dm table: propagate BDI_CAP_STABLE_WRITES to fix sporadic checksum errors
[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         unsigned integrity_added:1;
51
52         /*
53          * Indicates the rw permissions for the new logical
54          * device.  This should be a combination of FMODE_READ
55          * and FMODE_WRITE.
56          */
57         fmode_t mode;
58
59         /* a list of devices used by this table */
60         struct list_head devices;
61
62         /* events get handed up using this callback */
63         void (*event_fn)(void *);
64         void *event_context;
65
66         struct dm_md_mempools *mempools;
67
68         struct list_head target_callbacks;
69 };
70
71 /*
72  * Similar to ceiling(log_size(n))
73  */
74 static unsigned int int_log(unsigned int n, unsigned int base)
75 {
76         int result = 0;
77
78         while (n > 1) {
79                 n = dm_div_up(n, base);
80                 result++;
81         }
82
83         return result;
84 }
85
86 /*
87  * Calculate the index of the child node of the n'th node k'th key.
88  */
89 static inline unsigned int get_child(unsigned int n, unsigned int k)
90 {
91         return (n * CHILDREN_PER_NODE) + k;
92 }
93
94 /*
95  * Return the n'th node of level l from table t.
96  */
97 static inline sector_t *get_node(struct dm_table *t,
98                                  unsigned int l, unsigned int n)
99 {
100         return t->index[l] + (n * KEYS_PER_NODE);
101 }
102
103 /*
104  * Return the highest key that you could lookup from the n'th
105  * node on level l of the btree.
106  */
107 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
108 {
109         for (; l < t->depth - 1; l++)
110                 n = get_child(n, CHILDREN_PER_NODE - 1);
111
112         if (n >= t->counts[l])
113                 return (sector_t) - 1;
114
115         return get_node(t, l, n)[KEYS_PER_NODE - 1];
116 }
117
118 /*
119  * Fills in a level of the btree based on the highs of the level
120  * below it.
121  */
122 static int setup_btree_index(unsigned int l, struct dm_table *t)
123 {
124         unsigned int n, k;
125         sector_t *node;
126
127         for (n = 0U; n < t->counts[l]; n++) {
128                 node = get_node(t, l, n);
129
130                 for (k = 0U; k < KEYS_PER_NODE; k++)
131                         node[k] = high(t, l + 1, get_child(n, k));
132         }
133
134         return 0;
135 }
136
137 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
138 {
139         unsigned long size;
140         void *addr;
141
142         /*
143          * Check that we're not going to overflow.
144          */
145         if (nmemb > (ULONG_MAX / elem_size))
146                 return NULL;
147
148         size = nmemb * elem_size;
149         addr = vzalloc(size);
150
151         return addr;
152 }
153 EXPORT_SYMBOL(dm_vcalloc);
154
155 /*
156  * highs, and targets are managed as dynamic arrays during a
157  * table load.
158  */
159 static int alloc_targets(struct dm_table *t, unsigned int num)
160 {
161         sector_t *n_highs;
162         struct dm_target *n_targets;
163
164         /*
165          * Allocate both the target array and offset array at once.
166          * Append an empty entry to catch sectors beyond the end of
167          * the device.
168          */
169         n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
170                                           sizeof(sector_t));
171         if (!n_highs)
172                 return -ENOMEM;
173
174         n_targets = (struct dm_target *) (n_highs + num);
175
176         memset(n_highs, -1, sizeof(*n_highs) * num);
177         vfree(t->highs);
178
179         t->num_allocated = num;
180         t->highs = n_highs;
181         t->targets = n_targets;
182
183         return 0;
184 }
185
186 int dm_table_create(struct dm_table **result, fmode_t mode,
187                     unsigned num_targets, struct mapped_device *md)
188 {
189         struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
190
191         if (!t)
192                 return -ENOMEM;
193
194         INIT_LIST_HEAD(&t->devices);
195         INIT_LIST_HEAD(&t->target_callbacks);
196
197         if (!num_targets)
198                 num_targets = KEYS_PER_NODE;
199
200         num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
201
202         if (!num_targets) {
203                 kfree(t);
204                 return -ENOMEM;
205         }
206
207         if (alloc_targets(t, num_targets)) {
208                 kfree(t);
209                 return -ENOMEM;
210         }
211
212         t->type = DM_TYPE_NONE;
213         t->mode = mode;
214         t->md = md;
215         *result = t;
216         return 0;
217 }
218
219 static void free_devices(struct list_head *devices, struct mapped_device *md)
220 {
221         struct list_head *tmp, *next;
222
223         list_for_each_safe(tmp, next, devices) {
224                 struct dm_dev_internal *dd =
225                     list_entry(tmp, struct dm_dev_internal, list);
226                 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
227                        dm_device_name(md), dd->dm_dev->name);
228                 dm_put_table_device(md, dd->dm_dev);
229                 kfree(dd);
230         }
231 }
232
233 void dm_table_destroy(struct dm_table *t)
234 {
235         unsigned int i;
236
237         if (!t)
238                 return;
239
240         /* free the indexes */
241         if (t->depth >= 2)
242                 vfree(t->index[t->depth - 2]);
243
244         /* free the targets */
245         for (i = 0; i < t->num_targets; i++) {
246                 struct dm_target *tgt = t->targets + i;
247
248                 if (tgt->type->dtr)
249                         tgt->type->dtr(tgt);
250
251                 dm_put_target_type(tgt->type);
252         }
253
254         vfree(t->highs);
255
256         /* free the device list */
257         free_devices(&t->devices, t->md);
258
259         dm_free_md_mempools(t->mempools);
260
261         kfree(t);
262 }
263
264 /*
265  * See if we've already got a device in the list.
266  */
267 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
268 {
269         struct dm_dev_internal *dd;
270
271         list_for_each_entry (dd, l, list)
272                 if (dd->dm_dev->bdev->bd_dev == dev)
273                         return dd;
274
275         return NULL;
276 }
277
278 /*
279  * If possible, this checks an area of a destination device is invalid.
280  */
281 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
282                                   sector_t start, sector_t len, void *data)
283 {
284         struct request_queue *q;
285         struct queue_limits *limits = data;
286         struct block_device *bdev = dev->bdev;
287         sector_t dev_size =
288                 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
289         unsigned short logical_block_size_sectors =
290                 limits->logical_block_size >> SECTOR_SHIFT;
291         char b[BDEVNAME_SIZE];
292
293         /*
294          * Some devices exist without request functions,
295          * such as loop devices not yet bound to backing files.
296          * Forbid the use of such devices.
297          */
298         q = bdev_get_queue(bdev);
299         if (!q || !q->make_request_fn) {
300                 DMWARN("%s: %s is not yet initialised: "
301                        "start=%llu, len=%llu, dev_size=%llu",
302                        dm_device_name(ti->table->md), bdevname(bdev, b),
303                        (unsigned long long)start,
304                        (unsigned long long)len,
305                        (unsigned long long)dev_size);
306                 return 1;
307         }
308
309         if (!dev_size)
310                 return 0;
311
312         if ((start >= dev_size) || (start + len > dev_size)) {
313                 DMWARN("%s: %s too small for target: "
314                        "start=%llu, len=%llu, dev_size=%llu",
315                        dm_device_name(ti->table->md), bdevname(bdev, b),
316                        (unsigned long long)start,
317                        (unsigned long long)len,
318                        (unsigned long long)dev_size);
319                 return 1;
320         }
321
322         /*
323          * If the target is mapped to zoned block device(s), check
324          * that the zones are not partially mapped.
325          */
326         if (bdev_zoned_model(bdev) != BLK_ZONED_NONE) {
327                 unsigned int zone_sectors = bdev_zone_sectors(bdev);
328
329                 if (start & (zone_sectors - 1)) {
330                         DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s",
331                                dm_device_name(ti->table->md),
332                                (unsigned long long)start,
333                                zone_sectors, bdevname(bdev, b));
334                         return 1;
335                 }
336
337                 /*
338                  * Note: The last zone of a zoned block device may be smaller
339                  * than other zones. So for a target mapping the end of a
340                  * zoned block device with such a zone, len would not be zone
341                  * aligned. We do not allow such last smaller zone to be part
342                  * of the mapping here to ensure that mappings with multiple
343                  * devices do not end up with a smaller zone in the middle of
344                  * the sector range.
345                  */
346                 if (len & (zone_sectors - 1)) {
347                         DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s",
348                                dm_device_name(ti->table->md),
349                                (unsigned long long)len,
350                                zone_sectors, bdevname(bdev, b));
351                         return 1;
352                 }
353         }
354
355         if (logical_block_size_sectors <= 1)
356                 return 0;
357
358         if (start & (logical_block_size_sectors - 1)) {
359                 DMWARN("%s: start=%llu not aligned to h/w "
360                        "logical block size %u of %s",
361                        dm_device_name(ti->table->md),
362                        (unsigned long long)start,
363                        limits->logical_block_size, bdevname(bdev, b));
364                 return 1;
365         }
366
367         if (len & (logical_block_size_sectors - 1)) {
368                 DMWARN("%s: len=%llu not aligned to h/w "
369                        "logical block size %u of %s",
370                        dm_device_name(ti->table->md),
371                        (unsigned long long)len,
372                        limits->logical_block_size, bdevname(bdev, b));
373                 return 1;
374         }
375
376         return 0;
377 }
378
379 /*
380  * This upgrades the mode on an already open dm_dev, being
381  * careful to leave things as they were if we fail to reopen the
382  * device and not to touch the existing bdev field in case
383  * it is accessed concurrently inside dm_table_any_congested().
384  */
385 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
386                         struct mapped_device *md)
387 {
388         int r;
389         struct dm_dev *old_dev, *new_dev;
390
391         old_dev = dd->dm_dev;
392
393         r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
394                                 dd->dm_dev->mode | new_mode, &new_dev);
395         if (r)
396                 return r;
397
398         dd->dm_dev = new_dev;
399         dm_put_table_device(md, old_dev);
400
401         return 0;
402 }
403
404 /*
405  * Convert the path to a device
406  */
407 dev_t dm_get_dev_t(const char *path)
408 {
409         dev_t dev;
410         struct block_device *bdev;
411
412         bdev = lookup_bdev(path);
413         if (IS_ERR(bdev))
414                 dev = name_to_dev_t(path);
415         else {
416                 dev = bdev->bd_dev;
417                 bdput(bdev);
418         }
419
420         return dev;
421 }
422 EXPORT_SYMBOL_GPL(dm_get_dev_t);
423
424 /*
425  * Add a device to the list, or just increment the usage count if
426  * it's already present.
427  */
428 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
429                   struct dm_dev **result)
430 {
431         int r;
432         dev_t dev;
433         struct dm_dev_internal *dd;
434         struct dm_table *t = ti->table;
435
436         BUG_ON(!t);
437
438         dev = dm_get_dev_t(path);
439         if (!dev)
440                 return -ENODEV;
441
442         dd = find_device(&t->devices, dev);
443         if (!dd) {
444                 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
445                 if (!dd)
446                         return -ENOMEM;
447
448                 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
449                         kfree(dd);
450                         return r;
451                 }
452
453                 refcount_set(&dd->count, 1);
454                 list_add(&dd->list, &t->devices);
455                 goto out;
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         }
462         refcount_inc(&dd->count);
463 out:
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 *size, char **old_argv)
551 {
552         char **argv;
553         unsigned new_size;
554         gfp_t gfp;
555
556         if (*size) {
557                 new_size = *size * 2;
558                 gfp = GFP_KERNEL;
559         } else {
560                 new_size = 8;
561                 gfp = GFP_NOIO;
562         }
563         argv = kmalloc_array(new_size, sizeof(*argv), gfp);
564         if (argv) {
565                 memcpy(argv, old_argv, *size * sizeof(*argv));
566                 *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                 table_type == DM_TYPE_NVME_BIO_BASED);
870 }
871
872 static bool __table_type_request_based(enum dm_queue_mode table_type)
873 {
874         return table_type == DM_TYPE_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         return bdev_dax_supported(dev->bdev, PAGE_SIZE);
887 }
888
889 static bool dm_table_supports_dax(struct dm_table *t)
890 {
891         struct dm_target *ti;
892         unsigned i;
893
894         /* Ensure that all targets support DAX. */
895         for (i = 0; i < dm_table_get_num_targets(t); i++) {
896                 ti = dm_table_get_target(t, i);
897
898                 if (!ti->type->direct_access)
899                         return false;
900
901                 if (!ti->type->iterate_devices ||
902                     !ti->type->iterate_devices(ti, device_supports_dax, NULL))
903                         return false;
904         }
905
906         return true;
907 }
908
909 static bool dm_table_does_not_support_partial_completion(struct dm_table *t);
910
911 struct verify_rq_based_data {
912         unsigned sq_count;
913         unsigned mq_count;
914 };
915
916 static int device_is_rq_based(struct dm_target *ti, struct dm_dev *dev,
917                               sector_t start, sector_t len, void *data)
918 {
919         struct request_queue *q = bdev_get_queue(dev->bdev);
920         struct verify_rq_based_data *v = data;
921
922         if (queue_is_mq(q))
923                 v->mq_count++;
924         else
925                 v->sq_count++;
926
927         return queue_is_mq(q);
928 }
929
930 static int dm_table_determine_type(struct dm_table *t)
931 {
932         unsigned i;
933         unsigned bio_based = 0, request_based = 0, hybrid = 0;
934         struct verify_rq_based_data v = {.sq_count = 0, .mq_count = 0};
935         struct dm_target *tgt;
936         struct list_head *devices = dm_table_get_devices(t);
937         enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
938
939         if (t->type != DM_TYPE_NONE) {
940                 /* target already set the table's type */
941                 if (t->type == DM_TYPE_BIO_BASED) {
942                         /* possibly upgrade to a variant of bio-based */
943                         goto verify_bio_based;
944                 }
945                 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
946                 BUG_ON(t->type == DM_TYPE_NVME_BIO_BASED);
947                 goto verify_rq_based;
948         }
949
950         for (i = 0; i < t->num_targets; i++) {
951                 tgt = t->targets + i;
952                 if (dm_target_hybrid(tgt))
953                         hybrid = 1;
954                 else if (dm_target_request_based(tgt))
955                         request_based = 1;
956                 else
957                         bio_based = 1;
958
959                 if (bio_based && request_based) {
960                         DMERR("Inconsistent table: different target types"
961                               " can't be mixed up");
962                         return -EINVAL;
963                 }
964         }
965
966         if (hybrid && !bio_based && !request_based) {
967                 /*
968                  * The targets can work either way.
969                  * Determine the type from the live device.
970                  * Default to bio-based if device is new.
971                  */
972                 if (__table_type_request_based(live_md_type))
973                         request_based = 1;
974                 else
975                         bio_based = 1;
976         }
977
978         if (bio_based) {
979 verify_bio_based:
980                 /* We must use this table as bio-based */
981                 t->type = DM_TYPE_BIO_BASED;
982                 if (dm_table_supports_dax(t) ||
983                     (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
984                         t->type = DM_TYPE_DAX_BIO_BASED;
985                 } else {
986                         /* Check if upgrading to NVMe bio-based is valid or required */
987                         tgt = dm_table_get_immutable_target(t);
988                         if (tgt && !tgt->max_io_len && dm_table_does_not_support_partial_completion(t)) {
989                                 t->type = DM_TYPE_NVME_BIO_BASED;
990                                 goto verify_rq_based; /* must be stacked directly on NVMe (blk-mq) */
991                         } else if (list_empty(devices) && live_md_type == DM_TYPE_NVME_BIO_BASED) {
992                                 t->type = DM_TYPE_NVME_BIO_BASED;
993                         }
994                 }
995                 return 0;
996         }
997
998         BUG_ON(!request_based); /* No targets in this table */
999
1000         t->type = DM_TYPE_REQUEST_BASED;
1001
1002 verify_rq_based:
1003         /*
1004          * Request-based dm supports only tables that have a single target now.
1005          * To support multiple targets, request splitting support is needed,
1006          * and that needs lots of changes in the block-layer.
1007          * (e.g. request completion process for partial completion.)
1008          */
1009         if (t->num_targets > 1) {
1010                 DMERR("%s DM doesn't support multiple targets",
1011                       t->type == DM_TYPE_NVME_BIO_BASED ? "nvme bio-based" : "request-based");
1012                 return -EINVAL;
1013         }
1014
1015         if (list_empty(devices)) {
1016                 int srcu_idx;
1017                 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
1018
1019                 /* inherit live table's type */
1020                 if (live_table)
1021                         t->type = live_table->type;
1022                 dm_put_live_table(t->md, srcu_idx);
1023                 return 0;
1024         }
1025
1026         tgt = dm_table_get_immutable_target(t);
1027         if (!tgt) {
1028                 DMERR("table load rejected: immutable target is required");
1029                 return -EINVAL;
1030         } else if (tgt->max_io_len) {
1031                 DMERR("table load rejected: immutable target that splits IO is not supported");
1032                 return -EINVAL;
1033         }
1034
1035         /* Non-request-stackable devices can't be used for request-based dm */
1036         if (!tgt->type->iterate_devices ||
1037             !tgt->type->iterate_devices(tgt, device_is_rq_based, &v)) {
1038                 DMERR("table load rejected: including non-request-stackable devices");
1039                 return -EINVAL;
1040         }
1041         if (v.sq_count > 0) {
1042                 DMERR("table load rejected: not all devices are blk-mq request-stackable");
1043                 return -EINVAL;
1044         }
1045
1046         return 0;
1047 }
1048
1049 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
1050 {
1051         return t->type;
1052 }
1053
1054 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
1055 {
1056         return t->immutable_target_type;
1057 }
1058
1059 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
1060 {
1061         /* Immutable target is implicitly a singleton */
1062         if (t->num_targets > 1 ||
1063             !dm_target_is_immutable(t->targets[0].type))
1064                 return NULL;
1065
1066         return t->targets;
1067 }
1068
1069 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1070 {
1071         struct dm_target *ti;
1072         unsigned i;
1073
1074         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1075                 ti = dm_table_get_target(t, i);
1076                 if (dm_target_is_wildcard(ti->type))
1077                         return ti;
1078         }
1079
1080         return NULL;
1081 }
1082
1083 bool dm_table_bio_based(struct dm_table *t)
1084 {
1085         return __table_type_bio_based(dm_table_get_type(t));
1086 }
1087
1088 bool dm_table_request_based(struct dm_table *t)
1089 {
1090         return __table_type_request_based(dm_table_get_type(t));
1091 }
1092
1093 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1094 {
1095         enum dm_queue_mode type = dm_table_get_type(t);
1096         unsigned per_io_data_size = 0;
1097         unsigned min_pool_size = 0;
1098         struct dm_target *ti;
1099         unsigned i;
1100
1101         if (unlikely(type == DM_TYPE_NONE)) {
1102                 DMWARN("no table type is set, can't allocate mempools");
1103                 return -EINVAL;
1104         }
1105
1106         if (__table_type_bio_based(type))
1107                 for (i = 0; i < t->num_targets; i++) {
1108                         ti = t->targets + i;
1109                         per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1110                         min_pool_size = max(min_pool_size, ti->num_flush_bios);
1111                 }
1112
1113         t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1114                                            per_io_data_size, min_pool_size);
1115         if (!t->mempools)
1116                 return -ENOMEM;
1117
1118         return 0;
1119 }
1120
1121 void dm_table_free_md_mempools(struct dm_table *t)
1122 {
1123         dm_free_md_mempools(t->mempools);
1124         t->mempools = NULL;
1125 }
1126
1127 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1128 {
1129         return t->mempools;
1130 }
1131
1132 static int setup_indexes(struct dm_table *t)
1133 {
1134         int i;
1135         unsigned int total = 0;
1136         sector_t *indexes;
1137
1138         /* allocate the space for *all* the indexes */
1139         for (i = t->depth - 2; i >= 0; i--) {
1140                 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1141                 total += t->counts[i];
1142         }
1143
1144         indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1145         if (!indexes)
1146                 return -ENOMEM;
1147
1148         /* set up internal nodes, bottom-up */
1149         for (i = t->depth - 2; i >= 0; i--) {
1150                 t->index[i] = indexes;
1151                 indexes += (KEYS_PER_NODE * t->counts[i]);
1152                 setup_btree_index(i, t);
1153         }
1154
1155         return 0;
1156 }
1157
1158 /*
1159  * Builds the btree to index the map.
1160  */
1161 static int dm_table_build_index(struct dm_table *t)
1162 {
1163         int r = 0;
1164         unsigned int leaf_nodes;
1165
1166         /* how many indexes will the btree have ? */
1167         leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1168         t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1169
1170         /* leaf layer has already been set up */
1171         t->counts[t->depth - 1] = leaf_nodes;
1172         t->index[t->depth - 1] = t->highs;
1173
1174         if (t->depth >= 2)
1175                 r = setup_indexes(t);
1176
1177         return r;
1178 }
1179
1180 static bool integrity_profile_exists(struct gendisk *disk)
1181 {
1182         return !!blk_get_integrity(disk);
1183 }
1184
1185 /*
1186  * Get a disk whose integrity profile reflects the table's profile.
1187  * Returns NULL if integrity support was inconsistent or unavailable.
1188  */
1189 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1190 {
1191         struct list_head *devices = dm_table_get_devices(t);
1192         struct dm_dev_internal *dd = NULL;
1193         struct gendisk *prev_disk = NULL, *template_disk = NULL;
1194         unsigned i;
1195
1196         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1197                 struct dm_target *ti = dm_table_get_target(t, i);
1198                 if (!dm_target_passes_integrity(ti->type))
1199                         goto no_integrity;
1200         }
1201
1202         list_for_each_entry(dd, devices, list) {
1203                 template_disk = dd->dm_dev->bdev->bd_disk;
1204                 if (!integrity_profile_exists(template_disk))
1205                         goto no_integrity;
1206                 else if (prev_disk &&
1207                          blk_integrity_compare(prev_disk, template_disk) < 0)
1208                         goto no_integrity;
1209                 prev_disk = template_disk;
1210         }
1211
1212         return template_disk;
1213
1214 no_integrity:
1215         if (prev_disk)
1216                 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1217                        dm_device_name(t->md),
1218                        prev_disk->disk_name,
1219                        template_disk->disk_name);
1220         return NULL;
1221 }
1222
1223 /*
1224  * Register the mapped device for blk_integrity support if the
1225  * underlying devices have an integrity profile.  But all devices may
1226  * not have matching profiles (checking all devices isn't reliable
1227  * during table load because this table may use other DM device(s) which
1228  * must be resumed before they will have an initialized integity
1229  * profile).  Consequently, stacked DM devices force a 2 stage integrity
1230  * profile validation: First pass during table load, final pass during
1231  * resume.
1232  */
1233 static int dm_table_register_integrity(struct dm_table *t)
1234 {
1235         struct mapped_device *md = t->md;
1236         struct gendisk *template_disk = NULL;
1237
1238         /* If target handles integrity itself do not register it here. */
1239         if (t->integrity_added)
1240                 return 0;
1241
1242         template_disk = dm_table_get_integrity_disk(t);
1243         if (!template_disk)
1244                 return 0;
1245
1246         if (!integrity_profile_exists(dm_disk(md))) {
1247                 t->integrity_supported = true;
1248                 /*
1249                  * Register integrity profile during table load; we can do
1250                  * this because the final profile must match during resume.
1251                  */
1252                 blk_integrity_register(dm_disk(md),
1253                                        blk_get_integrity(template_disk));
1254                 return 0;
1255         }
1256
1257         /*
1258          * If DM device already has an initialized integrity
1259          * profile the new profile should not conflict.
1260          */
1261         if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1262                 DMWARN("%s: conflict with existing integrity profile: "
1263                        "%s profile mismatch",
1264                        dm_device_name(t->md),
1265                        template_disk->disk_name);
1266                 return 1;
1267         }
1268
1269         /* Preserve existing integrity profile */
1270         t->integrity_supported = true;
1271         return 0;
1272 }
1273
1274 /*
1275  * Prepares the table for use by building the indices,
1276  * setting the type, and allocating mempools.
1277  */
1278 int dm_table_complete(struct dm_table *t)
1279 {
1280         int r;
1281
1282         r = dm_table_determine_type(t);
1283         if (r) {
1284                 DMERR("unable to determine table type");
1285                 return r;
1286         }
1287
1288         r = dm_table_build_index(t);
1289         if (r) {
1290                 DMERR("unable to build btrees");
1291                 return r;
1292         }
1293
1294         r = dm_table_register_integrity(t);
1295         if (r) {
1296                 DMERR("could not register integrity profile.");
1297                 return r;
1298         }
1299
1300         r = dm_table_alloc_md_mempools(t, t->md);
1301         if (r)
1302                 DMERR("unable to allocate mempools");
1303
1304         return r;
1305 }
1306
1307 static DEFINE_MUTEX(_event_lock);
1308 void dm_table_event_callback(struct dm_table *t,
1309                              void (*fn)(void *), void *context)
1310 {
1311         mutex_lock(&_event_lock);
1312         t->event_fn = fn;
1313         t->event_context = context;
1314         mutex_unlock(&_event_lock);
1315 }
1316
1317 void dm_table_event(struct dm_table *t)
1318 {
1319         /*
1320          * You can no longer call dm_table_event() from interrupt
1321          * context, use a bottom half instead.
1322          */
1323         BUG_ON(in_interrupt());
1324
1325         mutex_lock(&_event_lock);
1326         if (t->event_fn)
1327                 t->event_fn(t->event_context);
1328         mutex_unlock(&_event_lock);
1329 }
1330 EXPORT_SYMBOL(dm_table_event);
1331
1332 sector_t dm_table_get_size(struct dm_table *t)
1333 {
1334         return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1335 }
1336 EXPORT_SYMBOL(dm_table_get_size);
1337
1338 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1339 {
1340         if (index >= t->num_targets)
1341                 return NULL;
1342
1343         return t->targets + index;
1344 }
1345
1346 /*
1347  * Search the btree for the correct target.
1348  *
1349  * Caller should check returned pointer with dm_target_is_valid()
1350  * to trap I/O beyond end of device.
1351  */
1352 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1353 {
1354         unsigned int l, n = 0, k = 0;
1355         sector_t *node;
1356
1357         for (l = 0; l < t->depth; l++) {
1358                 n = get_child(n, k);
1359                 node = get_node(t, l, n);
1360
1361                 for (k = 0; k < KEYS_PER_NODE; k++)
1362                         if (node[k] >= sector)
1363                                 break;
1364         }
1365
1366         return &t->targets[(KEYS_PER_NODE * n) + k];
1367 }
1368
1369 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1370                         sector_t start, sector_t len, void *data)
1371 {
1372         unsigned *num_devices = data;
1373
1374         (*num_devices)++;
1375
1376         return 0;
1377 }
1378
1379 /*
1380  * Check whether a table has no data devices attached using each
1381  * target's iterate_devices method.
1382  * Returns false if the result is unknown because a target doesn't
1383  * support iterate_devices.
1384  */
1385 bool dm_table_has_no_data_devices(struct dm_table *table)
1386 {
1387         struct dm_target *ti;
1388         unsigned i, num_devices;
1389
1390         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1391                 ti = dm_table_get_target(table, i);
1392
1393                 if (!ti->type->iterate_devices)
1394                         return false;
1395
1396                 num_devices = 0;
1397                 ti->type->iterate_devices(ti, count_device, &num_devices);
1398                 if (num_devices)
1399                         return false;
1400         }
1401
1402         return true;
1403 }
1404
1405 static int device_is_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1406                                  sector_t start, sector_t len, void *data)
1407 {
1408         struct request_queue *q = bdev_get_queue(dev->bdev);
1409         enum blk_zoned_model *zoned_model = data;
1410
1411         return q && blk_queue_zoned_model(q) == *zoned_model;
1412 }
1413
1414 static bool dm_table_supports_zoned_model(struct dm_table *t,
1415                                           enum blk_zoned_model zoned_model)
1416 {
1417         struct dm_target *ti;
1418         unsigned i;
1419
1420         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1421                 ti = dm_table_get_target(t, i);
1422
1423                 if (zoned_model == BLK_ZONED_HM &&
1424                     !dm_target_supports_zoned_hm(ti->type))
1425                         return false;
1426
1427                 if (!ti->type->iterate_devices ||
1428                     !ti->type->iterate_devices(ti, device_is_zoned_model, &zoned_model))
1429                         return false;
1430         }
1431
1432         return true;
1433 }
1434
1435 static int device_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1436                                        sector_t start, sector_t len, void *data)
1437 {
1438         struct request_queue *q = bdev_get_queue(dev->bdev);
1439         unsigned int *zone_sectors = data;
1440
1441         return q && blk_queue_zone_sectors(q) == *zone_sectors;
1442 }
1443
1444 static bool dm_table_matches_zone_sectors(struct dm_table *t,
1445                                           unsigned int zone_sectors)
1446 {
1447         struct dm_target *ti;
1448         unsigned i;
1449
1450         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1451                 ti = dm_table_get_target(t, i);
1452
1453                 if (!ti->type->iterate_devices ||
1454                     !ti->type->iterate_devices(ti, device_matches_zone_sectors, &zone_sectors))
1455                         return false;
1456         }
1457
1458         return true;
1459 }
1460
1461 static int validate_hardware_zoned_model(struct dm_table *table,
1462                                          enum blk_zoned_model zoned_model,
1463                                          unsigned int zone_sectors)
1464 {
1465         if (zoned_model == BLK_ZONED_NONE)
1466                 return 0;
1467
1468         if (!dm_table_supports_zoned_model(table, zoned_model)) {
1469                 DMERR("%s: zoned model is not consistent across all devices",
1470                       dm_device_name(table->md));
1471                 return -EINVAL;
1472         }
1473
1474         /* Check zone size validity and compatibility */
1475         if (!zone_sectors || !is_power_of_2(zone_sectors))
1476                 return -EINVAL;
1477
1478         if (!dm_table_matches_zone_sectors(table, zone_sectors)) {
1479                 DMERR("%s: zone sectors is not consistent across all devices",
1480                       dm_device_name(table->md));
1481                 return -EINVAL;
1482         }
1483
1484         return 0;
1485 }
1486
1487 /*
1488  * Establish the new table's queue_limits and validate them.
1489  */
1490 int dm_calculate_queue_limits(struct dm_table *table,
1491                               struct queue_limits *limits)
1492 {
1493         struct dm_target *ti;
1494         struct queue_limits ti_limits;
1495         unsigned i;
1496         enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1497         unsigned int zone_sectors = 0;
1498
1499         blk_set_stacking_limits(limits);
1500
1501         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1502                 blk_set_stacking_limits(&ti_limits);
1503
1504                 ti = dm_table_get_target(table, i);
1505
1506                 if (!ti->type->iterate_devices)
1507                         goto combine_limits;
1508
1509                 /*
1510                  * Combine queue limits of all the devices this target uses.
1511                  */
1512                 ti->type->iterate_devices(ti, dm_set_device_limits,
1513                                           &ti_limits);
1514
1515                 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1516                         /*
1517                          * After stacking all limits, validate all devices
1518                          * in table support this zoned model and zone sectors.
1519                          */
1520                         zoned_model = ti_limits.zoned;
1521                         zone_sectors = ti_limits.chunk_sectors;
1522                 }
1523
1524                 /* Set I/O hints portion of queue limits */
1525                 if (ti->type->io_hints)
1526                         ti->type->io_hints(ti, &ti_limits);
1527
1528                 /*
1529                  * Check each device area is consistent with the target's
1530                  * overall queue limits.
1531                  */
1532                 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1533                                               &ti_limits))
1534                         return -EINVAL;
1535
1536 combine_limits:
1537                 /*
1538                  * Merge this target's queue limits into the overall limits
1539                  * for the table.
1540                  */
1541                 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1542                         DMWARN("%s: adding target device "
1543                                "(start sect %llu len %llu) "
1544                                "caused an alignment inconsistency",
1545                                dm_device_name(table->md),
1546                                (unsigned long long) ti->begin,
1547                                (unsigned long long) ti->len);
1548
1549                 /*
1550                  * FIXME: this should likely be moved to blk_stack_limits(), would
1551                  * also eliminate limits->zoned stacking hack in dm_set_device_limits()
1552                  */
1553                 if (limits->zoned == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1554                         /*
1555                          * By default, the stacked limits zoned model is set to
1556                          * BLK_ZONED_NONE in blk_set_stacking_limits(). Update
1557                          * this model using the first target model reported
1558                          * that is not BLK_ZONED_NONE. This will be either the
1559                          * first target device zoned model or the model reported
1560                          * by the target .io_hints.
1561                          */
1562                         limits->zoned = ti_limits.zoned;
1563                 }
1564         }
1565
1566         /*
1567          * Verify that the zoned model and zone sectors, as determined before
1568          * any .io_hints override, are the same across all devices in the table.
1569          * - this is especially relevant if .io_hints is emulating a disk-managed
1570          *   zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1571          * BUT...
1572          */
1573         if (limits->zoned != BLK_ZONED_NONE) {
1574                 /*
1575                  * ...IF the above limits stacking determined a zoned model
1576                  * validate that all of the table's devices conform to it.
1577                  */
1578                 zoned_model = limits->zoned;
1579                 zone_sectors = limits->chunk_sectors;
1580         }
1581         if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1582                 return -EINVAL;
1583
1584         return validate_hardware_logical_block_alignment(table, limits);
1585 }
1586
1587 /*
1588  * Verify that all devices have an integrity profile that matches the
1589  * DM device's registered integrity profile.  If the profiles don't
1590  * match then unregister the DM device's integrity profile.
1591  */
1592 static void dm_table_verify_integrity(struct dm_table *t)
1593 {
1594         struct gendisk *template_disk = NULL;
1595
1596         if (t->integrity_added)
1597                 return;
1598
1599         if (t->integrity_supported) {
1600                 /*
1601                  * Verify that the original integrity profile
1602                  * matches all the devices in this table.
1603                  */
1604                 template_disk = dm_table_get_integrity_disk(t);
1605                 if (template_disk &&
1606                     blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1607                         return;
1608         }
1609
1610         if (integrity_profile_exists(dm_disk(t->md))) {
1611                 DMWARN("%s: unable to establish an integrity profile",
1612                        dm_device_name(t->md));
1613                 blk_integrity_unregister(dm_disk(t->md));
1614         }
1615 }
1616
1617 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1618                                 sector_t start, sector_t len, void *data)
1619 {
1620         unsigned long flush = (unsigned long) data;
1621         struct request_queue *q = bdev_get_queue(dev->bdev);
1622
1623         return q && (q->queue_flags & flush);
1624 }
1625
1626 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1627 {
1628         struct dm_target *ti;
1629         unsigned i;
1630
1631         /*
1632          * Require at least one underlying device to support flushes.
1633          * t->devices includes internal dm devices such as mirror logs
1634          * so we need to use iterate_devices here, which targets
1635          * supporting flushes must provide.
1636          */
1637         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1638                 ti = dm_table_get_target(t, i);
1639
1640                 if (!ti->num_flush_bios)
1641                         continue;
1642
1643                 if (ti->flush_supported)
1644                         return true;
1645
1646                 if (ti->type->iterate_devices &&
1647                     ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1648                         return true;
1649         }
1650
1651         return false;
1652 }
1653
1654 static int device_dax_write_cache_enabled(struct dm_target *ti,
1655                                           struct dm_dev *dev, sector_t start,
1656                                           sector_t len, void *data)
1657 {
1658         struct dax_device *dax_dev = dev->dax_dev;
1659
1660         if (!dax_dev)
1661                 return false;
1662
1663         if (dax_write_cache_enabled(dax_dev))
1664                 return true;
1665         return false;
1666 }
1667
1668 static int dm_table_supports_dax_write_cache(struct dm_table *t)
1669 {
1670         struct dm_target *ti;
1671         unsigned i;
1672
1673         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1674                 ti = dm_table_get_target(t, i);
1675
1676                 if (ti->type->iterate_devices &&
1677                     ti->type->iterate_devices(ti,
1678                                 device_dax_write_cache_enabled, NULL))
1679                         return true;
1680         }
1681
1682         return false;
1683 }
1684
1685 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1686                             sector_t start, sector_t len, void *data)
1687 {
1688         struct request_queue *q = bdev_get_queue(dev->bdev);
1689
1690         return q && blk_queue_nonrot(q);
1691 }
1692
1693 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1694                              sector_t start, sector_t len, void *data)
1695 {
1696         struct request_queue *q = bdev_get_queue(dev->bdev);
1697
1698         return q && !blk_queue_add_random(q);
1699 }
1700
1701 static bool dm_table_all_devices_attribute(struct dm_table *t,
1702                                            iterate_devices_callout_fn func)
1703 {
1704         struct dm_target *ti;
1705         unsigned i;
1706
1707         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1708                 ti = dm_table_get_target(t, i);
1709
1710                 if (!ti->type->iterate_devices ||
1711                     !ti->type->iterate_devices(ti, func, NULL))
1712                         return false;
1713         }
1714
1715         return true;
1716 }
1717
1718 static int device_no_partial_completion(struct dm_target *ti, struct dm_dev *dev,
1719                                         sector_t start, sector_t len, void *data)
1720 {
1721         char b[BDEVNAME_SIZE];
1722
1723         /* For now, NVMe devices are the only devices of this class */
1724         return (strncmp(bdevname(dev->bdev, b), "nvme", 4) == 0);
1725 }
1726
1727 static bool dm_table_does_not_support_partial_completion(struct dm_table *t)
1728 {
1729         return dm_table_all_devices_attribute(t, device_no_partial_completion);
1730 }
1731
1732 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1733                                          sector_t start, sector_t len, void *data)
1734 {
1735         struct request_queue *q = bdev_get_queue(dev->bdev);
1736
1737         return q && !q->limits.max_write_same_sectors;
1738 }
1739
1740 static bool dm_table_supports_write_same(struct dm_table *t)
1741 {
1742         struct dm_target *ti;
1743         unsigned i;
1744
1745         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1746                 ti = dm_table_get_target(t, i);
1747
1748                 if (!ti->num_write_same_bios)
1749                         return false;
1750
1751                 if (!ti->type->iterate_devices ||
1752                     ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1753                         return false;
1754         }
1755
1756         return true;
1757 }
1758
1759 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1760                                            sector_t start, sector_t len, void *data)
1761 {
1762         struct request_queue *q = bdev_get_queue(dev->bdev);
1763
1764         return q && !q->limits.max_write_zeroes_sectors;
1765 }
1766
1767 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1768 {
1769         struct dm_target *ti;
1770         unsigned i = 0;
1771
1772         while (i < dm_table_get_num_targets(t)) {
1773                 ti = dm_table_get_target(t, i++);
1774
1775                 if (!ti->num_write_zeroes_bios)
1776                         return false;
1777
1778                 if (!ti->type->iterate_devices ||
1779                     ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1780                         return false;
1781         }
1782
1783         return true;
1784 }
1785
1786 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1787                                       sector_t start, sector_t len, void *data)
1788 {
1789         struct request_queue *q = bdev_get_queue(dev->bdev);
1790
1791         return q && !blk_queue_discard(q);
1792 }
1793
1794 static bool dm_table_supports_discards(struct dm_table *t)
1795 {
1796         struct dm_target *ti;
1797         unsigned i;
1798
1799         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1800                 ti = dm_table_get_target(t, i);
1801
1802                 if (!ti->num_discard_bios)
1803                         return false;
1804
1805                 /*
1806                  * Either the target provides discard support (as implied by setting
1807                  * 'discards_supported') or it relies on _all_ data devices having
1808                  * discard support.
1809                  */
1810                 if (!ti->discards_supported &&
1811                     (!ti->type->iterate_devices ||
1812                      ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1813                         return false;
1814         }
1815
1816         return true;
1817 }
1818
1819 static int device_not_secure_erase_capable(struct dm_target *ti,
1820                                            struct dm_dev *dev, sector_t start,
1821                                            sector_t len, void *data)
1822 {
1823         struct request_queue *q = bdev_get_queue(dev->bdev);
1824
1825         return q && !blk_queue_secure_erase(q);
1826 }
1827
1828 static bool dm_table_supports_secure_erase(struct dm_table *t)
1829 {
1830         struct dm_target *ti;
1831         unsigned int i;
1832
1833         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1834                 ti = dm_table_get_target(t, i);
1835
1836                 if (!ti->num_secure_erase_bios)
1837                         return false;
1838
1839                 if (!ti->type->iterate_devices ||
1840                     ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1841                         return false;
1842         }
1843
1844         return true;
1845 }
1846
1847 static int device_requires_stable_pages(struct dm_target *ti,
1848                                         struct dm_dev *dev, sector_t start,
1849                                         sector_t len, void *data)
1850 {
1851         struct request_queue *q = bdev_get_queue(dev->bdev);
1852
1853         return q && bdi_cap_stable_pages_required(q->backing_dev_info);
1854 }
1855
1856 /*
1857  * If any underlying device requires stable pages, a table must require
1858  * them as well.  Only targets that support iterate_devices are considered:
1859  * don't want error, zero, etc to require stable pages.
1860  */
1861 static bool dm_table_requires_stable_pages(struct dm_table *t)
1862 {
1863         struct dm_target *ti;
1864         unsigned i;
1865
1866         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1867                 ti = dm_table_get_target(t, i);
1868
1869                 if (ti->type->iterate_devices &&
1870                     ti->type->iterate_devices(ti, device_requires_stable_pages, NULL))
1871                         return true;
1872         }
1873
1874         return false;
1875 }
1876
1877 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1878                                struct queue_limits *limits)
1879 {
1880         bool wc = false, fua = false;
1881
1882         /*
1883          * Copy table's limits to the DM device's request_queue
1884          */
1885         q->limits = *limits;
1886
1887         if (!dm_table_supports_discards(t)) {
1888                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
1889                 /* Must also clear discard limits... */
1890                 q->limits.max_discard_sectors = 0;
1891                 q->limits.max_hw_discard_sectors = 0;
1892                 q->limits.discard_granularity = 0;
1893                 q->limits.discard_alignment = 0;
1894                 q->limits.discard_misaligned = 0;
1895         } else
1896                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
1897
1898         if (dm_table_supports_secure_erase(t))
1899                 blk_queue_flag_set(QUEUE_FLAG_SECERASE, q);
1900
1901         if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1902                 wc = true;
1903                 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1904                         fua = true;
1905         }
1906         blk_queue_write_cache(q, wc, fua);
1907
1908         if (dm_table_supports_dax(t))
1909                 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
1910         else
1911                 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
1912
1913         if (dm_table_supports_dax_write_cache(t))
1914                 dax_write_cache(t->md->dax_dev, true);
1915
1916         /* Ensure that all underlying devices are non-rotational. */
1917         if (dm_table_all_devices_attribute(t, device_is_nonrot))
1918                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
1919         else
1920                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
1921
1922         if (!dm_table_supports_write_same(t))
1923                 q->limits.max_write_same_sectors = 0;
1924         if (!dm_table_supports_write_zeroes(t))
1925                 q->limits.max_write_zeroes_sectors = 0;
1926
1927         dm_table_verify_integrity(t);
1928
1929         /*
1930          * Some devices don't use blk_integrity but still want stable pages
1931          * because they do their own checksumming.
1932          */
1933         if (dm_table_requires_stable_pages(t))
1934                 q->backing_dev_info->capabilities |= BDI_CAP_STABLE_WRITES;
1935         else
1936                 q->backing_dev_info->capabilities &= ~BDI_CAP_STABLE_WRITES;
1937
1938         /*
1939          * Determine whether or not this queue's I/O timings contribute
1940          * to the entropy pool, Only request-based targets use this.
1941          * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1942          * have it set.
1943          */
1944         if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1945                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
1946
1947         /*
1948          * For a zoned target, the number of zones should be updated for the
1949          * correct value to be exposed in sysfs queue/nr_zones. For a BIO based
1950          * target, this is all that is needed. For a request based target, the
1951          * queue zone bitmaps must also be updated.
1952          * Use blk_revalidate_disk_zones() to handle this.
1953          */
1954         if (blk_queue_is_zoned(q))
1955                 blk_revalidate_disk_zones(t->md->disk);
1956
1957         /* Allow reads to exceed readahead limits */
1958         q->backing_dev_info->io_pages = limits->max_sectors >> (PAGE_SHIFT - 9);
1959 }
1960
1961 unsigned int dm_table_get_num_targets(struct dm_table *t)
1962 {
1963         return t->num_targets;
1964 }
1965
1966 struct list_head *dm_table_get_devices(struct dm_table *t)
1967 {
1968         return &t->devices;
1969 }
1970
1971 fmode_t dm_table_get_mode(struct dm_table *t)
1972 {
1973         return t->mode;
1974 }
1975 EXPORT_SYMBOL(dm_table_get_mode);
1976
1977 enum suspend_mode {
1978         PRESUSPEND,
1979         PRESUSPEND_UNDO,
1980         POSTSUSPEND,
1981 };
1982
1983 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1984 {
1985         int i = t->num_targets;
1986         struct dm_target *ti = t->targets;
1987
1988         lockdep_assert_held(&t->md->suspend_lock);
1989
1990         while (i--) {
1991                 switch (mode) {
1992                 case PRESUSPEND:
1993                         if (ti->type->presuspend)
1994                                 ti->type->presuspend(ti);
1995                         break;
1996                 case PRESUSPEND_UNDO:
1997                         if (ti->type->presuspend_undo)
1998                                 ti->type->presuspend_undo(ti);
1999                         break;
2000                 case POSTSUSPEND:
2001                         if (ti->type->postsuspend)
2002                                 ti->type->postsuspend(ti);
2003                         break;
2004                 }
2005                 ti++;
2006         }
2007 }
2008
2009 void dm_table_presuspend_targets(struct dm_table *t)
2010 {
2011         if (!t)
2012                 return;
2013
2014         suspend_targets(t, PRESUSPEND);
2015 }
2016
2017 void dm_table_presuspend_undo_targets(struct dm_table *t)
2018 {
2019         if (!t)
2020                 return;
2021
2022         suspend_targets(t, PRESUSPEND_UNDO);
2023 }
2024
2025 void dm_table_postsuspend_targets(struct dm_table *t)
2026 {
2027         if (!t)
2028                 return;
2029
2030         suspend_targets(t, POSTSUSPEND);
2031 }
2032
2033 int dm_table_resume_targets(struct dm_table *t)
2034 {
2035         int i, r = 0;
2036
2037         lockdep_assert_held(&t->md->suspend_lock);
2038
2039         for (i = 0; i < t->num_targets; i++) {
2040                 struct dm_target *ti = t->targets + i;
2041
2042                 if (!ti->type->preresume)
2043                         continue;
2044
2045                 r = ti->type->preresume(ti);
2046                 if (r) {
2047                         DMERR("%s: %s: preresume failed, error = %d",
2048                               dm_device_name(t->md), ti->type->name, r);
2049                         return r;
2050                 }
2051         }
2052
2053         for (i = 0; i < t->num_targets; i++) {
2054                 struct dm_target *ti = t->targets + i;
2055
2056                 if (ti->type->resume)
2057                         ti->type->resume(ti);
2058         }
2059
2060         return 0;
2061 }
2062
2063 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
2064 {
2065         list_add(&cb->list, &t->target_callbacks);
2066 }
2067 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
2068
2069 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
2070 {
2071         struct dm_dev_internal *dd;
2072         struct list_head *devices = dm_table_get_devices(t);
2073         struct dm_target_callbacks *cb;
2074         int r = 0;
2075
2076         list_for_each_entry(dd, devices, list) {
2077                 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
2078                 char b[BDEVNAME_SIZE];
2079
2080                 if (likely(q))
2081                         r |= bdi_congested(q->backing_dev_info, bdi_bits);
2082                 else
2083                         DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
2084                                      dm_device_name(t->md),
2085                                      bdevname(dd->dm_dev->bdev, b));
2086         }
2087
2088         list_for_each_entry(cb, &t->target_callbacks, list)
2089                 if (cb->congested_fn)
2090                         r |= cb->congested_fn(cb, bdi_bits);
2091
2092         return r;
2093 }
2094
2095 struct mapped_device *dm_table_get_md(struct dm_table *t)
2096 {
2097         return t->md;
2098 }
2099 EXPORT_SYMBOL(dm_table_get_md);
2100
2101 const char *dm_table_device_name(struct dm_table *t)
2102 {
2103         return dm_device_name(t->md);
2104 }
2105 EXPORT_SYMBOL_GPL(dm_table_device_name);
2106
2107 void dm_table_run_md_queue_async(struct dm_table *t)
2108 {
2109         struct mapped_device *md;
2110         struct request_queue *queue;
2111
2112         if (!dm_table_request_based(t))
2113                 return;
2114
2115         md = dm_table_get_md(t);
2116         queue = dm_get_md_queue(md);
2117         if (queue)
2118                 blk_mq_run_hw_queues(queue, true);
2119 }
2120 EXPORT_SYMBOL(dm_table_run_md_queue_async);
2121