsigned long long)len,
275                                zone_sectors, bdev);
276                         return 1;
277                 }
278         }
279
280         if (logical_block_size_sectors <= 1)
281                 return 0;
282
283         if (start & (logical_block_size_sectors - 1)) {
284                 DMWARN("%s: start=%llu not aligned to h/w "
285                        "logical block size %u of %pg",
286                        dm_device_name(ti->table->md),
287                        (unsigned long long)start,
288                        limits->logical_block_size, bdev);
289                 return 1;
290         }
291
292         if (len & (logical_block_size_sectors - 1)) {
293                 DMWARN("%s: len=%llu not aligned to h/w "
294                        "logical block size %u of %pg",
295                        dm_device_name(ti->table->md),
296                        (unsigned long long)len,
297                        limits->logical_block_size, bdev);
298                 return 1;
299         }
300
301         return 0;
302 }
303
304 /*
305  * This upgrades the mode on an already open dm_dev, being
306  * careful to leave things as they were if we fail to reopen the
307  * device and not to touch the existing bdev field in case
308  * it is accessed concurrently.
309  */
310 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
311                         struct mapped_device *md)
312 {
313         int r;
314         struct dm_dev *old_dev, *new_dev;
315
316         old_dev = dd->dm_dev;
317
318         r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
319                                 dd->dm_dev->mode | new_mode, &new_dev);
320         if (r)
321                 return r;
322
323         dd->dm_dev = new_dev;
324         dm_put_table_device(md, old_dev);
325
326         return 0;
327 }
328
329 /*
330  * Convert the path to a device
331  */
332 dev_t dm_get_dev_t(const char *path)
333 {
334         dev_t dev;
335
336         if (lookup_bdev(path, &dev))
337                 dev = name_to_dev_t(path);
338         return dev;
339 }
340 EXPORT_SYMBOL_GPL(dm_get_dev_t);
341
342 /*
343  * Add a device to the list, or just increment the usage count if
344  * it's already present.
345  */
346 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
347                   struct dm_dev **result)
348 {
349         int r;
350         dev_t dev;
351         unsigned int major, minor;
352         char dummy;
353         struct dm_dev_internal *dd;
354         struct dm_table *t = ti->table;
355
356         BUG_ON(!t);
357
358         if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
359                 /* Extract the major/minor numbers */
360                 dev = MKDEV(major, minor);
361                 if (MAJOR(dev) != major || MINOR(dev) != minor)
362                         return -EOVERFLOW;
363         } else {
364                 dev = dm_get_dev_t(path);
365                 if (!dev)
366                         return -ENODEV;
367         }
368
369         dd = find_device(&t->devices, dev);
370         if (!dd) {
371                 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
372                 if (!dd)
373                         return -ENOMEM;
374
375                 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
376                         kfree(dd);
377                         return r;
378                 }
379
380                 refcount_set(&dd->count, 1);
381                 list_add(&dd->list, &t->devices);
382                 goto out;
383
384         } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
385                 r = upgrade_mode(dd, mode, t->md);
386                 if (r)
387                         return r;
388         }
389         refcount_inc(&dd->count);
390 out:
391         *result = dd->dm_dev;
392         return 0;
393 }
394 EXPORT_SYMBOL(dm_get_device);
395
396 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
397                                 sector_t start, sector_t len, void *data)
398 {
399         struct queue_limits *limits = data;
400         struct block_device *bdev = dev->bdev;
401         struct request_queue *q = bdev_get_queue(bdev);
402
403         if (unlikely(!q)) {
404                 DMWARN("%s: Cannot set limits for nonexistent device %pg",
405                        dm_device_name(ti->table->md), bdev);
406                 return 0;
407         }
408
409         if (blk_stack_limits(limits, &q->limits,
410                         get_start_sect(bdev) + start) < 0)
411                 DMWARN("%s: adding target device %pg caused an alignment inconsistency: "
412                        "physical_block_size=%u, logical_block_size=%u, "
413                        "alignment_offset=%u, start=%llu",
414                        dm_device_name(ti->table->md), bdev,
415                        q->limits.physical_block_size,
416                        q->limits.logical_block_size,
417                        q->limits.alignment_offset,
418                        (unsigned long long) start << SECTOR_SHIFT);
419         return 0;
420 }
421
422 /*
423  * Decrement a device's use count and remove it if necessary.
424  */
425 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
426 {
427         int found = 0;
428         struct list_head *devices = &ti->table->devices;
429         struct dm_dev_internal *dd;
430
431         list_for_each_entry(dd, devices, list) {
432                 if (dd->dm_dev == d) {
433                         found = 1;
434                         break;
435                 }
436         }
437         if (!found) {
438                 DMWARN("%s: device %s not in table devices list",
439                        dm_device_name(ti->table->md), d->name);
440                 return;
441         }
442         if (refcount_dec_and_test(&dd->count)) {
443                 dm_put_table_device(ti->table->md, d);
444                 list_del(&dd->list);
445                 kfree(dd);
446         }
447 }
448 EXPORT_SYMBOL(dm_put_device);
449
450 /*
451  * Checks to see if the target joins onto the end of the table.
452  */
453 static int adjoin(struct dm_table *table, struct dm_target *ti)
454 {
455         struct dm_target *prev;
456
457         if (!table->num_targets)
458                 return !ti->begin;
459
460         prev = &table->targets[table->num_targets - 1];
461         return (ti->begin == (prev->begin + prev->len));
462 }
463
464 /*
465  * Used to dynamically allocate the arg array.
466  *
467  * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
468  * process messages even if some device is suspended. These messages have a
469  * small fixed number of arguments.
470  *
471  * On the other hand, dm-switch needs to process bulk data using messages and
472  * excessive use of GFP_NOIO could cause trouble.
473  */
474 static char **realloc_argv(unsigned *size, char **old_argv)
475 {
476         char **argv;
477         unsigned new_size;
478         gfp_t gfp;
479
480         if (*size) {
481                 new_size = *size * 2;
482                 gfp = GFP_KERNEL;
483         } else {
484                 new_size = 8;
485                 gfp = GFP_NOIO;
486         }
487         argv = kmalloc_array(new_size, sizeof(*argv), gfp);
488         if (argv && old_argv) {
489                 memcpy(argv, old_argv, *size * sizeof(*argv));
490                 *size = new_size;
491         }
492
493         kfree(old_argv);
494         return argv;
495 }
496
497 /*
498  * Destructively splits up the argument list to pass to ctr.
499  */
500 int dm_split_args(int *argc, char ***argvp, char *input)
501 {
502         char *start, *end = input, *out, **argv = NULL;
503         unsigned array_size = 0;
504
505         *argc = 0;
506
507         if (!input) {
508                 *argvp = NULL;
509                 return 0;
510         }
511
512         argv = realloc_argv(&array_size, argv);
513         if (!argv)
514                 return -ENOMEM;
515
516         while (1) {
517                 /* Skip whitespace */
518                 start = skip_spaces(end);
519
520                 if (!*start)
521                         break;  /* success, we hit the end */
522
523                 /* 'out' is used to remove any back-quotes */
524                 end = out = start;
525                 while (*end) {
526                         /* Everything apart from '\0' can be quoted */
527                         if (*end == '\\' && *(end + 1)) {
528                                 *out++ = *(end + 1);
529                                 end += 2;
530                                 continue;
531                         }
532
533                         if (isspace(*end))
534                                 break;  /* end of token */
535
536                         *out++ = *end++;
537                 }
538
539                 /* have we already filled the array ? */
540                 if ((*argc + 1) > array_size) {
541                         argv = realloc_argv(&array_size, argv);
542                         if (!argv)
543                                 return -ENOMEM;
544                 }
545
546                 /* we know this is whitespace */
547                 if (*end)
548                         end++;
549
550                 /* terminate the string and put it in the array */
551                 *out = '\0';
552                 argv[*argc] = start;
553                 (*argc)++;
554         }
555
556         *argvp = argv;
557         return 0;
558 }
559
560 /*
561  * Impose necessary and sufficient conditions on a devices's table such
562  * that any incoming bio which respects its logical_block_size can be
563  * processed successfully.  If it falls across the boundary between
564  * two or more targets, the size of each piece it gets split into must
565  * be compatible with the logical_block_size of the target processing it.
566  */
567 static int validate_hardware_logical_block_alignment(struct dm_table *table,
568                                                  struct queue_limits *limits)
569 {
570         /*
571          * This function uses arithmetic modulo the logical_block_size
572          * (in units of 512-byte sectors).
573          */
574         unsigned short device_logical_block_size_sects =
575                 limits->logical_block_size >> SECTOR_SHIFT;
576
577         /*
578          * Offset of the start of the next table entry, mod logical_block_size.
579          */
580         unsigned short next_target_start = 0;
581
582         /*
583          * Given an aligned bio that extends beyond the end of a
584          * target, how many sectors must the next target handle?
585          */
586         unsigned short remaining = 0;
587
588         struct dm_target *ti;
589         struct queue_limits ti_limits;
590         unsigned i;
591
592         /*
593          * Check each entry in the table in turn.
594          */
595         for (i = 0; i < dm_table_get_num_targets(table); i++) {
596                 ti = dm_table_get_target(table, i);
597
598                 blk_set_stacking_limits(&ti_limits);
599
600                 /* combine all target devices' limits */
601                 if (ti->type->iterate_devices)
602                         ti->type->iterate_devices(ti, dm_set_device_limits,
603                                                   &ti_limits);
604
605                 /*
606                  * If the remaining sectors fall entirely within this
607                  * table entry are they compatible with its logical_block_size?
608                  */
609                 if (remaining < ti->len &&
610                     remaining & ((ti_limits.logical_block_size >>
611                                   SECTOR_SHIFT) - 1))
612                         break;  /* Error */
613
614                 next_target_start =
615                     (unsigned short) ((next_target_start + ti->len) &
616                                       (device_logical_block_size_sects - 1));
617                 remaining = next_target_start ?
618                     device_logical_block_size_sects - next_target_start : 0;
619         }
620
621         if (remaining) {
622                 DMWARN("%s: table line %u (start sect %llu len %llu) "
623                        "not aligned to h/w logical block size %u",
624                        dm_device_name(table->md), i,
625                        (unsigned long long) ti->begin,
626                        (unsigned long long) ti->len,
627                        limits->logical_block_size);
628                 return -EINVAL;
629         }
630
631         return 0;
632 }
633
634 int dm_table_add_target(struct dm_table *t, const char *type,
635                         sector_t start, sector_t len, char *params)
636 {
637         int r = -EINVAL, argc;
638         char **argv;
639         struct dm_target *tgt;
640
641         if (t->singleton) {
642                 DMERR("%s: target type %s must appear alone in table",
643                       dm_device_name(t->md), t->targets->type->name);
644                 return -EINVAL;
645         }
646
647         BUG_ON(t->num_targets >= t->num_allocated);
648
649         tgt = t->targets + t->num_targets;
650         memset(tgt, 0, sizeof(*tgt));
651
652         if (!len) {
653                 DMERR("%s: zero-length target", dm_device_name(t->md));
654                 return -EINVAL;
655         }
656
657         tgt->type = dm_get_target_type(type);
658         if (!tgt->type) {
659                 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
660                 return -EINVAL;
661         }
662
663         if (dm_target_needs_singleton(tgt->type)) {
664                 if (t->num_targets) {
665                         tgt->error = "singleton target type must appear alone in table";
666                         goto bad;
667                 }
668                 t->singleton = true;
669         }
670
671         if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
672                 tgt->error = "target type may not be included in a read-only table";
673                 goto bad;
674         }
675
676         if (t->immutable_target_type) {
677                 if (t->immutable_target_type != tgt->type) {
678                         tgt->error = "immutable target type cannot be mixed with other target types";
679                         goto bad;
680                 }
681         } else if (dm_target_is_immutable(tgt->type)) {
682                 if (t->num_targets) {
683                         tgt->error = "immutable target type cannot be mixed with other target types";
684                         goto bad;
685                 }
686                 t->immutable_target_type = tgt->type;
687         }
688
689         if (dm_target_has_integrity(tgt->type))
690                 t->integrity_added = 1;
691
692         tgt->table = t;
693         tgt->begin = start;
694         tgt->len = len;
695         tgt->error = "Unknown error";
696
697         /*
698          * Does this target adjoin the previous one ?
699          */
700         if (!adjoin(t, tgt)) {
701                 tgt->error = "Gap in table";
702                 goto bad;
703         }
704
705         r = dm_split_args(&argc, &argv, params);
706         if (r) {
707                 tgt->error = "couldn't split parameters";
708                 goto bad;
709         }
710
711         r = tgt->type->ctr(tgt, argc, argv);
712         kfree(argv);
713         if (r)
714                 goto bad;
715
716         t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
717
718         if (!tgt->num_discard_bios && tgt->discards_supported)
719                 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
720                        dm_device_name(t->md), type);
721
722         return 0;
723
724  bad:
725         DMERR("%s: %s: %s (%pe)", dm_device_name(t->md), type, tgt->error, ERR_PTR(r));
726         dm_put_target_type(tgt->type);
727         return r;
728 }
729
730 /*
731  * Target argument parsing helpers.
732  */
733 static int validate_next_arg(const struct dm_arg *arg,
734                              struct dm_arg_set *arg_set,
735                              unsigned *value, char **error, unsigned grouped)
736 {
737         const char *arg_str = dm_shift_arg(arg_set);
738         char dummy;
739
740         if (!arg_str ||
741             (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
742             (*value < arg->min) ||
743             (*value > arg->max) ||
744             (grouped && arg_set->argc < *value)) {
745                 *error = arg->error;
746                 return -EINVAL;
747         }
748
749         return 0;
750 }
751
752 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
753                 unsigned *value, char **error)
754 {
755         return validate_next_arg(arg, arg_set, value, error, 0);
756 }
757 EXPORT_SYMBOL(dm_read_arg);
758
759 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
760                       unsigned *value, char **error)
761 {
762         return validate_next_arg(arg, arg_set, value, error, 1);
763 }
764 EXPORT_SYMBOL(dm_read_arg_group);
765
766 const char *dm_shift_arg(struct dm_arg_set *as)
767 {
768         char *r;
769
770         if (as->argc) {
771                 as->argc--;
772                 r = *as->argv;
773                 as->argv++;
774                 return r;
775         }
776
777         return NULL;
778 }
779 EXPORT_SYMBOL(dm_shift_arg);
780
781 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
782 {
783         BUG_ON(as->argc < num_args);
784         as->argc -= num_args;
785         as->argv += num_args;
786 }
787 EXPORT_SYMBOL(dm_consume_args);
788
789 static bool __table_type_bio_based(enum dm_queue_mode table_type)
790 {
791         return (table_type == DM_TYPE_BIO_BASED ||
792                 table_type == DM_TYPE_DAX_BIO_BASED);
793 }
794
795 static bool __table_type_request_based(enum dm_queue_mode table_type)
796 {
797         return table_type == DM_TYPE_REQUEST_BASED;
798 }
799
800 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
801 {
802         t->type = type;
803 }
804 EXPORT_SYMBOL_GPL(dm_table_set_type);
805
806 /* validate the dax capability of the target device span */
807 static int device_not_dax_capable(struct dm_target *ti, struct dm_dev *dev,
808                         sector_t start, sector_t len, void *data)
809 {
810         if (dev->dax_dev)
811                 return false;
812
813         DMDEBUG("%pg: error: dax unsupported by block device", dev->bdev);
814         return true;
815 }
816
817 /* Check devices support synchronous DAX */
818 static int device_not_dax_synchronous_capable(struct dm_target *ti, struct dm_dev *dev,
819                                               sector_t start, sector_t len, void *data)
820 {
821         return !dev->dax_dev || !dax_synchronous(dev->dax_dev);
822 }
823
824 static bool dm_table_supports_dax(struct dm_table *t,
825                            iterate_devices_callout_fn iterate_fn)
826 {
827         struct dm_target *ti;
828         unsigned i;
829
830         /* Ensure that all targets support DAX. */
831         for (i = 0; i < dm_table_get_num_targets(t); i++) {
832                 ti = dm_table_get_target(t, i);
833
834                 if (!ti->type->direct_access)
835                         return false;
836
837                 if (!ti->type->iterate_devices ||
838                     ti->type->iterate_devices(ti, iterate_fn, NULL))
839                         return false;
840         }
841
842         return true;
843 }
844
845 static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
846                                   sector_t start, sector_t len, void *data)
847 {
848         struct block_device *bdev = dev->bdev;
849         struct request_queue *q = bdev_get_queue(bdev);
850
851         /* request-based cannot stack on partitions! */
852         if (bdev_is_partition(bdev))
853                 return false;
854
855         return queue_is_mq(q);
856 }
857
858 static int dm_table_determine_type(struct dm_table *t)
859 {
860         unsigned i;
861         unsigned bio_based = 0, request_based = 0, hybrid = 0;
862         struct dm_target *tgt;
863         struct list_head *devices = dm_table_get_devices(t);
864         enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
865
866         if (t->type != DM_TYPE_NONE) {
867                 /* target already set the table's type */
868                 if (t->type == DM_TYPE_BIO_BASED) {
869                         /* possibly upgrade to a variant of bio-based */
870                         goto verify_bio_based;
871                 }
872                 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
873                 goto verify_rq_based;
874         }
875
876         for (i = 0; i < t->num_targets; i++) {
877                 tgt = t->targets + i;
878                 if (dm_target_hybrid(tgt))
879                         hybrid = 1;
880                 else if (dm_target_request_based(tgt))
881                         request_based = 1;
882                 else
883                         bio_based = 1;
884
885                 if (bio_based && request_based) {
886                         DMERR("Inconsistent table: different target types"
887                               " can't be mixed up");
888                         return -EINVAL;
889                 }
890         }
891
892         if (hybrid && !bio_based && !request_based) {
893                 /*
894                  * The targets can work either way.
895                  * Determine the type from the live device.
896                  * Default to bio-based if device is new.
897                  */
898                 if (__table_type_request_based(live_md_type))
899                         request_based = 1;
900                 else
901                         bio_based = 1;
902         }
903
904         if (bio_based) {
905 verify_bio_based:
906                 /* We must use this table as bio-based */
907                 t->type = DM_TYPE_BIO_BASED;
908                 if (dm_table_supports_dax(t, device_not_dax_capable) ||
909                     (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
910                         t->type = DM_TYPE_DAX_BIO_BASED;
911                 }
912                 return 0;
913         }
914
915         BUG_ON(!request_based); /* No targets in this table */
916
917         t->type = DM_TYPE_REQUEST_BASED;
918
919 verify_rq_based:
920         /*
921          * Request-based dm supports only tables that have a single target now.
922          * To support multiple targets, request splitting support is needed,
923          * and that needs lots of changes in the block-layer.
924          * (e.g. request completion process for partial completion.)
925          */
926         if (t->num_targets > 1) {
927                 DMERR("request-based DM doesn't support multiple targets");
928                 return -EINVAL;
929         }
930
931         if (list_empty(devices)) {
932                 int srcu_idx;
933                 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
934
935                 /* inherit live table's type */
936                 if (live_table)
937                         t->type = live_table->type;
938                 dm_put_live_table(t->md, srcu_idx);
939                 return 0;
940         }
941
942         tgt = dm_table_get_immutable_target(t);
943         if (!tgt) {
944                 DMERR("table load rejected: immutable target is required");
945                 return -EINVAL;
946         } else if (tgt->max_io_len) {
947                 DMERR("table load rejected: immutable target that splits IO is not supported");
948                 return -EINVAL;
949         }
950
951         /* Non-request-stackable devices can't be used for request-based dm */
952         if (!tgt->type->iterate_devices ||
953             !tgt->type->iterate_devices(tgt, device_is_rq_stackable, NULL)) {
954                 DMERR("table load rejected: including non-request-stackable devices");
955                 return -EINVAL;
956         }
957
958         return 0;
959 }
960
961 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
962 {
963         return t->type;
964 }
965
966 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
967 {
968         return t->immutable_target_type;
969 }
970
971 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
972 {
973         /* Immutable target is implicitly a singleton */
974         if (t->num_targets > 1 ||
975             !dm_target_is_immutable(t->targets[0].type))
976                 return NULL;
977
978         return t->targets;
979 }
980
981 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
982 {
983         struct dm_target *ti;
984         unsigned i;
985
986         for (i = 0; i < dm_table_get_num_targets(t); i++) {
987                 ti = dm_table_get_target(t, i);
988                 if (dm_target_is_wildcard(ti->type))
989                         return ti;
990         }
991
992         return NULL;
993 }
994
995 bool dm_table_bio_based(struct dm_table *t)
996 {
997         return __table_type_bio_based(dm_table_get_type(t));
998 }
999
1000 bool dm_table_request_based(struct dm_table *t)
1001 {
1002         return __table_type_request_based(dm_table_get_type(t));
1003 }
1004
1005 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1006 {
1007         enum dm_queue_mode type = dm_table_get_type(t);
1008         unsigned per_io_data_size = 0;
1009         unsigned min_pool_size = 0;
1010         struct dm_target *ti;
1011         unsigned i;
1012
1013         if (unlikely(type == DM_TYPE_NONE)) {
1014                 DMWARN("no table type is set, can't allocate mempools");
1015                 return -EINVAL;
1016         }
1017
1018         if (__table_type_bio_based(type))
1019                 for (i = 0; i < t->num_targets; i++) {
1020                         ti = t->targets + i;
1021                         per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1022                         min_pool_size = max(min_pool_size, ti->num_flush_bios);
1023                 }
1024
1025         t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1026                                            per_io_data_size, min_pool_size);
1027         if (!t->mempools)
1028                 return -ENOMEM;
1029
1030         return 0;
1031 }
1032
1033 void dm_table_free_md_mempools(struct dm_table *t)
1034 {
1035         dm_free_md_mempools(t->mempools);
1036         t->mempools = NULL;
1037 }
1038
1039 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1040 {
1041         return t->mempools;
1042 }
1043
1044 static int setup_indexes(struct dm_table *t)
1045 {
1046         int i;
1047         unsigned int total = 0;
1048         sector_t *indexes;
1049
1050         /* allocate the space for *all* the indexes */
1051         for (i = t->depth - 2; i >= 0; i--) {
1052                 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1053                 total += t->counts[i];
1054         }
1055
1056         indexes = kvcalloc(total, NODE_SIZE, GFP_KERNEL);
1057         if (!indexes)
1058                 return -ENOMEM;
1059
1060         /* set up internal nodes, bottom-up */
1061         for (i = t->depth - 2; i >= 0; i--) {
1062                 t->index[i] = indexes;
1063                 indexes += (KEYS_PER_NODE * t->counts[i]);
1064                 setup_btree_index(i, t);
1065         }
1066
1067         return 0;
1068 }
1069
1070 /*
1071  * Builds the btree to index the map.
1072  */
1073 static int dm_table_build_index(struct dm_table *t)
1074 {
1075         int r = 0;
1076         unsigned int leaf_nodes;
1077
1078         /* how many indexes will the btree have ? */
1079         leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1080         t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1081
1082         /* leaf layer has already been set up */
1083         t->counts[t->depth - 1] = leaf_nodes;
1084         t->index[t->depth - 1] = t->highs;
1085
1086         if (t->depth >= 2)
1087                 r = setup_indexes(t);
1088
1089         return r;
1090 }
1091
1092 static bool integrity_profile_exists(struct gendisk *disk)
1093 {
1094         return !!blk_get_integrity(disk);
1095 }
1096
1097 /*
1098  * Get a disk whose integrity profile reflects the table's profile.
1099  * Returns NULL if integrity support was inconsistent or unavailable.
1100  */
1101 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1102 {
1103         struct list_head *devices = dm_table_get_devices(t);
1104         struct dm_dev_internal *dd = NULL;
1105         struct gendisk *prev_disk = NULL, *template_disk = NULL;
1106         unsigned i;
1107
1108         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1109                 struct dm_target *ti = dm_table_get_target(t, i);
1110                 if (!dm_target_passes_integrity(ti->type))
1111                         goto no_integrity;
1112         }
1113
1114         list_for_each_entry(dd, devices, list) {
1115                 template_disk = dd->dm_dev->bdev->bd_disk;
1116                 if (!integrity_profile_exists(template_disk))
1117                         goto no_integrity;
1118                 else if (prev_disk &&
1119                          blk_integrity_compare(prev_disk, template_disk) < 0)
1120                         goto no_integrity;
1121                 prev_disk = template_disk;
1122         }
1123
1124         return template_disk;
1125
1126 no_integrity:
1127         if (prev_disk)
1128                 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1129                        dm_device_name(t->md),
1130                        prev_disk->disk_name,
1131                        template_disk->disk_name);
1132         return NULL;
1133 }
1134
1135 /*
1136  * Register the mapped device for blk_integrity support if the
1137  * underlying devices have an integrity profile.  But all devices may
1138  * not have matching profiles (checking all devices isn't reliable
1139  * during table load because this table may use other DM device(s) which
1140  * must be resumed before they will have an initialized integity
1141  * profile).  Consequently, stacked DM devices force a 2 stage integrity
1142  * profile validation: First pass during table load, final pass during
1143  * resume.
1144  */
1145 static int dm_table_register_integrity(struct dm_table *t)
1146 {
1147         struct mapped_device *md = t->md;
1148         struct gendisk *template_disk = NULL;
1149
1150         /* If target handles integrity itself do not register it here. */
1151         if (t->integrity_added)
1152                 return 0;
1153
1154         template_disk = dm_table_get_integrity_disk(t);
1155         if (!template_disk)
1156                 return 0;
1157
1158         if (!integrity_profile_exists(dm_disk(md))) {
1159                 t->integrity_supported = true;
1160                 /*
1161                  * Register integrity profile during table load; we can do
1162                  * this because the final profile must match during resume.
1163                  */
1164                 blk_integrity_register(dm_disk(md),
1165                                        blk_get_integrity(template_disk));
1166                 return 0;
1167         }
1168
1169         /*
1170          * If DM device already has an initialized integrity
1171          * profile the new profile should not conflict.
1172          */
1173         if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1174                 DMWARN("%s: conflict with existing integrity profile: "
1175                        "%s profile mismatch",
1176                        dm_device_name(t->md),
1177                        template_disk->disk_name);
1178                 return 1;
1179         }
1180
1181         /* Preserve existing integrity profile */
1182         t->integrity_supported = true;
1183         return 0;
1184 }
1185
1186 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1187
1188 struct dm_crypto_profile {
1189         struct blk_crypto_profile profile;
1190         struct mapped_device *md;
1191 };
1192
1193 struct dm_keyslot_evict_args {
1194         const struct blk_crypto_key *key;
1195         int err;
1196 };
1197
1198 static int dm_keyslot_evict_callback(struct dm_target *ti, struct dm_dev *dev,
1199                                      sector_t start, sector_t len, void *data)
1200 {
1201         struct dm_keyslot_evict_args *args = data;
1202         int err;
1203
1204         err = blk_crypto_evict_key(bdev_get_queue(dev->bdev), args->key);
1205         if (!args->err)
1206                 args->err = err;
1207         /* Always try to evict the key from all devices. */
1208         return 0;
1209 }
1210
1211 /*
1212  * When an inline encryption key is evicted from a device-mapper device, evict
1213  * it from all the underlying devices.
1214  */
1215 static int dm_keyslot_evict(struct blk_crypto_profile *profile,
1216                             const struct blk_crypto_key *key, unsigned int slot)
1217 {
1218         struct mapped_device *md =
1219                 container_of(profile, struct dm_crypto_profile, profile)->md;
1220         struct dm_keyslot_evict_args args = { key };
1221         struct dm_table *t;
1222         int srcu_idx;
1223         int i;
1224         struct dm_target *ti;
1225
1226         t = dm_get_live_table(md, &srcu_idx);
1227         if (!t)
1228                 return 0;
1229         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1230                 ti = dm_table_get_target(t, i);
1231                 if (!ti->type->iterate_devices)
1232                         continue;
1233                 ti->type->iterate_devices(ti, dm_keyslot_evict_callback, &args);
1234         }
1235         dm_put_live_table(md, srcu_idx);
1236         return args.err;
1237 }
1238
1239 static int
1240 device_intersect_crypto_capabilities(struct dm_target *ti, struct dm_dev *dev,
1241                                      sector_t start, sector_t len, void *data)
1242 {
1243         struct blk_crypto_profile *parent = data;
1244         struct blk_crypto_profile *child =
1245                 bdev_get_queue(dev->bdev)->crypto_profile;
1246
1247         blk_crypto_intersect_capabilities(parent, child);
1248         return 0;
1249 }
1250
1251 void dm_destroy_crypto_profile(struct blk_crypto_profile *profile)
1252 {
1253         struct dm_crypto_profile *dmcp = container_of(profile,
1254                                                       struct dm_crypto_profile,
1255                                                       profile);
1256
1257         if (!profile)
1258                 return;
1259
1260         blk_crypto_profile_destroy(profile);
1261         kfree(dmcp);
1262 }
1263
1264 static void dm_table_destroy_crypto_profile(struct dm_table *t)
1265 {
1266         dm_destroy_crypto_profile(t->crypto_profile);
1267         t->crypto_profile = NULL;
1268 }
1269
1270 /*
1271  * Constructs and initializes t->crypto_profile with a crypto profile that
1272  * represents the common set of crypto capabilities of the devices described by
1273  * the dm_table.  However, if the constructed crypto profile doesn't support all
1274  * crypto capabilities that are supported by the current mapped_device, it
1275  * returns an error instead, since we don't support removing crypto capabilities
1276  * on table changes.  Finally, if the constructed crypto profile is "empty" (has
1277  * no crypto capabilities at all), it just sets t->crypto_profile to NULL.
1278  */
1279 static int dm_table_construct_crypto_profile(struct dm_table *t)
1280 {
1281         struct dm_crypto_profile *dmcp;
1282         struct blk_crypto_profile *profile;
1283         struct dm_target *ti;
1284         unsigned int i;
1285         bool empty_profile = true;
1286
1287         dmcp = kmalloc(sizeof(*dmcp), GFP_KERNEL);
1288         if (!dmcp)
1289                 return -ENOMEM;
1290         dmcp->md = t->md;
1291
1292         profile = &dmcp->profile;
1293         blk_crypto_profile_init(profile, 0);
1294         profile->ll_ops.keyslot_evict = dm_keyslot_evict;
1295         profile->max_dun_bytes_supported = UINT_MAX;
1296         memset(profile->modes_supported, 0xFF,
1297                sizeof(profile->modes_supported));
1298
1299         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1300                 ti = dm_table_get_target(t, i);
1301
1302                 if (!dm_target_passes_crypto(ti->type)) {
1303                         blk_crypto_intersect_capabilities(profile, NULL);
1304                         break;
1305                 }
1306                 if (!ti->type->iterate_devices)
1307                         continue;
1308                 ti->type->iterate_devices(ti,
1309                                           device_intersect_crypto_capabilities,
1310                                           profile);
1311         }
1312
1313         if (t->md->queue &&
1314             !blk_crypto_has_capabilities(profile,
1315                                          t->md->queue->crypto_profile)) {
1316                 DMWARN("Inline encryption capabilities of new DM table were more restrictive than the old table's. This is not supported!");
1317                 dm_destroy_crypto_profile(profile);
1318                 return -EINVAL;
1319         }
1320
1321         /*
1322          * If the new profile doesn't actually support any crypto capabilities,
1323          * we may as well represent it with a NULL profile.
1324          */
1325         for (i = 0; i < ARRAY_SIZE(profile->modes_supported); i++) {
1326                 if (profile->modes_supported[i]) {
1327                         empty_profile = false;
1328                         break;
1329                 }
1330         }
1331
1332         if (empty_profile) {
1333                 dm_destroy_crypto_profile(profile);
1334                 profile = NULL;
1335         }
1336
1337         /*
1338          * t->crypto_profile is only set temporarily while the table is being
1339          * set up, and it gets set to NULL after the profile has been
1340          * transferred to the request_queue.
1341          */
1342         t->crypto_profile = profile;
1343
1344         return 0;
1345 }
1346
1347 static void dm_update_crypto_profile(struct request_queue *q,
1348                                      struct dm_table *t)
1349 {
1350         if (!t->crypto_profile)
1351                 return;
1352
1353         /* Make the crypto profile less restrictive. */
1354         if (!q->crypto_profile) {
1355                 blk_crypto_register(t->crypto_profile, q);
1356         } else {
1357                 blk_crypto_update_capabilities(q->crypto_profile,
1358                                                t->crypto_profile);
1359                 dm_destroy_crypto_profile(t->crypto_profile);
1360         }
1361         t->crypto_profile = NULL;
1362 }
1363
1364 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1365
1366 static int dm_table_construct_crypto_profile(struct dm_table *t)
1367 {
1368         return 0;
1369 }
1370
1371 void dm_destroy_crypto_profile(struct blk_crypto_profile *profile)
1372 {
1373 }
1374
1375 static void dm_table_destroy_crypto_profile(struct dm_table *t)
1376 {
1377 }
1378
1379 static void dm_update_crypto_profile(struct request_queue *q,
1380                                      struct dm_table *t)
1381 {
1382 }
1383
1384 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1385
1386 /*
1387  * Prepares the table for use by building the indices,
1388  * setting the type, and allocating mempools.
1389  */
1390 int dm_table_complete(struct dm_table *t)
1391 {
1392         int r;
1393
1394         r = dm_table_determine_type(t);
1395         if (r) {
1396                 DMERR("unable to determine table type");
1397                 return r;
1398         }
1399
1400         r = dm_table_build_index(t);
1401         if (r) {
1402                 DMERR("unable to build btrees");
1403                 return r;
1404         }
1405
1406         r = dm_table_register_integrity(t);
1407         if (r) {
1408                 DMERR("could not register integrity profile.");
1409                 return r;
1410         }
1411
1412         r = dm_table_construct_crypto_profile(t);
1413         if (r) {
1414                 DMERR("could not construct crypto profile.");
1415                 return r;
1416         }
1417
1418         r = dm_table_alloc_md_mempools(t, t->md);
1419         if (r)
1420                 DMERR("unable to allocate mempools");
1421
1422         return r;
1423 }
1424
1425 static DEFINE_MUTEX(_event_lock);
1426 void dm_table_event_callback(struct dm_table *t,
1427                              void (*fn)(void *), void *context)
1428 {
1429         mutex_lock(&_event_lock);
1430         t->event_fn = fn;
1431         t->event_context = context;
1432         mutex_unlock(&_event_lock);
1433 }
1434
1435 void dm_table_event(struct dm_table *t)
1436 {
1437         mutex_lock(&_event_lock);
1438         if (t->event_fn)
1439                 t->event_fn(t->event_context);
1440         mutex_unlock(&_event_lock);
1441 }
1442 EXPORT_SYMBOL(dm_table_event);
1443
1444 inline sector_t dm_table_get_size(struct dm_table *t)
1445 {
1446         return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1447 }
1448 EXPORT_SYMBOL(dm_table_get_size);
1449
1450 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1451 {
1452         if (index >= t->num_targets)
1453                 return NULL;
1454
1455         return t->targets + index;
1456 }
1457
1458 /*
1459  * Search the btree for the correct target.
1460  *
1461  * Caller should check returned pointer for NULL
1462  * to trap I/O beyond end of device.
1463  */
1464 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1465 {
1466         unsigned int l, n = 0, k = 0;
1467         sector_t *node;
1468
1469         if (unlikely(sector >= dm_table_get_size(t)))
1470                 return NULL;
1471
1472         for (l = 0; l < t->depth; l++) {
1473                 n = get_child(n, k);
1474                 node = get_node(t, l, n);
1475
1476                 for (k = 0; k < KEYS_PER_NODE; k++)
1477                         if (node[k] >= sector)
1478                                 break;
1479         }
1480
1481         return &t->targets[(KEYS_PER_NODE * n) + k];
1482 }
1483
1484 static int device_not_poll_capable(struct dm_target *ti, struct dm_dev *dev,
1485                                    sector_t start, sector_t len, void *data)
1486 {
1487         struct request_queue *q = bdev_get_queue(dev->bdev);
1488
1489         return !test_bit(QUEUE_FLAG_POLL, &q->queue_flags);
1490 }
1491
1492 /*
1493  * type->iterate_devices() should be called when the sanity check needs to
1494  * iterate and check all underlying data devices. iterate_devices() will
1495  * iterate all underlying data devices until it encounters a non-zero return
1496  * code, returned by whether the input iterate_devices_callout_fn, or
1497  * iterate_devices() itself internally.
1498  *
1499  * For some target type (e.g. dm-stripe), one call of iterate_devices() may
1500  * iterate multiple underlying devices internally, in which case a non-zero
1501  * return code returned by iterate_devices_callout_fn will stop the iteration
1502  * in advance.
1503  *
1504  * Cases requiring _any_ underlying device supporting some kind of attribute,
1505  * should use the iteration structure like dm_table_any_dev_attr(), or call
1506  * it directly. @func should handle semantics of positive examples, e.g.
1507  * capable of something.
1508  *
1509  * Cases requiring _all_ underlying devices supporting some kind of attribute,
1510  * should use the iteration structure like dm_table_supports_nowait() or
1511  * dm_table_supports_discards(). Or introduce dm_table_all_devs_attr() that
1512  * uses an @anti_func that handle semantics of counter examples, e.g. not
1513  * capable of something. So: return !dm_table_any_dev_attr(t, anti_func, data);
1514  */
1515 static bool dm_table_any_dev_attr(struct dm_table *t,
1516                                   iterate_devices_callout_fn func, void *data)
1517 {
1518         struct dm_target *ti;
1519         unsigned int i;
1520
1521         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1522                 ti = dm_table_get_target(t, i);
1523
1524                 if (ti->type->iterate_devices &&
1525                     ti->type->iterate_devices(ti, func, data))
1526                         return true;
1527         }
1528
1529         return false;
1530 }
1531
1532 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1533                         sector_t start, sector_t len, void *data)
1534 {
1535         unsigned *num_devices = data;
1536
1537         (*num_devices)++;
1538
1539         return 0;
1540 }
1541
1542 static int dm_table_supports_poll(struct dm_table *t)
1543 {
1544         return !dm_table_any_dev_attr(t, device_not_poll_capable, NULL);
1545 }
1546
1547 /*
1548  * Check whether a table has no data devices attached using each
1549  * target's iterate_devices method.
1550  * Returns false if the result is unknown because a target doesn't
1551  * support iterate_devices.
1552  */
1553 bool dm_table_has_no_data_devices(struct dm_table *table)
1554 {
1555         struct dm_target *ti;
1556         unsigned i, num_devices;
1557
1558         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1559                 ti = dm_table_get_target(table, i);
1560
1561                 if (!ti->type->iterate_devices)
1562                         return false;
1563
1564                 num_devices = 0;
1565                 ti->type->iterate_devices(ti, count_device, &num_devices);
1566                 if (num_devices)
1567                         return false;
1568         }
1569
1570         return true;
1571 }
1572
1573 static int device_not_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1574                                   sector_t start, sector_t len, void *data)
1575 {
1576         struct request_queue *q = bdev_get_queue(dev->bdev);
1577         enum blk_zoned_model *zoned_model = data;
1578
1579         return blk_queue_zoned_model(q) != *zoned_model;
1580 }
1581
1582 /*
1583  * Check the device zoned model based on the target feature flag. If the target
1584  * has the DM_TARGET_ZONED_HM feature flag set, host-managed zoned devices are
1585  * also accepted but all devices must have the same zoned model. If the target
1586  * has the DM_TARGET_MIXED_ZONED_MODEL feature set, the devices can have any
1587  * zoned model with all zoned devices having the same zone size.
1588  */
1589 static bool dm_table_supports_zoned_model(struct dm_table *t,
1590                                           enum blk_zoned_model zoned_model)
1591 {
1592         struct dm_target *ti;
1593         unsigned i;
1594
1595         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1596                 ti = dm_table_get_target(t, i);
1597
1598                 if (dm_target_supports_zoned_hm(ti->type)) {
1599                         if (!ti->type->iterate_devices ||
1600                             ti->type->iterate_devices(ti, device_not_zoned_model,
1601                                                       &zoned_model))
1602                                 return false;
1603                 } else if (!dm_target_supports_mixed_zoned_model(ti->type)) {
1604                         if (zoned_model == BLK_ZONED_HM)
1605                                 return false;
1606                 }
1607         }
1608
1609         return true;
1610 }
1611
1612 static int device_not_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1613                                            sector_t start, sector_t len, void *data)
1614 {
1615         struct request_queue *q = bdev_get_queue(dev->bdev);
1616         unsigned int *zone_sectors = data;
1617
1618         if (!blk_queue_is_zoned(q))
1619                 return 0;
1620
1621         return blk_queue_zone_sectors(q) != *zone_sectors;
1622 }
1623
1624 /*
1625  * Check consistency of zoned model and zone sectors across all targets. For
1626  * zone sectors, if the destination device is a zoned block device, it shall
1627  * have the specified zone_sectors.
1628  */
1629 static int validate_hardware_zoned_model(struct dm_table *table,
1630                                          enum blk_zoned_model zoned_model,
1631                                          unsigned int zone_sectors)
1632 {
1633         if (zoned_model == BLK_ZONED_NONE)
1634                 return 0;
1635
1636         if (!dm_table_supports_zoned_model(table, zoned_model)) {
1637                 DMERR("%s: zoned model is not consistent across all devices",
1638                       dm_device_name(table->md));
1639                 return -EINVAL;
1640         }
1641
1642         /* Check zone size validity and compatibility */
1643         if (!zone_sectors || !is_power_of_2(zone_sectors))
1644                 return -EINVAL;
1645
1646         if (dm_table_any_dev_attr(table, device_not_matches_zone_sectors, &zone_sectors)) {
1647                 DMERR("%s: zone sectors is not consistent across all zoned devices",
1648                       dm_device_name(table->md));
1649                 return -EINVAL;
1650         }
1651
1652         return 0;
1653 }
1654
1655 /*
1656  * Establish the new table's queue_limits and validate them.
1657  */
1658 int dm_calculate_queue_limits(struct dm_table *table,
1659                               struct queue_limits *limits)
1660 {
1661         struct dm_target *ti;
1662         struct queue_limits ti_limits;
1663         unsigned i;
1664         enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1665         unsigned int zone_sectors = 0;
1666
1667         blk_set_stacking_limits(limits);
1668
1669         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1670                 blk_set_stacking_limits(&ti_limits);
1671
1672                 ti = dm_table_get_target(table, i);
1673
1674                 if (!ti->type->iterate_devices)
1675                         goto combine_limits;
1676
1677                 /*
1678                  * Combine queue limits of all the devices this target uses.
1679                  */
1680                 ti->type->iterate_devices(ti, dm_set_device_limits,
1681                                           &ti_limits);
1682
1683                 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1684                         /*
1685                          * After stacking all limits, validate all devices
1686                          * in table support this zoned model and zone sectors.
1687                          */
1688                         zoned_model = ti_limits.zoned;
1689                         zone_sectors = ti_limits.chunk_sectors;
1690                 }
1691
1692                 /* Set I/O hints portion of queue limits */
1693                 if (ti->type->io_hints)
1694                         ti->type->io_hints(ti, &ti_limits);
1695
1696                 /*
1697                  * Check each device area is consistent with the target's
1698                  * overall queue limits.
1699                  */
1700                 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1701                                               &ti_limits))
1702                         return -EINVAL;
1703
1704 combine_limits:
1705                 /*
1706                  * Merge this target's queue limits into the overall limits
1707                  * for the table.
1708                  */
1709                 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1710                         DMWARN("%s: adding target device "
1711                                "(start sect %llu len %llu) "
1712                                "caused an alignment inconsistency",
1713                                dm_device_name(table->md),
1714                                (unsigned long long) ti->begin,
1715                                (unsigned long long) ti->len);
1716         }
1717
1718         /*
1719          * Verify that the zoned model and zone sectors, as determined before
1720          * any .io_hints override, are the same across all devices in the table.
1721          * - this is especially relevant if .io_hints is emulating a disk-managed
1722          *   zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1723          * BUT...
1724          */
1725         if (limits->zoned != BLK_ZONED_NONE) {
1726                 /*
1727                  * ...IF the above limits stacking determined a zoned model
1728                  * validate that all of the table's devices conform to it.
1729                  */
1730                 zoned_model = limits->zoned;
1731                 zone_sectors = limits->chunk_sectors;
1732         }
1733         if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1734                 return -EINVAL;
1735
1736         return validate_hardware_logical_block_alignment(table, limits);
1737 }
1738
1739 /*
1740  * Verify that all devices have an integrity profile that matches the
1741  * DM device's registered integrity profile.  If the profiles don't
1742  * match then unregister the DM device's integrity profile.
1743  */
1744 static void dm_table_verify_integrity(struct dm_table *t)
1745 {
1746         struct gendisk *template_disk = NULL;
1747
1748         if (t->integrity_added)
1749                 return;
1750
1751         if (t->integrity_supported) {
1752                 /*
1753                  * Verify that the original integrity profile
1754                  * matches all the devices in this table.
1755                  */
1756                 template_disk = dm_table_get_integrity_disk(t);
1757                 if (template_disk &&
1758                     blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1759                         return;
1760         }
1761
1762         if (integrity_profile_exists(dm_disk(t->md))) {
1763                 DMWARN("%s: unable to establish an integrity profile",
1764                        dm_device_name(t->md));
1765                 blk_integrity_unregister(dm_disk(t->md));
1766         }
1767 }
1768
1769 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1770                                 sector_t start, sector_t len, void *data)
1771 {
1772         unsigned long flush = (unsigned long) data;
1773         struct request_queue *q = bdev_get_queue(dev->bdev);
1774
1775         return (q->queue_flags & flush);
1776 }
1777
1778 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1779 {
1780         struct dm_target *ti;
1781         unsigned i;
1782
1783         /*
1784          * Require at least one underlying device to support flushes.
1785          * t->devices includes internal dm devices such as mirror logs
1786          * so we need to use iterate_devices here, which targets
1787          * supporting flushes must provide.
1788          */
1789         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1790                 ti = dm_table_get_target(t, i);
1791
1792                 if (!ti->num_flush_bios)
1793                         continue;
1794
1795                 if (ti->flush_supported)
1796                         return true;
1797
1798                 if (ti->type->iterate_devices &&
1799                     ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1800                         return true;
1801         }
1802
1803         return false;
1804 }
1805
1806 static int device_dax_write_cache_enabled(struct dm_target *ti,
1807                                           struct dm_dev *dev, sector_t start,
1808                                           sector_t len, void *data)
1809 {
1810         struct dax_device *dax_dev = dev->dax_dev;
1811
1812         if (!dax_dev)
1813                 return false;
1814
1815         if (dax_write_cache_enabled(dax_dev))
1816                 return true;
1817         return false;
1818 }
1819
1820 static int device_is_rotational(struct dm_target *ti, struct dm_dev *dev,
1821                                 sector_t start, sector_t len, void *data)
1822 {
1823         struct request_queue *q = bdev_get_queue(dev->bdev);
1824
1825         return !blk_queue_nonrot(q);
1826 }
1827
1828 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1829                              sector_t start, sector_t len, void *data)
1830 {
1831         struct request_queue *q = bdev_get_queue(dev->bdev);
1832
1833         return !blk_queue_add_random(q);
1834 }
1835
1836 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1837                                            sector_t start, sector_t len, void *data)
1838 {
1839         struct request_queue *q = bdev_get_queue(dev->bdev);
1840
1841         return !q->limits.max_write_zeroes_sectors;
1842 }
1843
1844 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1845 {
1846         struct dm_target *ti;
1847         unsigned i = 0;
1848
1849         while (i < dm_table_get_num_targets(t)) {
1850                 ti = dm_table_get_target(t, i++);
1851
1852                 if (!ti->num_write_zeroes_bios)
1853                         return false;
1854
1855                 if (!ti->type->iterate_devices ||
1856                     ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1857                         return false;
1858         }
1859
1860         return true;
1861 }
1862
1863 static int device_not_nowait_capable(struct dm_target *ti, struct dm_dev *dev,
1864                                      sector_t start, sector_t len, void *data)
1865 {
1866         struct request_queue *q = bdev_get_queue(dev->bdev);
1867
1868         return !blk_queue_nowait(q);
1869 }
1870
1871 static bool dm_table_supports_nowait(struct dm_table *t)
1872 {
1873         struct dm_target *ti;
1874         unsigned i = 0;
1875
1876         while (i < dm_table_get_num_targets(t)) {
1877                 ti = dm_table_get_target(t, i++);
1878
1879                 if (!dm_target_supports_nowait(ti->type))
1880                         return false;
1881
1882                 if (!ti->type->iterate_devices ||
1883                     ti->type->iterate_devices(ti, device_not_nowait_capable, NULL))
1884                         return false;
1885         }
1886
1887         return true;
1888 }
1889
1890 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1891                                       sector_t start, sector_t len, void *data)
1892 {
1893         struct request_queue *q = bdev_get_queue(dev->bdev);
1894
1895         return !blk_queue_discard(q);
1896 }
1897
1898 static bool dm_table_supports_discards(struct dm_table *t)
1899 {
1900         struct dm_target *ti;
1901         unsigned i;
1902
1903         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1904                 ti = dm_table_get_target(t, i);
1905
1906                 if (!ti->num_discard_bios)
1907                         return false;
1908
1909                 /*
1910                  * Either the target provides discard support (as implied by setting
1911                  * 'discards_supported') or it relies on _all_ data devices having
1912                  * discard support.
1913                  */
1914                 if (!ti->discards_supported &&
1915                     (!ti->type->iterate_devices ||
1916                      ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1917                         return false;
1918         }
1919
1920         return true;
1921 }
1922
1923 static int device_not_secure_erase_capable(struct dm_target *ti,
1924                                            struct dm_dev *dev, sector_t start,
1925                                            sector_t len, void *data)
1926 {
1927         struct request_queue *q = bdev_get_queue(dev->bdev);
1928
1929         return !blk_queue_secure_erase(q);
1930 }
1931
1932 static bool dm_table_supports_secure_erase(struct dm_table *t)
1933 {
1934         struct dm_target *ti;
1935         unsigned int i;
1936
1937         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1938                 ti = dm_table_get_target(t, i);
1939
1940                 if (!ti->num_secure_erase_bios)
1941                         return false;
1942
1943                 if (!ti->type->iterate_devices ||
1944                     ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1945                         return false;
1946         }
1947
1948         return true;
1949 }
1950
1951 static int device_requires_stable_pages(struct dm_target *ti,
1952                                         struct dm_dev *dev, sector_t start,
1953                                         sector_t len, void *data)
1954 {
1955         struct request_queue *q = bdev_get_queue(dev->bdev);
1956
1957         return blk_queue_stable_writes(q);
1958 }
1959
1960 int dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1961                               struct queue_limits *limits)
1962 {
1963         bool wc = false, fua = false;
1964         int r;
1965
1966         /*
1967          * Copy table's limits to the DM device's request_queue
1968          */
1969         q->limits = *limits;
1970
1971         if (dm_table_supports_nowait(t))
1972                 blk_queue_flag_set(QUEUE_FLAG_NOWAIT, q);
1973         else
1974                 blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, q);
1975
1976         if (!dm_table_supports_discards(t)) {
1977                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
1978                 /* Must also clear discard limits... */
1979                 q->limits.max_discard_sectors = 0;
1980                 q->limits.max_hw_discard_sectors = 0;
1981                 q->limits.discard_granularity = 0;
1982                 q->limits.discard_alignment = 0;
1983                 q->limits.discard_misaligned = 0;
1984         } else
1985                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
1986
1987         if (dm_table_supports_secure_erase(t))
1988                 blk_queue_flag_set(QUEUE_FLAG_SECERASE, q);
1989
1990         if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1991                 wc = true;
1992                 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1993                         fua = true;
1994         }
1995         blk_queue_write_cache(q, wc, fua);
1996
1997         if (dm_table_supports_dax(t, device_not_dax_capable)) {
1998                 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
1999                 if (dm_table_supports_dax(t, device_not_dax_synchronous_capable))
2000                         set_dax_synchronous(t->md->dax_dev);
2001         }
2002         else
2003                 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
2004
2005         if (dm_table_any_dev_attr(t, device_dax_write_cache_enabled, NULL))
2006                 dax_write_cache(t->md->dax_dev, true);
2007
2008         /* Ensure that all underlying devices are non-rotational. */
2009         if (dm_table_any_dev_attr(t, device_is_rotational, NULL))
2010                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
2011         else
2012                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2013
2014         if (!dm_table_supports_write_zeroes(t))
2015                 q->limits.max_write_zeroes_sectors = 0;
2016
2017         dm_table_verify_integrity(t);
2018
2019         /*
2020          * Some devices don't use blk_integrity but still want stable pages
2021          * because they do their own checksumming.
2022          * If any underlying device requires stable pages, a table must require
2023          * them as well.  Only targets that support iterate_devices are considered:
2024          * don't want error, zero, etc to require stable pages.
2025          */
2026         if (dm_table_any_dev_attr(t, device_requires_stable_pages, NULL))
2027                 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q);
2028         else
2029                 blk_queue_flag_clear(QUEUE_FLAG_STABLE_WRITES, q);
2030
2031         /*
2032          * Determine whether or not this queue's I/O timings contribute
2033          * to the entropy pool, Only request-based targets use this.
2034          * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
2035          * have it set.
2036          */
2037         if (blk_queue_add_random(q) &&
2038             dm_table_any_dev_attr(t, device_is_not_random, NULL))
2039                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2040
2041         /*
2042          * For a zoned target, setup the zones related queue attributes
2043          * and resources necessary for zone append emulation if necessary.
2044          */
2045         if (blk_queue_is_zoned(q)) {
2046                 r = dm_set_zones_restrictions(t, q);
2047                 if (r)
2048                         return r;
2049         }
2050
2051         dm_update_crypto_profile(q, t);
2052         disk_update_readahead(t->md->disk);
2053
2054         /*
2055          * Check for request-based device is left to
2056          * dm_mq_init_request_queue()->blk_mq_init_allocated_queue().
2057          *
2058          * For bio-based device, only set QUEUE_FLAG_POLL when all
2059          * underlying devices supporting polling.
2060          */
2061         if (__table_type_bio_based(t->type)) {
2062                 if (dm_table_supports_poll(t))
2063                         blk_queue_flag_set(QUEUE_FLAG_POLL, q);
2064                 else
2065                         blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
2066         }
2067
2068         return 0;
2069 }
2070
2071 unsigned int dm_table_get_num_targets(struct dm_table *t)
2072 {
2073         return t->num_targets;
2074 }
2075
2076 struct list_head *dm_table_get_devices(struct dm_table *t)
2077 {
2078         return &t->devices;
2079 }
2080
2081 fmode_t dm_table_get_mode(struct dm_table *t)
2082 {
2083         return t->mode;
2084 }
2085 EXPORT_SYMBOL(dm_table_get_mode);
2086
2087 enum suspend_mode {
2088         PRESUSPEND,
2089         PRESUSPEND_UNDO,
2090         POSTSUSPEND,
2091 };
2092
2093 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
2094 {
2095         int i = t->num_targets;
2096         struct dm_target *ti = t->targets;
2097
2098         lockdep_assert_held(&t->md->suspend_lock);
2099
2100         while (i--) {
2101                 switch (mode) {
2102                 case PRESUSPEND:
2103                         if (ti->type->presuspend)
2104                                 ti->type->presuspend(ti);
2105                         break;
2106                 case PRESUSPEND_UNDO:
2107                         if (ti->type->presuspend_undo)
2108                                 ti->type->presuspend_undo(ti);
2109                         break;
2110                 case POSTSUSPEND:
2111                         if (ti->type->postsuspend)
2112                                 ti->type->postsuspend(ti);
2113                         break;
2114                 }
2115                 ti++;
2116         }
2117 }
2118
2119 void dm_table_presuspend_targets(struct dm_table *t)
2120 {
2121         if (!t)
2122                 return;
2123
2124         suspend_targets(t, PRESUSPEND);
2125 }
2126
2127 void dm_table_presuspend_undo_targets(struct dm_table *t)
2128 {
2129         if (!t)
2130                 return;
2131
2132         suspend_targets(t, PRESUSPEND_UNDO);
2133 }
2134
2135 void dm_table_postsuspend_targets(struct dm_table *t)
2136 {
2137         if (!t)
2138                 return;
2139
2140         suspend_targets(t, POSTSUSPEND);
2141 }
2142
2143 int dm_table_resume_targets(struct dm_table *t)
2144 {
2145         int i, r = 0;
2146
2147         lockdep_assert_held(&t->md->suspend_lock);
2148
2149         for (i = 0; i < t->num_targets; i++) {
2150                 struct dm_target *ti = t->targets + i;
2151
2152                 if (!ti->type->preresume)
2153                         continue;
2154
2155                 r = ti->type->preresume(ti);
2156                 if (r) {
2157                         DMERR("%s: %s: preresume failed, error = %d",
2158                               dm_device_name(t->md), ti->type->name, r);
2159                         return r;
2160                 }
2161         }
2162
2163         for (i = 0; i < t->num_targets; i++) {
2164                 struct dm_target *ti = t->targets + i;
2165
2166                 if (ti->type->resume)
2167                         ti->type->resume(ti);
2168         }
2169
2170         return 0;
2171 }
2172
2173 struct mapped_device *dm_table_get_md(struct dm_table *t)
2174 {
2175         return t->md;
2176 }
2177 EXPORT_SYMBOL(dm_table_get_md);
2178
2179 const char *dm_table_device_name(struct dm_table *t)
2180 {
2181         return dm_device_name(t->md);
2182 }
2183 EXPORT_SYMBOL_GPL(dm_table_device_name);
2184
2185 void dm_table_run_md_queue_async(struct dm_table *t)
2186 {
2187         if (!dm_table_request_based(t))
2188                 return;
2189
2190         if (t->md->queue)
2191                 blk_mq_run_hw_queues(t->md->queue, true);
2192 }
2193 EXPORT_SYMBOL(dm_table_run_md_queue_async);
2194