Loading...
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.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
22#define DM_MSG_PREFIX "table"
23
24#define MAX_DEPTH 16
25#define NODE_SIZE L1_CACHE_BYTES
26#define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
27#define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
28
29/*
30 * The table has always exactly one reference from either mapped_device->map
31 * or hash_cell->new_map. This reference is not counted in table->holders.
32 * A pair of dm_create_table/dm_destroy_table functions is used for table
33 * creation/destruction.
34 *
35 * Temporary references from the other code increase table->holders. A pair
36 * of dm_table_get/dm_table_put functions is used to manipulate it.
37 *
38 * When the table is about to be destroyed, we wait for table->holders to
39 * drop to zero.
40 */
41
42struct dm_table {
43 struct mapped_device *md;
44 atomic_t holders;
45 unsigned type;
46
47 /* btree table */
48 unsigned int depth;
49 unsigned int counts[MAX_DEPTH]; /* in nodes */
50 sector_t *index[MAX_DEPTH];
51
52 unsigned int num_targets;
53 unsigned int num_allocated;
54 sector_t *highs;
55 struct dm_target *targets;
56
57 unsigned integrity_supported:1;
58
59 /*
60 * Indicates the rw permissions for the new logical
61 * device. This should be a combination of FMODE_READ
62 * and FMODE_WRITE.
63 */
64 fmode_t mode;
65
66 /* a list of devices used by this table */
67 struct list_head devices;
68
69 /* events get handed up using this callback */
70 void (*event_fn)(void *);
71 void *event_context;
72
73 struct dm_md_mempools *mempools;
74
75 struct list_head target_callbacks;
76};
77
78/*
79 * Similar to ceiling(log_size(n))
80 */
81static unsigned int int_log(unsigned int n, unsigned int base)
82{
83 int result = 0;
84
85 while (n > 1) {
86 n = dm_div_up(n, base);
87 result++;
88 }
89
90 return result;
91}
92
93/*
94 * Calculate the index of the child node of the n'th node k'th key.
95 */
96static inline unsigned int get_child(unsigned int n, unsigned int k)
97{
98 return (n * CHILDREN_PER_NODE) + k;
99}
100
101/*
102 * Return the n'th node of level l from table t.
103 */
104static inline sector_t *get_node(struct dm_table *t,
105 unsigned int l, unsigned int n)
106{
107 return t->index[l] + (n * KEYS_PER_NODE);
108}
109
110/*
111 * Return the highest key that you could lookup from the n'th
112 * node on level l of the btree.
113 */
114static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
115{
116 for (; l < t->depth - 1; l++)
117 n = get_child(n, CHILDREN_PER_NODE - 1);
118
119 if (n >= t->counts[l])
120 return (sector_t) - 1;
121
122 return get_node(t, l, n)[KEYS_PER_NODE - 1];
123}
124
125/*
126 * Fills in a level of the btree based on the highs of the level
127 * below it.
128 */
129static int setup_btree_index(unsigned int l, struct dm_table *t)
130{
131 unsigned int n, k;
132 sector_t *node;
133
134 for (n = 0U; n < t->counts[l]; n++) {
135 node = get_node(t, l, n);
136
137 for (k = 0U; k < KEYS_PER_NODE; k++)
138 node[k] = high(t, l + 1, get_child(n, k));
139 }
140
141 return 0;
142}
143
144void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
145{
146 unsigned long size;
147 void *addr;
148
149 /*
150 * Check that we're not going to overflow.
151 */
152 if (nmemb > (ULONG_MAX / elem_size))
153 return NULL;
154
155 size = nmemb * elem_size;
156 addr = vzalloc(size);
157
158 return addr;
159}
160EXPORT_SYMBOL(dm_vcalloc);
161
162/*
163 * highs, and targets are managed as dynamic arrays during a
164 * table load.
165 */
166static int alloc_targets(struct dm_table *t, unsigned int num)
167{
168 sector_t *n_highs;
169 struct dm_target *n_targets;
170 int n = t->num_targets;
171
172 /*
173 * Allocate both the target array and offset array at once.
174 * Append an empty entry to catch sectors beyond the end of
175 * the device.
176 */
177 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
178 sizeof(sector_t));
179 if (!n_highs)
180 return -ENOMEM;
181
182 n_targets = (struct dm_target *) (n_highs + num);
183
184 if (n) {
185 memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
186 memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
187 }
188
189 memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
190 vfree(t->highs);
191
192 t->num_allocated = num;
193 t->highs = n_highs;
194 t->targets = n_targets;
195
196 return 0;
197}
198
199int dm_table_create(struct dm_table **result, fmode_t mode,
200 unsigned num_targets, struct mapped_device *md)
201{
202 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
203
204 if (!t)
205 return -ENOMEM;
206
207 INIT_LIST_HEAD(&t->devices);
208 INIT_LIST_HEAD(&t->target_callbacks);
209 atomic_set(&t->holders, 0);
210
211 if (!num_targets)
212 num_targets = KEYS_PER_NODE;
213
214 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
215
216 if (alloc_targets(t, num_targets)) {
217 kfree(t);
218 t = NULL;
219 return -ENOMEM;
220 }
221
222 t->mode = mode;
223 t->md = md;
224 *result = t;
225 return 0;
226}
227
228static void free_devices(struct list_head *devices)
229{
230 struct list_head *tmp, *next;
231
232 list_for_each_safe(tmp, next, devices) {
233 struct dm_dev_internal *dd =
234 list_entry(tmp, struct dm_dev_internal, list);
235 DMWARN("dm_table_destroy: dm_put_device call missing for %s",
236 dd->dm_dev.name);
237 kfree(dd);
238 }
239}
240
241void dm_table_destroy(struct dm_table *t)
242{
243 unsigned int i;
244
245 if (!t)
246 return;
247
248 while (atomic_read(&t->holders))
249 msleep(1);
250 smp_mb();
251
252 /* free the indexes */
253 if (t->depth >= 2)
254 vfree(t->index[t->depth - 2]);
255
256 /* free the targets */
257 for (i = 0; i < t->num_targets; i++) {
258 struct dm_target *tgt = t->targets + i;
259
260 if (tgt->type->dtr)
261 tgt->type->dtr(tgt);
262
263 dm_put_target_type(tgt->type);
264 }
265
266 vfree(t->highs);
267
268 /* free the device list */
269 if (t->devices.next != &t->devices)
270 free_devices(&t->devices);
271
272 dm_free_md_mempools(t->mempools);
273
274 kfree(t);
275}
276
277void dm_table_get(struct dm_table *t)
278{
279 atomic_inc(&t->holders);
280}
281EXPORT_SYMBOL(dm_table_get);
282
283void dm_table_put(struct dm_table *t)
284{
285 if (!t)
286 return;
287
288 smp_mb__before_atomic_dec();
289 atomic_dec(&t->holders);
290}
291EXPORT_SYMBOL(dm_table_put);
292
293/*
294 * Checks to see if we need to extend highs or targets.
295 */
296static inline int check_space(struct dm_table *t)
297{
298 if (t->num_targets >= t->num_allocated)
299 return alloc_targets(t, t->num_allocated * 2);
300
301 return 0;
302}
303
304/*
305 * See if we've already got a device in the list.
306 */
307static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
308{
309 struct dm_dev_internal *dd;
310
311 list_for_each_entry (dd, l, list)
312 if (dd->dm_dev.bdev->bd_dev == dev)
313 return dd;
314
315 return NULL;
316}
317
318/*
319 * Open a device so we can use it as a map destination.
320 */
321static int open_dev(struct dm_dev_internal *d, dev_t dev,
322 struct mapped_device *md)
323{
324 static char *_claim_ptr = "I belong to device-mapper";
325 struct block_device *bdev;
326
327 int r;
328
329 BUG_ON(d->dm_dev.bdev);
330
331 bdev = blkdev_get_by_dev(dev, d->dm_dev.mode | FMODE_EXCL, _claim_ptr);
332 if (IS_ERR(bdev))
333 return PTR_ERR(bdev);
334
335 r = bd_link_disk_holder(bdev, dm_disk(md));
336 if (r) {
337 blkdev_put(bdev, d->dm_dev.mode | FMODE_EXCL);
338 return r;
339 }
340
341 d->dm_dev.bdev = bdev;
342 return 0;
343}
344
345/*
346 * Close a device that we've been using.
347 */
348static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
349{
350 if (!d->dm_dev.bdev)
351 return;
352
353 bd_unlink_disk_holder(d->dm_dev.bdev, dm_disk(md));
354 blkdev_put(d->dm_dev.bdev, d->dm_dev.mode | FMODE_EXCL);
355 d->dm_dev.bdev = NULL;
356}
357
358/*
359 * If possible, this checks an area of a destination device is invalid.
360 */
361static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
362 sector_t start, sector_t len, void *data)
363{
364 struct request_queue *q;
365 struct queue_limits *limits = data;
366 struct block_device *bdev = dev->bdev;
367 sector_t dev_size =
368 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
369 unsigned short logical_block_size_sectors =
370 limits->logical_block_size >> SECTOR_SHIFT;
371 char b[BDEVNAME_SIZE];
372
373 /*
374 * Some devices exist without request functions,
375 * such as loop devices not yet bound to backing files.
376 * Forbid the use of such devices.
377 */
378 q = bdev_get_queue(bdev);
379 if (!q || !q->make_request_fn) {
380 DMWARN("%s: %s is not yet initialised: "
381 "start=%llu, len=%llu, dev_size=%llu",
382 dm_device_name(ti->table->md), bdevname(bdev, b),
383 (unsigned long long)start,
384 (unsigned long long)len,
385 (unsigned long long)dev_size);
386 return 1;
387 }
388
389 if (!dev_size)
390 return 0;
391
392 if ((start >= dev_size) || (start + len > dev_size)) {
393 DMWARN("%s: %s too small for target: "
394 "start=%llu, len=%llu, dev_size=%llu",
395 dm_device_name(ti->table->md), bdevname(bdev, b),
396 (unsigned long long)start,
397 (unsigned long long)len,
398 (unsigned long long)dev_size);
399 return 1;
400 }
401
402 if (logical_block_size_sectors <= 1)
403 return 0;
404
405 if (start & (logical_block_size_sectors - 1)) {
406 DMWARN("%s: start=%llu not aligned to h/w "
407 "logical block size %u of %s",
408 dm_device_name(ti->table->md),
409 (unsigned long long)start,
410 limits->logical_block_size, bdevname(bdev, b));
411 return 1;
412 }
413
414 if (len & (logical_block_size_sectors - 1)) {
415 DMWARN("%s: len=%llu not aligned to h/w "
416 "logical block size %u of %s",
417 dm_device_name(ti->table->md),
418 (unsigned long long)len,
419 limits->logical_block_size, bdevname(bdev, b));
420 return 1;
421 }
422
423 return 0;
424}
425
426/*
427 * This upgrades the mode on an already open dm_dev, being
428 * careful to leave things as they were if we fail to reopen the
429 * device and not to touch the existing bdev field in case
430 * it is accessed concurrently inside dm_table_any_congested().
431 */
432static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
433 struct mapped_device *md)
434{
435 int r;
436 struct dm_dev_internal dd_new, dd_old;
437
438 dd_new = dd_old = *dd;
439
440 dd_new.dm_dev.mode |= new_mode;
441 dd_new.dm_dev.bdev = NULL;
442
443 r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md);
444 if (r)
445 return r;
446
447 dd->dm_dev.mode |= new_mode;
448 close_dev(&dd_old, md);
449
450 return 0;
451}
452
453/*
454 * Add a device to the list, or just increment the usage count if
455 * it's already present.
456 */
457int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
458 struct dm_dev **result)
459{
460 int r;
461 dev_t uninitialized_var(dev);
462 struct dm_dev_internal *dd;
463 unsigned int major, minor;
464 struct dm_table *t = ti->table;
465
466 BUG_ON(!t);
467
468 if (sscanf(path, "%u:%u", &major, &minor) == 2) {
469 /* Extract the major/minor numbers */
470 dev = MKDEV(major, minor);
471 if (MAJOR(dev) != major || MINOR(dev) != minor)
472 return -EOVERFLOW;
473 } else {
474 /* convert the path to a device */
475 struct block_device *bdev = lookup_bdev(path);
476
477 if (IS_ERR(bdev))
478 return PTR_ERR(bdev);
479 dev = bdev->bd_dev;
480 bdput(bdev);
481 }
482
483 dd = find_device(&t->devices, dev);
484 if (!dd) {
485 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
486 if (!dd)
487 return -ENOMEM;
488
489 dd->dm_dev.mode = mode;
490 dd->dm_dev.bdev = NULL;
491
492 if ((r = open_dev(dd, dev, t->md))) {
493 kfree(dd);
494 return r;
495 }
496
497 format_dev_t(dd->dm_dev.name, dev);
498
499 atomic_set(&dd->count, 0);
500 list_add(&dd->list, &t->devices);
501
502 } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
503 r = upgrade_mode(dd, mode, t->md);
504 if (r)
505 return r;
506 }
507 atomic_inc(&dd->count);
508
509 *result = &dd->dm_dev;
510 return 0;
511}
512EXPORT_SYMBOL(dm_get_device);
513
514int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
515 sector_t start, sector_t len, void *data)
516{
517 struct queue_limits *limits = data;
518 struct block_device *bdev = dev->bdev;
519 struct request_queue *q = bdev_get_queue(bdev);
520 char b[BDEVNAME_SIZE];
521
522 if (unlikely(!q)) {
523 DMWARN("%s: Cannot set limits for nonexistent device %s",
524 dm_device_name(ti->table->md), bdevname(bdev, b));
525 return 0;
526 }
527
528 if (bdev_stack_limits(limits, bdev, start) < 0)
529 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
530 "physical_block_size=%u, logical_block_size=%u, "
531 "alignment_offset=%u, start=%llu",
532 dm_device_name(ti->table->md), bdevname(bdev, b),
533 q->limits.physical_block_size,
534 q->limits.logical_block_size,
535 q->limits.alignment_offset,
536 (unsigned long long) start << SECTOR_SHIFT);
537
538 /*
539 * Check if merge fn is supported.
540 * If not we'll force DM to use PAGE_SIZE or
541 * smaller I/O, just to be safe.
542 */
543 if (dm_queue_merge_is_compulsory(q) && !ti->type->merge)
544 blk_limits_max_hw_sectors(limits,
545 (unsigned int) (PAGE_SIZE >> 9));
546 return 0;
547}
548EXPORT_SYMBOL_GPL(dm_set_device_limits);
549
550/*
551 * Decrement a device's use count and remove it if necessary.
552 */
553void dm_put_device(struct dm_target *ti, struct dm_dev *d)
554{
555 struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
556 dm_dev);
557
558 if (atomic_dec_and_test(&dd->count)) {
559 close_dev(dd, ti->table->md);
560 list_del(&dd->list);
561 kfree(dd);
562 }
563}
564EXPORT_SYMBOL(dm_put_device);
565
566/*
567 * Checks to see if the target joins onto the end of the table.
568 */
569static int adjoin(struct dm_table *table, struct dm_target *ti)
570{
571 struct dm_target *prev;
572
573 if (!table->num_targets)
574 return !ti->begin;
575
576 prev = &table->targets[table->num_targets - 1];
577 return (ti->begin == (prev->begin + prev->len));
578}
579
580/*
581 * Used to dynamically allocate the arg array.
582 */
583static char **realloc_argv(unsigned *array_size, char **old_argv)
584{
585 char **argv;
586 unsigned new_size;
587
588 new_size = *array_size ? *array_size * 2 : 64;
589 argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
590 if (argv) {
591 memcpy(argv, old_argv, *array_size * sizeof(*argv));
592 *array_size = new_size;
593 }
594
595 kfree(old_argv);
596 return argv;
597}
598
599/*
600 * Destructively splits up the argument list to pass to ctr.
601 */
602int dm_split_args(int *argc, char ***argvp, char *input)
603{
604 char *start, *end = input, *out, **argv = NULL;
605 unsigned array_size = 0;
606
607 *argc = 0;
608
609 if (!input) {
610 *argvp = NULL;
611 return 0;
612 }
613
614 argv = realloc_argv(&array_size, argv);
615 if (!argv)
616 return -ENOMEM;
617
618 while (1) {
619 /* Skip whitespace */
620 start = skip_spaces(end);
621
622 if (!*start)
623 break; /* success, we hit the end */
624
625 /* 'out' is used to remove any back-quotes */
626 end = out = start;
627 while (*end) {
628 /* Everything apart from '\0' can be quoted */
629 if (*end == '\\' && *(end + 1)) {
630 *out++ = *(end + 1);
631 end += 2;
632 continue;
633 }
634
635 if (isspace(*end))
636 break; /* end of token */
637
638 *out++ = *end++;
639 }
640
641 /* have we already filled the array ? */
642 if ((*argc + 1) > array_size) {
643 argv = realloc_argv(&array_size, argv);
644 if (!argv)
645 return -ENOMEM;
646 }
647
648 /* we know this is whitespace */
649 if (*end)
650 end++;
651
652 /* terminate the string and put it in the array */
653 *out = '\0';
654 argv[*argc] = start;
655 (*argc)++;
656 }
657
658 *argvp = argv;
659 return 0;
660}
661
662/*
663 * Impose necessary and sufficient conditions on a devices's table such
664 * that any incoming bio which respects its logical_block_size can be
665 * processed successfully. If it falls across the boundary between
666 * two or more targets, the size of each piece it gets split into must
667 * be compatible with the logical_block_size of the target processing it.
668 */
669static int validate_hardware_logical_block_alignment(struct dm_table *table,
670 struct queue_limits *limits)
671{
672 /*
673 * This function uses arithmetic modulo the logical_block_size
674 * (in units of 512-byte sectors).
675 */
676 unsigned short device_logical_block_size_sects =
677 limits->logical_block_size >> SECTOR_SHIFT;
678
679 /*
680 * Offset of the start of the next table entry, mod logical_block_size.
681 */
682 unsigned short next_target_start = 0;
683
684 /*
685 * Given an aligned bio that extends beyond the end of a
686 * target, how many sectors must the next target handle?
687 */
688 unsigned short remaining = 0;
689
690 struct dm_target *uninitialized_var(ti);
691 struct queue_limits ti_limits;
692 unsigned i = 0;
693
694 /*
695 * Check each entry in the table in turn.
696 */
697 while (i < dm_table_get_num_targets(table)) {
698 ti = dm_table_get_target(table, i++);
699
700 blk_set_default_limits(&ti_limits);
701
702 /* combine all target devices' limits */
703 if (ti->type->iterate_devices)
704 ti->type->iterate_devices(ti, dm_set_device_limits,
705 &ti_limits);
706
707 /*
708 * If the remaining sectors fall entirely within this
709 * table entry are they compatible with its logical_block_size?
710 */
711 if (remaining < ti->len &&
712 remaining & ((ti_limits.logical_block_size >>
713 SECTOR_SHIFT) - 1))
714 break; /* Error */
715
716 next_target_start =
717 (unsigned short) ((next_target_start + ti->len) &
718 (device_logical_block_size_sects - 1));
719 remaining = next_target_start ?
720 device_logical_block_size_sects - next_target_start : 0;
721 }
722
723 if (remaining) {
724 DMWARN("%s: table line %u (start sect %llu len %llu) "
725 "not aligned to h/w logical block size %u",
726 dm_device_name(table->md), i,
727 (unsigned long long) ti->begin,
728 (unsigned long long) ti->len,
729 limits->logical_block_size);
730 return -EINVAL;
731 }
732
733 return 0;
734}
735
736int dm_table_add_target(struct dm_table *t, const char *type,
737 sector_t start, sector_t len, char *params)
738{
739 int r = -EINVAL, argc;
740 char **argv;
741 struct dm_target *tgt;
742
743 if ((r = check_space(t)))
744 return r;
745
746 tgt = t->targets + t->num_targets;
747 memset(tgt, 0, sizeof(*tgt));
748
749 if (!len) {
750 DMERR("%s: zero-length target", dm_device_name(t->md));
751 return -EINVAL;
752 }
753
754 tgt->type = dm_get_target_type(type);
755 if (!tgt->type) {
756 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
757 type);
758 return -EINVAL;
759 }
760
761 tgt->table = t;
762 tgt->begin = start;
763 tgt->len = len;
764 tgt->error = "Unknown error";
765
766 /*
767 * Does this target adjoin the previous one ?
768 */
769 if (!adjoin(t, tgt)) {
770 tgt->error = "Gap in table";
771 r = -EINVAL;
772 goto bad;
773 }
774
775 r = dm_split_args(&argc, &argv, params);
776 if (r) {
777 tgt->error = "couldn't split parameters (insufficient memory)";
778 goto bad;
779 }
780
781 r = tgt->type->ctr(tgt, argc, argv);
782 kfree(argv);
783 if (r)
784 goto bad;
785
786 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
787
788 if (!tgt->num_discard_requests && tgt->discards_supported)
789 DMWARN("%s: %s: ignoring discards_supported because num_discard_requests is zero.",
790 dm_device_name(t->md), type);
791
792 return 0;
793
794 bad:
795 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
796 dm_put_target_type(tgt->type);
797 return r;
798}
799
800/*
801 * Target argument parsing helpers.
802 */
803static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
804 unsigned *value, char **error, unsigned grouped)
805{
806 const char *arg_str = dm_shift_arg(arg_set);
807
808 if (!arg_str ||
809 (sscanf(arg_str, "%u", value) != 1) ||
810 (*value < arg->min) ||
811 (*value > arg->max) ||
812 (grouped && arg_set->argc < *value)) {
813 *error = arg->error;
814 return -EINVAL;
815 }
816
817 return 0;
818}
819
820int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
821 unsigned *value, char **error)
822{
823 return validate_next_arg(arg, arg_set, value, error, 0);
824}
825EXPORT_SYMBOL(dm_read_arg);
826
827int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
828 unsigned *value, char **error)
829{
830 return validate_next_arg(arg, arg_set, value, error, 1);
831}
832EXPORT_SYMBOL(dm_read_arg_group);
833
834const char *dm_shift_arg(struct dm_arg_set *as)
835{
836 char *r;
837
838 if (as->argc) {
839 as->argc--;
840 r = *as->argv;
841 as->argv++;
842 return r;
843 }
844
845 return NULL;
846}
847EXPORT_SYMBOL(dm_shift_arg);
848
849void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
850{
851 BUG_ON(as->argc < num_args);
852 as->argc -= num_args;
853 as->argv += num_args;
854}
855EXPORT_SYMBOL(dm_consume_args);
856
857static int dm_table_set_type(struct dm_table *t)
858{
859 unsigned i;
860 unsigned bio_based = 0, request_based = 0;
861 struct dm_target *tgt;
862 struct dm_dev_internal *dd;
863 struct list_head *devices;
864
865 for (i = 0; i < t->num_targets; i++) {
866 tgt = t->targets + i;
867 if (dm_target_request_based(tgt))
868 request_based = 1;
869 else
870 bio_based = 1;
871
872 if (bio_based && request_based) {
873 DMWARN("Inconsistent table: different target types"
874 " can't be mixed up");
875 return -EINVAL;
876 }
877 }
878
879 if (bio_based) {
880 /* We must use this table as bio-based */
881 t->type = DM_TYPE_BIO_BASED;
882 return 0;
883 }
884
885 BUG_ON(!request_based); /* No targets in this table */
886
887 /* Non-request-stackable devices can't be used for request-based dm */
888 devices = dm_table_get_devices(t);
889 list_for_each_entry(dd, devices, list) {
890 if (!blk_queue_stackable(bdev_get_queue(dd->dm_dev.bdev))) {
891 DMWARN("table load rejected: including"
892 " non-request-stackable devices");
893 return -EINVAL;
894 }
895 }
896
897 /*
898 * Request-based dm supports only tables that have a single target now.
899 * To support multiple targets, request splitting support is needed,
900 * and that needs lots of changes in the block-layer.
901 * (e.g. request completion process for partial completion.)
902 */
903 if (t->num_targets > 1) {
904 DMWARN("Request-based dm doesn't support multiple targets yet");
905 return -EINVAL;
906 }
907
908 t->type = DM_TYPE_REQUEST_BASED;
909
910 return 0;
911}
912
913unsigned dm_table_get_type(struct dm_table *t)
914{
915 return t->type;
916}
917
918bool dm_table_request_based(struct dm_table *t)
919{
920 return dm_table_get_type(t) == DM_TYPE_REQUEST_BASED;
921}
922
923int dm_table_alloc_md_mempools(struct dm_table *t)
924{
925 unsigned type = dm_table_get_type(t);
926
927 if (unlikely(type == DM_TYPE_NONE)) {
928 DMWARN("no table type is set, can't allocate mempools");
929 return -EINVAL;
930 }
931
932 t->mempools = dm_alloc_md_mempools(type, t->integrity_supported);
933 if (!t->mempools)
934 return -ENOMEM;
935
936 return 0;
937}
938
939void dm_table_free_md_mempools(struct dm_table *t)
940{
941 dm_free_md_mempools(t->mempools);
942 t->mempools = NULL;
943}
944
945struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
946{
947 return t->mempools;
948}
949
950static int setup_indexes(struct dm_table *t)
951{
952 int i;
953 unsigned int total = 0;
954 sector_t *indexes;
955
956 /* allocate the space for *all* the indexes */
957 for (i = t->depth - 2; i >= 0; i--) {
958 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
959 total += t->counts[i];
960 }
961
962 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
963 if (!indexes)
964 return -ENOMEM;
965
966 /* set up internal nodes, bottom-up */
967 for (i = t->depth - 2; i >= 0; i--) {
968 t->index[i] = indexes;
969 indexes += (KEYS_PER_NODE * t->counts[i]);
970 setup_btree_index(i, t);
971 }
972
973 return 0;
974}
975
976/*
977 * Builds the btree to index the map.
978 */
979static int dm_table_build_index(struct dm_table *t)
980{
981 int r = 0;
982 unsigned int leaf_nodes;
983
984 /* how many indexes will the btree have ? */
985 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
986 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
987
988 /* leaf layer has already been set up */
989 t->counts[t->depth - 1] = leaf_nodes;
990 t->index[t->depth - 1] = t->highs;
991
992 if (t->depth >= 2)
993 r = setup_indexes(t);
994
995 return r;
996}
997
998/*
999 * Get a disk whose integrity profile reflects the table's profile.
1000 * If %match_all is true, all devices' profiles must match.
1001 * If %match_all is false, all devices must at least have an
1002 * allocated integrity profile; but uninitialized is ok.
1003 * Returns NULL if integrity support was inconsistent or unavailable.
1004 */
1005static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t,
1006 bool match_all)
1007{
1008 struct list_head *devices = dm_table_get_devices(t);
1009 struct dm_dev_internal *dd = NULL;
1010 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1011
1012 list_for_each_entry(dd, devices, list) {
1013 template_disk = dd->dm_dev.bdev->bd_disk;
1014 if (!blk_get_integrity(template_disk))
1015 goto no_integrity;
1016 if (!match_all && !blk_integrity_is_initialized(template_disk))
1017 continue; /* skip uninitialized profiles */
1018 else if (prev_disk &&
1019 blk_integrity_compare(prev_disk, template_disk) < 0)
1020 goto no_integrity;
1021 prev_disk = template_disk;
1022 }
1023
1024 return template_disk;
1025
1026no_integrity:
1027 if (prev_disk)
1028 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1029 dm_device_name(t->md),
1030 prev_disk->disk_name,
1031 template_disk->disk_name);
1032 return NULL;
1033}
1034
1035/*
1036 * Register the mapped device for blk_integrity support if
1037 * the underlying devices have an integrity profile. But all devices
1038 * may not have matching profiles (checking all devices isn't reliable
1039 * during table load because this table may use other DM device(s) which
1040 * must be resumed before they will have an initialized integity profile).
1041 * Stacked DM devices force a 2 stage integrity profile validation:
1042 * 1 - during load, validate all initialized integrity profiles match
1043 * 2 - during resume, validate all integrity profiles match
1044 */
1045static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md)
1046{
1047 struct gendisk *template_disk = NULL;
1048
1049 template_disk = dm_table_get_integrity_disk(t, false);
1050 if (!template_disk)
1051 return 0;
1052
1053 if (!blk_integrity_is_initialized(dm_disk(md))) {
1054 t->integrity_supported = 1;
1055 return blk_integrity_register(dm_disk(md), NULL);
1056 }
1057
1058 /*
1059 * If DM device already has an initalized integrity
1060 * profile the new profile should not conflict.
1061 */
1062 if (blk_integrity_is_initialized(template_disk) &&
1063 blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1064 DMWARN("%s: conflict with existing integrity profile: "
1065 "%s profile mismatch",
1066 dm_device_name(t->md),
1067 template_disk->disk_name);
1068 return 1;
1069 }
1070
1071 /* Preserve existing initialized integrity profile */
1072 t->integrity_supported = 1;
1073 return 0;
1074}
1075
1076/*
1077 * Prepares the table for use by building the indices,
1078 * setting the type, and allocating mempools.
1079 */
1080int dm_table_complete(struct dm_table *t)
1081{
1082 int r;
1083
1084 r = dm_table_set_type(t);
1085 if (r) {
1086 DMERR("unable to set table type");
1087 return r;
1088 }
1089
1090 r = dm_table_build_index(t);
1091 if (r) {
1092 DMERR("unable to build btrees");
1093 return r;
1094 }
1095
1096 r = dm_table_prealloc_integrity(t, t->md);
1097 if (r) {
1098 DMERR("could not register integrity profile.");
1099 return r;
1100 }
1101
1102 r = dm_table_alloc_md_mempools(t);
1103 if (r)
1104 DMERR("unable to allocate mempools");
1105
1106 return r;
1107}
1108
1109static DEFINE_MUTEX(_event_lock);
1110void dm_table_event_callback(struct dm_table *t,
1111 void (*fn)(void *), void *context)
1112{
1113 mutex_lock(&_event_lock);
1114 t->event_fn = fn;
1115 t->event_context = context;
1116 mutex_unlock(&_event_lock);
1117}
1118
1119void dm_table_event(struct dm_table *t)
1120{
1121 /*
1122 * You can no longer call dm_table_event() from interrupt
1123 * context, use a bottom half instead.
1124 */
1125 BUG_ON(in_interrupt());
1126
1127 mutex_lock(&_event_lock);
1128 if (t->event_fn)
1129 t->event_fn(t->event_context);
1130 mutex_unlock(&_event_lock);
1131}
1132EXPORT_SYMBOL(dm_table_event);
1133
1134sector_t dm_table_get_size(struct dm_table *t)
1135{
1136 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1137}
1138EXPORT_SYMBOL(dm_table_get_size);
1139
1140struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1141{
1142 if (index >= t->num_targets)
1143 return NULL;
1144
1145 return t->targets + index;
1146}
1147
1148/*
1149 * Search the btree for the correct target.
1150 *
1151 * Caller should check returned pointer with dm_target_is_valid()
1152 * to trap I/O beyond end of device.
1153 */
1154struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1155{
1156 unsigned int l, n = 0, k = 0;
1157 sector_t *node;
1158
1159 for (l = 0; l < t->depth; l++) {
1160 n = get_child(n, k);
1161 node = get_node(t, l, n);
1162
1163 for (k = 0; k < KEYS_PER_NODE; k++)
1164 if (node[k] >= sector)
1165 break;
1166 }
1167
1168 return &t->targets[(KEYS_PER_NODE * n) + k];
1169}
1170
1171/*
1172 * Establish the new table's queue_limits and validate them.
1173 */
1174int dm_calculate_queue_limits(struct dm_table *table,
1175 struct queue_limits *limits)
1176{
1177 struct dm_target *uninitialized_var(ti);
1178 struct queue_limits ti_limits;
1179 unsigned i = 0;
1180
1181 blk_set_default_limits(limits);
1182
1183 while (i < dm_table_get_num_targets(table)) {
1184 blk_set_default_limits(&ti_limits);
1185
1186 ti = dm_table_get_target(table, i++);
1187
1188 if (!ti->type->iterate_devices)
1189 goto combine_limits;
1190
1191 /*
1192 * Combine queue limits of all the devices this target uses.
1193 */
1194 ti->type->iterate_devices(ti, dm_set_device_limits,
1195 &ti_limits);
1196
1197 /* Set I/O hints portion of queue limits */
1198 if (ti->type->io_hints)
1199 ti->type->io_hints(ti, &ti_limits);
1200
1201 /*
1202 * Check each device area is consistent with the target's
1203 * overall queue limits.
1204 */
1205 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1206 &ti_limits))
1207 return -EINVAL;
1208
1209combine_limits:
1210 /*
1211 * Merge this target's queue limits into the overall limits
1212 * for the table.
1213 */
1214 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1215 DMWARN("%s: adding target device "
1216 "(start sect %llu len %llu) "
1217 "caused an alignment inconsistency",
1218 dm_device_name(table->md),
1219 (unsigned long long) ti->begin,
1220 (unsigned long long) ti->len);
1221 }
1222
1223 return validate_hardware_logical_block_alignment(table, limits);
1224}
1225
1226/*
1227 * Set the integrity profile for this device if all devices used have
1228 * matching profiles. We're quite deep in the resume path but still
1229 * don't know if all devices (particularly DM devices this device
1230 * may be stacked on) have matching profiles. Even if the profiles
1231 * don't match we have no way to fail (to resume) at this point.
1232 */
1233static void dm_table_set_integrity(struct dm_table *t)
1234{
1235 struct gendisk *template_disk = NULL;
1236
1237 if (!blk_get_integrity(dm_disk(t->md)))
1238 return;
1239
1240 template_disk = dm_table_get_integrity_disk(t, true);
1241 if (template_disk)
1242 blk_integrity_register(dm_disk(t->md),
1243 blk_get_integrity(template_disk));
1244 else if (blk_integrity_is_initialized(dm_disk(t->md)))
1245 DMWARN("%s: device no longer has a valid integrity profile",
1246 dm_device_name(t->md));
1247 else
1248 DMWARN("%s: unable to establish an integrity profile",
1249 dm_device_name(t->md));
1250}
1251
1252static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1253 sector_t start, sector_t len, void *data)
1254{
1255 unsigned flush = (*(unsigned *)data);
1256 struct request_queue *q = bdev_get_queue(dev->bdev);
1257
1258 return q && (q->flush_flags & flush);
1259}
1260
1261static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1262{
1263 struct dm_target *ti;
1264 unsigned i = 0;
1265
1266 /*
1267 * Require at least one underlying device to support flushes.
1268 * t->devices includes internal dm devices such as mirror logs
1269 * so we need to use iterate_devices here, which targets
1270 * supporting flushes must provide.
1271 */
1272 while (i < dm_table_get_num_targets(t)) {
1273 ti = dm_table_get_target(t, i++);
1274
1275 if (!ti->num_flush_requests)
1276 continue;
1277
1278 if (ti->type->iterate_devices &&
1279 ti->type->iterate_devices(ti, device_flush_capable, &flush))
1280 return 1;
1281 }
1282
1283 return 0;
1284}
1285
1286static bool dm_table_discard_zeroes_data(struct dm_table *t)
1287{
1288 struct dm_target *ti;
1289 unsigned i = 0;
1290
1291 /* Ensure that all targets supports discard_zeroes_data. */
1292 while (i < dm_table_get_num_targets(t)) {
1293 ti = dm_table_get_target(t, i++);
1294
1295 if (ti->discard_zeroes_data_unsupported)
1296 return 0;
1297 }
1298
1299 return 1;
1300}
1301
1302void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1303 struct queue_limits *limits)
1304{
1305 unsigned flush = 0;
1306
1307 /*
1308 * Copy table's limits to the DM device's request_queue
1309 */
1310 q->limits = *limits;
1311
1312 if (!dm_table_supports_discards(t))
1313 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1314 else
1315 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1316
1317 if (dm_table_supports_flush(t, REQ_FLUSH)) {
1318 flush |= REQ_FLUSH;
1319 if (dm_table_supports_flush(t, REQ_FUA))
1320 flush |= REQ_FUA;
1321 }
1322 blk_queue_flush(q, flush);
1323
1324 if (!dm_table_discard_zeroes_data(t))
1325 q->limits.discard_zeroes_data = 0;
1326
1327 dm_table_set_integrity(t);
1328
1329 /*
1330 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1331 * visible to other CPUs because, once the flag is set, incoming bios
1332 * are processed by request-based dm, which refers to the queue
1333 * settings.
1334 * Until the flag set, bios are passed to bio-based dm and queued to
1335 * md->deferred where queue settings are not needed yet.
1336 * Those bios are passed to request-based dm at the resume time.
1337 */
1338 smp_mb();
1339 if (dm_table_request_based(t))
1340 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1341}
1342
1343unsigned int dm_table_get_num_targets(struct dm_table *t)
1344{
1345 return t->num_targets;
1346}
1347
1348struct list_head *dm_table_get_devices(struct dm_table *t)
1349{
1350 return &t->devices;
1351}
1352
1353fmode_t dm_table_get_mode(struct dm_table *t)
1354{
1355 return t->mode;
1356}
1357EXPORT_SYMBOL(dm_table_get_mode);
1358
1359static void suspend_targets(struct dm_table *t, unsigned postsuspend)
1360{
1361 int i = t->num_targets;
1362 struct dm_target *ti = t->targets;
1363
1364 while (i--) {
1365 if (postsuspend) {
1366 if (ti->type->postsuspend)
1367 ti->type->postsuspend(ti);
1368 } else if (ti->type->presuspend)
1369 ti->type->presuspend(ti);
1370
1371 ti++;
1372 }
1373}
1374
1375void dm_table_presuspend_targets(struct dm_table *t)
1376{
1377 if (!t)
1378 return;
1379
1380 suspend_targets(t, 0);
1381}
1382
1383void dm_table_postsuspend_targets(struct dm_table *t)
1384{
1385 if (!t)
1386 return;
1387
1388 suspend_targets(t, 1);
1389}
1390
1391int dm_table_resume_targets(struct dm_table *t)
1392{
1393 int i, r = 0;
1394
1395 for (i = 0; i < t->num_targets; i++) {
1396 struct dm_target *ti = t->targets + i;
1397
1398 if (!ti->type->preresume)
1399 continue;
1400
1401 r = ti->type->preresume(ti);
1402 if (r)
1403 return r;
1404 }
1405
1406 for (i = 0; i < t->num_targets; i++) {
1407 struct dm_target *ti = t->targets + i;
1408
1409 if (ti->type->resume)
1410 ti->type->resume(ti);
1411 }
1412
1413 return 0;
1414}
1415
1416void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1417{
1418 list_add(&cb->list, &t->target_callbacks);
1419}
1420EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1421
1422int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1423{
1424 struct dm_dev_internal *dd;
1425 struct list_head *devices = dm_table_get_devices(t);
1426 struct dm_target_callbacks *cb;
1427 int r = 0;
1428
1429 list_for_each_entry(dd, devices, list) {
1430 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
1431 char b[BDEVNAME_SIZE];
1432
1433 if (likely(q))
1434 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1435 else
1436 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1437 dm_device_name(t->md),
1438 bdevname(dd->dm_dev.bdev, b));
1439 }
1440
1441 list_for_each_entry(cb, &t->target_callbacks, list)
1442 if (cb->congested_fn)
1443 r |= cb->congested_fn(cb, bdi_bits);
1444
1445 return r;
1446}
1447
1448int dm_table_any_busy_target(struct dm_table *t)
1449{
1450 unsigned i;
1451 struct dm_target *ti;
1452
1453 for (i = 0; i < t->num_targets; i++) {
1454 ti = t->targets + i;
1455 if (ti->type->busy && ti->type->busy(ti))
1456 return 1;
1457 }
1458
1459 return 0;
1460}
1461
1462struct mapped_device *dm_table_get_md(struct dm_table *t)
1463{
1464 return t->md;
1465}
1466EXPORT_SYMBOL(dm_table_get_md);
1467
1468static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1469 sector_t start, sector_t len, void *data)
1470{
1471 struct request_queue *q = bdev_get_queue(dev->bdev);
1472
1473 return q && blk_queue_discard(q);
1474}
1475
1476bool dm_table_supports_discards(struct dm_table *t)
1477{
1478 struct dm_target *ti;
1479 unsigned i = 0;
1480
1481 /*
1482 * Unless any target used by the table set discards_supported,
1483 * require at least one underlying device to support discards.
1484 * t->devices includes internal dm devices such as mirror logs
1485 * so we need to use iterate_devices here, which targets
1486 * supporting discard selectively must provide.
1487 */
1488 while (i < dm_table_get_num_targets(t)) {
1489 ti = dm_table_get_target(t, i++);
1490
1491 if (!ti->num_discard_requests)
1492 continue;
1493
1494 if (ti->discards_supported)
1495 return 1;
1496
1497 if (ti->type->iterate_devices &&
1498 ti->type->iterate_devices(ti, device_discard_capable, NULL))
1499 return 1;
1500 }
1501
1502 return 0;
1503}
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
32struct dm_table {
33 struct mapped_device *md;
34 enum dm_queue_mode type;
35
36 /* btree table */
37 unsigned int depth;
38 unsigned int counts[MAX_DEPTH]; /* in nodes */
39 sector_t *index[MAX_DEPTH];
40
41 unsigned int num_targets;
42 unsigned int num_allocated;
43 sector_t *highs;
44 struct dm_target *targets;
45
46 struct target_type *immutable_target_type;
47
48 bool integrity_supported:1;
49 bool singleton:1;
50 bool all_blk_mq:1;
51 unsigned integrity_added:1;
52
53 /*
54 * Indicates the rw permissions for the new logical
55 * device. This should be a combination of FMODE_READ
56 * and FMODE_WRITE.
57 */
58 fmode_t mode;
59
60 /* a list of devices used by this table */
61 struct list_head devices;
62
63 /* events get handed up using this callback */
64 void (*event_fn)(void *);
65 void *event_context;
66
67 struct dm_md_mempools *mempools;
68
69 struct list_head target_callbacks;
70};
71
72/*
73 * Similar to ceiling(log_size(n))
74 */
75static unsigned int int_log(unsigned int n, unsigned int base)
76{
77 int result = 0;
78
79 while (n > 1) {
80 n = dm_div_up(n, base);
81 result++;
82 }
83
84 return result;
85}
86
87/*
88 * Calculate the index of the child node of the n'th node k'th key.
89 */
90static inline unsigned int get_child(unsigned int n, unsigned int k)
91{
92 return (n * CHILDREN_PER_NODE) + k;
93}
94
95/*
96 * Return the n'th node of level l from table t.
97 */
98static inline sector_t *get_node(struct dm_table *t,
99 unsigned int l, unsigned int n)
100{
101 return t->index[l] + (n * KEYS_PER_NODE);
102}
103
104/*
105 * Return the highest key that you could lookup from the n'th
106 * node on level l of the btree.
107 */
108static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
109{
110 for (; l < t->depth - 1; l++)
111 n = get_child(n, CHILDREN_PER_NODE - 1);
112
113 if (n >= t->counts[l])
114 return (sector_t) - 1;
115
116 return get_node(t, l, n)[KEYS_PER_NODE - 1];
117}
118
119/*
120 * Fills in a level of the btree based on the highs of the level
121 * below it.
122 */
123static int setup_btree_index(unsigned int l, struct dm_table *t)
124{
125 unsigned int n, k;
126 sector_t *node;
127
128 for (n = 0U; n < t->counts[l]; n++) {
129 node = get_node(t, l, n);
130
131 for (k = 0U; k < KEYS_PER_NODE; k++)
132 node[k] = high(t, l + 1, get_child(n, k));
133 }
134
135 return 0;
136}
137
138void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
139{
140 unsigned long size;
141 void *addr;
142
143 /*
144 * Check that we're not going to overflow.
145 */
146 if (nmemb > (ULONG_MAX / elem_size))
147 return NULL;
148
149 size = nmemb * elem_size;
150 addr = vzalloc(size);
151
152 return addr;
153}
154EXPORT_SYMBOL(dm_vcalloc);
155
156/*
157 * highs, and targets are managed as dynamic arrays during a
158 * table load.
159 */
160static int alloc_targets(struct dm_table *t, unsigned int num)
161{
162 sector_t *n_highs;
163 struct dm_target *n_targets;
164
165 /*
166 * Allocate both the target array and offset array at once.
167 * Append an empty entry to catch sectors beyond the end of
168 * the device.
169 */
170 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
171 sizeof(sector_t));
172 if (!n_highs)
173 return -ENOMEM;
174
175 n_targets = (struct dm_target *) (n_highs + num);
176
177 memset(n_highs, -1, sizeof(*n_highs) * num);
178 vfree(t->highs);
179
180 t->num_allocated = num;
181 t->highs = n_highs;
182 t->targets = n_targets;
183
184 return 0;
185}
186
187int dm_table_create(struct dm_table **result, fmode_t mode,
188 unsigned num_targets, struct mapped_device *md)
189{
190 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
191
192 if (!t)
193 return -ENOMEM;
194
195 INIT_LIST_HEAD(&t->devices);
196 INIT_LIST_HEAD(&t->target_callbacks);
197
198 if (!num_targets)
199 num_targets = KEYS_PER_NODE;
200
201 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
202
203 if (!num_targets) {
204 kfree(t);
205 return -ENOMEM;
206 }
207
208 if (alloc_targets(t, num_targets)) {
209 kfree(t);
210 return -ENOMEM;
211 }
212
213 t->type = DM_TYPE_NONE;
214 t->mode = mode;
215 t->md = md;
216 *result = t;
217 return 0;
218}
219
220static void free_devices(struct list_head *devices, struct mapped_device *md)
221{
222 struct list_head *tmp, *next;
223
224 list_for_each_safe(tmp, next, devices) {
225 struct dm_dev_internal *dd =
226 list_entry(tmp, struct dm_dev_internal, list);
227 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
228 dm_device_name(md), dd->dm_dev->name);
229 dm_put_table_device(md, dd->dm_dev);
230 kfree(dd);
231 }
232}
233
234void dm_table_destroy(struct dm_table *t)
235{
236 unsigned int i;
237
238 if (!t)
239 return;
240
241 /* free the indexes */
242 if (t->depth >= 2)
243 vfree(t->index[t->depth - 2]);
244
245 /* free the targets */
246 for (i = 0; i < t->num_targets; i++) {
247 struct dm_target *tgt = t->targets + i;
248
249 if (tgt->type->dtr)
250 tgt->type->dtr(tgt);
251
252 dm_put_target_type(tgt->type);
253 }
254
255 vfree(t->highs);
256
257 /* free the device list */
258 free_devices(&t->devices, t->md);
259
260 dm_free_md_mempools(t->mempools);
261
262 kfree(t);
263}
264
265/*
266 * See if we've already got a device in the list.
267 */
268static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
269{
270 struct dm_dev_internal *dd;
271
272 list_for_each_entry (dd, l, list)
273 if (dd->dm_dev->bdev->bd_dev == dev)
274 return dd;
275
276 return NULL;
277}
278
279/*
280 * If possible, this checks an area of a destination device is invalid.
281 */
282static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
283 sector_t start, sector_t len, void *data)
284{
285 struct request_queue *q;
286 struct queue_limits *limits = data;
287 struct block_device *bdev = dev->bdev;
288 sector_t dev_size =
289 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
290 unsigned short logical_block_size_sectors =
291 limits->logical_block_size >> SECTOR_SHIFT;
292 char b[BDEVNAME_SIZE];
293
294 /*
295 * Some devices exist without request functions,
296 * such as loop devices not yet bound to backing files.
297 * Forbid the use of such devices.
298 */
299 q = bdev_get_queue(bdev);
300 if (!q || !q->make_request_fn) {
301 DMWARN("%s: %s is not yet initialised: "
302 "start=%llu, len=%llu, dev_size=%llu",
303 dm_device_name(ti->table->md), bdevname(bdev, b),
304 (unsigned long long)start,
305 (unsigned long long)len,
306 (unsigned long long)dev_size);
307 return 1;
308 }
309
310 if (!dev_size)
311 return 0;
312
313 if ((start >= dev_size) || (start + len > dev_size)) {
314 DMWARN("%s: %s too small for target: "
315 "start=%llu, len=%llu, dev_size=%llu",
316 dm_device_name(ti->table->md), bdevname(bdev, b),
317 (unsigned long long)start,
318 (unsigned long long)len,
319 (unsigned long long)dev_size);
320 return 1;
321 }
322
323 /*
324 * If the target is mapped to zoned block device(s), check
325 * that the zones are not partially mapped.
326 */
327 if (bdev_zoned_model(bdev) != BLK_ZONED_NONE) {
328 unsigned int zone_sectors = bdev_zone_sectors(bdev);
329
330 if (start & (zone_sectors - 1)) {
331 DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s",
332 dm_device_name(ti->table->md),
333 (unsigned long long)start,
334 zone_sectors, bdevname(bdev, b));
335 return 1;
336 }
337
338 /*
339 * Note: The last zone of a zoned block device may be smaller
340 * than other zones. So for a target mapping the end of a
341 * zoned block device with such a zone, len would not be zone
342 * aligned. We do not allow such last smaller zone to be part
343 * of the mapping here to ensure that mappings with multiple
344 * devices do not end up with a smaller zone in the middle of
345 * the sector range.
346 */
347 if (len & (zone_sectors - 1)) {
348 DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s",
349 dm_device_name(ti->table->md),
350 (unsigned long long)len,
351 zone_sectors, bdevname(bdev, b));
352 return 1;
353 }
354 }
355
356 if (logical_block_size_sectors <= 1)
357 return 0;
358
359 if (start & (logical_block_size_sectors - 1)) {
360 DMWARN("%s: start=%llu not aligned to h/w "
361 "logical block size %u of %s",
362 dm_device_name(ti->table->md),
363 (unsigned long long)start,
364 limits->logical_block_size, bdevname(bdev, b));
365 return 1;
366 }
367
368 if (len & (logical_block_size_sectors - 1)) {
369 DMWARN("%s: len=%llu not aligned to h/w "
370 "logical block size %u of %s",
371 dm_device_name(ti->table->md),
372 (unsigned long long)len,
373 limits->logical_block_size, bdevname(bdev, b));
374 return 1;
375 }
376
377 return 0;
378}
379
380/*
381 * This upgrades the mode on an already open dm_dev, being
382 * careful to leave things as they were if we fail to reopen the
383 * device and not to touch the existing bdev field in case
384 * it is accessed concurrently inside dm_table_any_congested().
385 */
386static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
387 struct mapped_device *md)
388{
389 int r;
390 struct dm_dev *old_dev, *new_dev;
391
392 old_dev = dd->dm_dev;
393
394 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
395 dd->dm_dev->mode | new_mode, &new_dev);
396 if (r)
397 return r;
398
399 dd->dm_dev = new_dev;
400 dm_put_table_device(md, old_dev);
401
402 return 0;
403}
404
405/*
406 * Convert the path to a device
407 */
408dev_t dm_get_dev_t(const char *path)
409{
410 dev_t dev;
411 struct block_device *bdev;
412
413 bdev = lookup_bdev(path);
414 if (IS_ERR(bdev))
415 dev = name_to_dev_t(path);
416 else {
417 dev = bdev->bd_dev;
418 bdput(bdev);
419 }
420
421 return dev;
422}
423EXPORT_SYMBOL_GPL(dm_get_dev_t);
424
425/*
426 * Add a device to the list, or just increment the usage count if
427 * it's already present.
428 */
429int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
430 struct dm_dev **result)
431{
432 int r;
433 dev_t dev;
434 struct dm_dev_internal *dd;
435 struct dm_table *t = ti->table;
436
437 BUG_ON(!t);
438
439 dev = dm_get_dev_t(path);
440 if (!dev)
441 return -ENODEV;
442
443 dd = find_device(&t->devices, dev);
444 if (!dd) {
445 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
446 if (!dd)
447 return -ENOMEM;
448
449 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
450 kfree(dd);
451 return r;
452 }
453
454 refcount_set(&dd->count, 1);
455 list_add(&dd->list, &t->devices);
456 goto out;
457
458 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
459 r = upgrade_mode(dd, mode, t->md);
460 if (r)
461 return r;
462 }
463 refcount_inc(&dd->count);
464out:
465 *result = dd->dm_dev;
466 return 0;
467}
468EXPORT_SYMBOL(dm_get_device);
469
470static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
471 sector_t start, sector_t len, void *data)
472{
473 struct queue_limits *limits = data;
474 struct block_device *bdev = dev->bdev;
475 struct request_queue *q = bdev_get_queue(bdev);
476 char b[BDEVNAME_SIZE];
477
478 if (unlikely(!q)) {
479 DMWARN("%s: Cannot set limits for nonexistent device %s",
480 dm_device_name(ti->table->md), bdevname(bdev, b));
481 return 0;
482 }
483
484 if (bdev_stack_limits(limits, bdev, start) < 0)
485 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
486 "physical_block_size=%u, logical_block_size=%u, "
487 "alignment_offset=%u, start=%llu",
488 dm_device_name(ti->table->md), bdevname(bdev, b),
489 q->limits.physical_block_size,
490 q->limits.logical_block_size,
491 q->limits.alignment_offset,
492 (unsigned long long) start << SECTOR_SHIFT);
493
494 limits->zoned = blk_queue_zoned_model(q);
495
496 return 0;
497}
498
499/*
500 * Decrement a device's use count and remove it if necessary.
501 */
502void dm_put_device(struct dm_target *ti, struct dm_dev *d)
503{
504 int found = 0;
505 struct list_head *devices = &ti->table->devices;
506 struct dm_dev_internal *dd;
507
508 list_for_each_entry(dd, devices, list) {
509 if (dd->dm_dev == d) {
510 found = 1;
511 break;
512 }
513 }
514 if (!found) {
515 DMWARN("%s: device %s not in table devices list",
516 dm_device_name(ti->table->md), d->name);
517 return;
518 }
519 if (refcount_dec_and_test(&dd->count)) {
520 dm_put_table_device(ti->table->md, d);
521 list_del(&dd->list);
522 kfree(dd);
523 }
524}
525EXPORT_SYMBOL(dm_put_device);
526
527/*
528 * Checks to see if the target joins onto the end of the table.
529 */
530static int adjoin(struct dm_table *table, struct dm_target *ti)
531{
532 struct dm_target *prev;
533
534 if (!table->num_targets)
535 return !ti->begin;
536
537 prev = &table->targets[table->num_targets - 1];
538 return (ti->begin == (prev->begin + prev->len));
539}
540
541/*
542 * Used to dynamically allocate the arg array.
543 *
544 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
545 * process messages even if some device is suspended. These messages have a
546 * small fixed number of arguments.
547 *
548 * On the other hand, dm-switch needs to process bulk data using messages and
549 * excessive use of GFP_NOIO could cause trouble.
550 */
551static char **realloc_argv(unsigned *array_size, char **old_argv)
552{
553 char **argv;
554 unsigned new_size;
555 gfp_t gfp;
556
557 if (*array_size) {
558 new_size = *array_size * 2;
559 gfp = GFP_KERNEL;
560 } else {
561 new_size = 8;
562 gfp = GFP_NOIO;
563 }
564 argv = kmalloc(new_size * sizeof(*argv), gfp);
565 if (argv) {
566 memcpy(argv, old_argv, *array_size * sizeof(*argv));
567 *array_size = new_size;
568 }
569
570 kfree(old_argv);
571 return argv;
572}
573
574/*
575 * Destructively splits up the argument list to pass to ctr.
576 */
577int dm_split_args(int *argc, char ***argvp, char *input)
578{
579 char *start, *end = input, *out, **argv = NULL;
580 unsigned array_size = 0;
581
582 *argc = 0;
583
584 if (!input) {
585 *argvp = NULL;
586 return 0;
587 }
588
589 argv = realloc_argv(&array_size, argv);
590 if (!argv)
591 return -ENOMEM;
592
593 while (1) {
594 /* Skip whitespace */
595 start = skip_spaces(end);
596
597 if (!*start)
598 break; /* success, we hit the end */
599
600 /* 'out' is used to remove any back-quotes */
601 end = out = start;
602 while (*end) {
603 /* Everything apart from '\0' can be quoted */
604 if (*end == '\\' && *(end + 1)) {
605 *out++ = *(end + 1);
606 end += 2;
607 continue;
608 }
609
610 if (isspace(*end))
611 break; /* end of token */
612
613 *out++ = *end++;
614 }
615
616 /* have we already filled the array ? */
617 if ((*argc + 1) > array_size) {
618 argv = realloc_argv(&array_size, argv);
619 if (!argv)
620 return -ENOMEM;
621 }
622
623 /* we know this is whitespace */
624 if (*end)
625 end++;
626
627 /* terminate the string and put it in the array */
628 *out = '\0';
629 argv[*argc] = start;
630 (*argc)++;
631 }
632
633 *argvp = argv;
634 return 0;
635}
636
637/*
638 * Impose necessary and sufficient conditions on a devices's table such
639 * that any incoming bio which respects its logical_block_size can be
640 * processed successfully. If it falls across the boundary between
641 * two or more targets, the size of each piece it gets split into must
642 * be compatible with the logical_block_size of the target processing it.
643 */
644static int validate_hardware_logical_block_alignment(struct dm_table *table,
645 struct queue_limits *limits)
646{
647 /*
648 * This function uses arithmetic modulo the logical_block_size
649 * (in units of 512-byte sectors).
650 */
651 unsigned short device_logical_block_size_sects =
652 limits->logical_block_size >> SECTOR_SHIFT;
653
654 /*
655 * Offset of the start of the next table entry, mod logical_block_size.
656 */
657 unsigned short next_target_start = 0;
658
659 /*
660 * Given an aligned bio that extends beyond the end of a
661 * target, how many sectors must the next target handle?
662 */
663 unsigned short remaining = 0;
664
665 struct dm_target *uninitialized_var(ti);
666 struct queue_limits ti_limits;
667 unsigned i;
668
669 /*
670 * Check each entry in the table in turn.
671 */
672 for (i = 0; i < dm_table_get_num_targets(table); i++) {
673 ti = dm_table_get_target(table, i);
674
675 blk_set_stacking_limits(&ti_limits);
676
677 /* combine all target devices' limits */
678 if (ti->type->iterate_devices)
679 ti->type->iterate_devices(ti, dm_set_device_limits,
680 &ti_limits);
681
682 /*
683 * If the remaining sectors fall entirely within this
684 * table entry are they compatible with its logical_block_size?
685 */
686 if (remaining < ti->len &&
687 remaining & ((ti_limits.logical_block_size >>
688 SECTOR_SHIFT) - 1))
689 break; /* Error */
690
691 next_target_start =
692 (unsigned short) ((next_target_start + ti->len) &
693 (device_logical_block_size_sects - 1));
694 remaining = next_target_start ?
695 device_logical_block_size_sects - next_target_start : 0;
696 }
697
698 if (remaining) {
699 DMWARN("%s: table line %u (start sect %llu len %llu) "
700 "not aligned to h/w logical block size %u",
701 dm_device_name(table->md), i,
702 (unsigned long long) ti->begin,
703 (unsigned long long) ti->len,
704 limits->logical_block_size);
705 return -EINVAL;
706 }
707
708 return 0;
709}
710
711int dm_table_add_target(struct dm_table *t, const char *type,
712 sector_t start, sector_t len, char *params)
713{
714 int r = -EINVAL, argc;
715 char **argv;
716 struct dm_target *tgt;
717
718 if (t->singleton) {
719 DMERR("%s: target type %s must appear alone in table",
720 dm_device_name(t->md), t->targets->type->name);
721 return -EINVAL;
722 }
723
724 BUG_ON(t->num_targets >= t->num_allocated);
725
726 tgt = t->targets + t->num_targets;
727 memset(tgt, 0, sizeof(*tgt));
728
729 if (!len) {
730 DMERR("%s: zero-length target", dm_device_name(t->md));
731 return -EINVAL;
732 }
733
734 tgt->type = dm_get_target_type(type);
735 if (!tgt->type) {
736 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
737 return -EINVAL;
738 }
739
740 if (dm_target_needs_singleton(tgt->type)) {
741 if (t->num_targets) {
742 tgt->error = "singleton target type must appear alone in table";
743 goto bad;
744 }
745 t->singleton = true;
746 }
747
748 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
749 tgt->error = "target type may not be included in a read-only table";
750 goto bad;
751 }
752
753 if (t->immutable_target_type) {
754 if (t->immutable_target_type != tgt->type) {
755 tgt->error = "immutable target type cannot be mixed with other target types";
756 goto bad;
757 }
758 } else if (dm_target_is_immutable(tgt->type)) {
759 if (t->num_targets) {
760 tgt->error = "immutable target type cannot be mixed with other target types";
761 goto bad;
762 }
763 t->immutable_target_type = tgt->type;
764 }
765
766 if (dm_target_has_integrity(tgt->type))
767 t->integrity_added = 1;
768
769 tgt->table = t;
770 tgt->begin = start;
771 tgt->len = len;
772 tgt->error = "Unknown error";
773
774 /*
775 * Does this target adjoin the previous one ?
776 */
777 if (!adjoin(t, tgt)) {
778 tgt->error = "Gap in table";
779 goto bad;
780 }
781
782 r = dm_split_args(&argc, &argv, params);
783 if (r) {
784 tgt->error = "couldn't split parameters (insufficient memory)";
785 goto bad;
786 }
787
788 r = tgt->type->ctr(tgt, argc, argv);
789 kfree(argv);
790 if (r)
791 goto bad;
792
793 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
794
795 if (!tgt->num_discard_bios && tgt->discards_supported)
796 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
797 dm_device_name(t->md), type);
798
799 return 0;
800
801 bad:
802 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
803 dm_put_target_type(tgt->type);
804 return r;
805}
806
807/*
808 * Target argument parsing helpers.
809 */
810static int validate_next_arg(const struct dm_arg *arg,
811 struct dm_arg_set *arg_set,
812 unsigned *value, char **error, unsigned grouped)
813{
814 const char *arg_str = dm_shift_arg(arg_set);
815 char dummy;
816
817 if (!arg_str ||
818 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
819 (*value < arg->min) ||
820 (*value > arg->max) ||
821 (grouped && arg_set->argc < *value)) {
822 *error = arg->error;
823 return -EINVAL;
824 }
825
826 return 0;
827}
828
829int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
830 unsigned *value, char **error)
831{
832 return validate_next_arg(arg, arg_set, value, error, 0);
833}
834EXPORT_SYMBOL(dm_read_arg);
835
836int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
837 unsigned *value, char **error)
838{
839 return validate_next_arg(arg, arg_set, value, error, 1);
840}
841EXPORT_SYMBOL(dm_read_arg_group);
842
843const char *dm_shift_arg(struct dm_arg_set *as)
844{
845 char *r;
846
847 if (as->argc) {
848 as->argc--;
849 r = *as->argv;
850 as->argv++;
851 return r;
852 }
853
854 return NULL;
855}
856EXPORT_SYMBOL(dm_shift_arg);
857
858void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
859{
860 BUG_ON(as->argc < num_args);
861 as->argc -= num_args;
862 as->argv += num_args;
863}
864EXPORT_SYMBOL(dm_consume_args);
865
866static bool __table_type_bio_based(enum dm_queue_mode table_type)
867{
868 return (table_type == DM_TYPE_BIO_BASED ||
869 table_type == DM_TYPE_DAX_BIO_BASED ||
870 table_type == DM_TYPE_NVME_BIO_BASED);
871}
872
873static bool __table_type_request_based(enum dm_queue_mode table_type)
874{
875 return (table_type == DM_TYPE_REQUEST_BASED ||
876 table_type == DM_TYPE_MQ_REQUEST_BASED);
877}
878
879void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
880{
881 t->type = type;
882}
883EXPORT_SYMBOL_GPL(dm_table_set_type);
884
885static int device_supports_dax(struct dm_target *ti, struct dm_dev *dev,
886 sector_t start, sector_t len, void *data)
887{
888 struct request_queue *q = bdev_get_queue(dev->bdev);
889
890 return q && blk_queue_dax(q);
891}
892
893static bool dm_table_supports_dax(struct dm_table *t)
894{
895 struct dm_target *ti;
896 unsigned i;
897
898 /* Ensure that all targets support DAX. */
899 for (i = 0; i < dm_table_get_num_targets(t); i++) {
900 ti = dm_table_get_target(t, i);
901
902 if (!ti->type->direct_access)
903 return false;
904
905 if (!ti->type->iterate_devices ||
906 !ti->type->iterate_devices(ti, device_supports_dax, NULL))
907 return false;
908 }
909
910 return true;
911}
912
913static bool dm_table_does_not_support_partial_completion(struct dm_table *t);
914
915struct verify_rq_based_data {
916 unsigned sq_count;
917 unsigned mq_count;
918};
919
920static int device_is_rq_based(struct dm_target *ti, struct dm_dev *dev,
921 sector_t start, sector_t len, void *data)
922{
923 struct request_queue *q = bdev_get_queue(dev->bdev);
924 struct verify_rq_based_data *v = data;
925
926 if (q->mq_ops)
927 v->mq_count++;
928 else
929 v->sq_count++;
930
931 return queue_is_rq_based(q);
932}
933
934static int dm_table_determine_type(struct dm_table *t)
935{
936 unsigned i;
937 unsigned bio_based = 0, request_based = 0, hybrid = 0;
938 struct verify_rq_based_data v = {.sq_count = 0, .mq_count = 0};
939 struct dm_target *tgt;
940 struct list_head *devices = dm_table_get_devices(t);
941 enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
942
943 if (t->type != DM_TYPE_NONE) {
944 /* target already set the table's type */
945 if (t->type == DM_TYPE_BIO_BASED) {
946 /* possibly upgrade to a variant of bio-based */
947 goto verify_bio_based;
948 }
949 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
950 BUG_ON(t->type == DM_TYPE_NVME_BIO_BASED);
951 goto verify_rq_based;
952 }
953
954 for (i = 0; i < t->num_targets; i++) {
955 tgt = t->targets + i;
956 if (dm_target_hybrid(tgt))
957 hybrid = 1;
958 else if (dm_target_request_based(tgt))
959 request_based = 1;
960 else
961 bio_based = 1;
962
963 if (bio_based && request_based) {
964 DMERR("Inconsistent table: different target types"
965 " can't be mixed up");
966 return -EINVAL;
967 }
968 }
969
970 if (hybrid && !bio_based && !request_based) {
971 /*
972 * The targets can work either way.
973 * Determine the type from the live device.
974 * Default to bio-based if device is new.
975 */
976 if (__table_type_request_based(live_md_type))
977 request_based = 1;
978 else
979 bio_based = 1;
980 }
981
982 if (bio_based) {
983verify_bio_based:
984 /* We must use this table as bio-based */
985 t->type = DM_TYPE_BIO_BASED;
986 if (dm_table_supports_dax(t) ||
987 (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
988 t->type = DM_TYPE_DAX_BIO_BASED;
989 } else {
990 /* Check if upgrading to NVMe bio-based is valid or required */
991 tgt = dm_table_get_immutable_target(t);
992 if (tgt && !tgt->max_io_len && dm_table_does_not_support_partial_completion(t)) {
993 t->type = DM_TYPE_NVME_BIO_BASED;
994 goto verify_rq_based; /* must be stacked directly on NVMe (blk-mq) */
995 } else if (list_empty(devices) && live_md_type == DM_TYPE_NVME_BIO_BASED) {
996 t->type = DM_TYPE_NVME_BIO_BASED;
997 }
998 }
999 return 0;
1000 }
1001
1002 BUG_ON(!request_based); /* No targets in this table */
1003
1004 /*
1005 * The only way to establish DM_TYPE_MQ_REQUEST_BASED is by
1006 * having a compatible target use dm_table_set_type.
1007 */
1008 t->type = DM_TYPE_REQUEST_BASED;
1009
1010verify_rq_based:
1011 /*
1012 * Request-based dm supports only tables that have a single target now.
1013 * To support multiple targets, request splitting support is needed,
1014 * and that needs lots of changes in the block-layer.
1015 * (e.g. request completion process for partial completion.)
1016 */
1017 if (t->num_targets > 1) {
1018 DMERR("%s DM doesn't support multiple targets",
1019 t->type == DM_TYPE_NVME_BIO_BASED ? "nvme bio-based" : "request-based");
1020 return -EINVAL;
1021 }
1022
1023 if (list_empty(devices)) {
1024 int srcu_idx;
1025 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
1026
1027 /* inherit live table's type and all_blk_mq */
1028 if (live_table) {
1029 t->type = live_table->type;
1030 t->all_blk_mq = live_table->all_blk_mq;
1031 }
1032 dm_put_live_table(t->md, srcu_idx);
1033 return 0;
1034 }
1035
1036 tgt = dm_table_get_immutable_target(t);
1037 if (!tgt) {
1038 DMERR("table load rejected: immutable target is required");
1039 return -EINVAL;
1040 } else if (tgt->max_io_len) {
1041 DMERR("table load rejected: immutable target that splits IO is not supported");
1042 return -EINVAL;
1043 }
1044
1045 /* Non-request-stackable devices can't be used for request-based dm */
1046 if (!tgt->type->iterate_devices ||
1047 !tgt->type->iterate_devices(tgt, device_is_rq_based, &v)) {
1048 DMERR("table load rejected: including non-request-stackable devices");
1049 return -EINVAL;
1050 }
1051 if (v.sq_count && v.mq_count) {
1052 DMERR("table load rejected: not all devices are blk-mq request-stackable");
1053 return -EINVAL;
1054 }
1055 t->all_blk_mq = v.mq_count > 0;
1056
1057 if (!t->all_blk_mq &&
1058 (t->type == DM_TYPE_MQ_REQUEST_BASED || t->type == DM_TYPE_NVME_BIO_BASED)) {
1059 DMERR("table load rejected: all devices are not blk-mq request-stackable");
1060 return -EINVAL;
1061 }
1062
1063 return 0;
1064}
1065
1066enum dm_queue_mode dm_table_get_type(struct dm_table *t)
1067{
1068 return t->type;
1069}
1070
1071struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
1072{
1073 return t->immutable_target_type;
1074}
1075
1076struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
1077{
1078 /* Immutable target is implicitly a singleton */
1079 if (t->num_targets > 1 ||
1080 !dm_target_is_immutable(t->targets[0].type))
1081 return NULL;
1082
1083 return t->targets;
1084}
1085
1086struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1087{
1088 struct dm_target *ti;
1089 unsigned i;
1090
1091 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1092 ti = dm_table_get_target(t, i);
1093 if (dm_target_is_wildcard(ti->type))
1094 return ti;
1095 }
1096
1097 return NULL;
1098}
1099
1100bool dm_table_bio_based(struct dm_table *t)
1101{
1102 return __table_type_bio_based(dm_table_get_type(t));
1103}
1104
1105bool dm_table_request_based(struct dm_table *t)
1106{
1107 return __table_type_request_based(dm_table_get_type(t));
1108}
1109
1110bool dm_table_all_blk_mq_devices(struct dm_table *t)
1111{
1112 return t->all_blk_mq;
1113}
1114
1115static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1116{
1117 enum dm_queue_mode type = dm_table_get_type(t);
1118 unsigned per_io_data_size = 0;
1119 unsigned min_pool_size = 0;
1120 struct dm_target *ti;
1121 unsigned i;
1122
1123 if (unlikely(type == DM_TYPE_NONE)) {
1124 DMWARN("no table type is set, can't allocate mempools");
1125 return -EINVAL;
1126 }
1127
1128 if (__table_type_bio_based(type))
1129 for (i = 0; i < t->num_targets; i++) {
1130 ti = t->targets + i;
1131 per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1132 min_pool_size = max(min_pool_size, ti->num_flush_bios);
1133 }
1134
1135 t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1136 per_io_data_size, min_pool_size);
1137 if (!t->mempools)
1138 return -ENOMEM;
1139
1140 return 0;
1141}
1142
1143void dm_table_free_md_mempools(struct dm_table *t)
1144{
1145 dm_free_md_mempools(t->mempools);
1146 t->mempools = NULL;
1147}
1148
1149struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1150{
1151 return t->mempools;
1152}
1153
1154static int setup_indexes(struct dm_table *t)
1155{
1156 int i;
1157 unsigned int total = 0;
1158 sector_t *indexes;
1159
1160 /* allocate the space for *all* the indexes */
1161 for (i = t->depth - 2; i >= 0; i--) {
1162 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1163 total += t->counts[i];
1164 }
1165
1166 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1167 if (!indexes)
1168 return -ENOMEM;
1169
1170 /* set up internal nodes, bottom-up */
1171 for (i = t->depth - 2; i >= 0; i--) {
1172 t->index[i] = indexes;
1173 indexes += (KEYS_PER_NODE * t->counts[i]);
1174 setup_btree_index(i, t);
1175 }
1176
1177 return 0;
1178}
1179
1180/*
1181 * Builds the btree to index the map.
1182 */
1183static int dm_table_build_index(struct dm_table *t)
1184{
1185 int r = 0;
1186 unsigned int leaf_nodes;
1187
1188 /* how many indexes will the btree have ? */
1189 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1190 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1191
1192 /* leaf layer has already been set up */
1193 t->counts[t->depth - 1] = leaf_nodes;
1194 t->index[t->depth - 1] = t->highs;
1195
1196 if (t->depth >= 2)
1197 r = setup_indexes(t);
1198
1199 return r;
1200}
1201
1202static bool integrity_profile_exists(struct gendisk *disk)
1203{
1204 return !!blk_get_integrity(disk);
1205}
1206
1207/*
1208 * Get a disk whose integrity profile reflects the table's profile.
1209 * Returns NULL if integrity support was inconsistent or unavailable.
1210 */
1211static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1212{
1213 struct list_head *devices = dm_table_get_devices(t);
1214 struct dm_dev_internal *dd = NULL;
1215 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1216 unsigned i;
1217
1218 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1219 struct dm_target *ti = dm_table_get_target(t, i);
1220 if (!dm_target_passes_integrity(ti->type))
1221 goto no_integrity;
1222 }
1223
1224 list_for_each_entry(dd, devices, list) {
1225 template_disk = dd->dm_dev->bdev->bd_disk;
1226 if (!integrity_profile_exists(template_disk))
1227 goto no_integrity;
1228 else if (prev_disk &&
1229 blk_integrity_compare(prev_disk, template_disk) < 0)
1230 goto no_integrity;
1231 prev_disk = template_disk;
1232 }
1233
1234 return template_disk;
1235
1236no_integrity:
1237 if (prev_disk)
1238 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1239 dm_device_name(t->md),
1240 prev_disk->disk_name,
1241 template_disk->disk_name);
1242 return NULL;
1243}
1244
1245/*
1246 * Register the mapped device for blk_integrity support if the
1247 * underlying devices have an integrity profile. But all devices may
1248 * not have matching profiles (checking all devices isn't reliable
1249 * during table load because this table may use other DM device(s) which
1250 * must be resumed before they will have an initialized integity
1251 * profile). Consequently, stacked DM devices force a 2 stage integrity
1252 * profile validation: First pass during table load, final pass during
1253 * resume.
1254 */
1255static int dm_table_register_integrity(struct dm_table *t)
1256{
1257 struct mapped_device *md = t->md;
1258 struct gendisk *template_disk = NULL;
1259
1260 /* If target handles integrity itself do not register it here. */
1261 if (t->integrity_added)
1262 return 0;
1263
1264 template_disk = dm_table_get_integrity_disk(t);
1265 if (!template_disk)
1266 return 0;
1267
1268 if (!integrity_profile_exists(dm_disk(md))) {
1269 t->integrity_supported = true;
1270 /*
1271 * Register integrity profile during table load; we can do
1272 * this because the final profile must match during resume.
1273 */
1274 blk_integrity_register(dm_disk(md),
1275 blk_get_integrity(template_disk));
1276 return 0;
1277 }
1278
1279 /*
1280 * If DM device already has an initialized integrity
1281 * profile the new profile should not conflict.
1282 */
1283 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1284 DMWARN("%s: conflict with existing integrity profile: "
1285 "%s profile mismatch",
1286 dm_device_name(t->md),
1287 template_disk->disk_name);
1288 return 1;
1289 }
1290
1291 /* Preserve existing integrity profile */
1292 t->integrity_supported = true;
1293 return 0;
1294}
1295
1296/*
1297 * Prepares the table for use by building the indices,
1298 * setting the type, and allocating mempools.
1299 */
1300int dm_table_complete(struct dm_table *t)
1301{
1302 int r;
1303
1304 r = dm_table_determine_type(t);
1305 if (r) {
1306 DMERR("unable to determine table type");
1307 return r;
1308 }
1309
1310 r = dm_table_build_index(t);
1311 if (r) {
1312 DMERR("unable to build btrees");
1313 return r;
1314 }
1315
1316 r = dm_table_register_integrity(t);
1317 if (r) {
1318 DMERR("could not register integrity profile.");
1319 return r;
1320 }
1321
1322 r = dm_table_alloc_md_mempools(t, t->md);
1323 if (r)
1324 DMERR("unable to allocate mempools");
1325
1326 return r;
1327}
1328
1329static DEFINE_MUTEX(_event_lock);
1330void dm_table_event_callback(struct dm_table *t,
1331 void (*fn)(void *), void *context)
1332{
1333 mutex_lock(&_event_lock);
1334 t->event_fn = fn;
1335 t->event_context = context;
1336 mutex_unlock(&_event_lock);
1337}
1338
1339void dm_table_event(struct dm_table *t)
1340{
1341 /*
1342 * You can no longer call dm_table_event() from interrupt
1343 * context, use a bottom half instead.
1344 */
1345 BUG_ON(in_interrupt());
1346
1347 mutex_lock(&_event_lock);
1348 if (t->event_fn)
1349 t->event_fn(t->event_context);
1350 mutex_unlock(&_event_lock);
1351}
1352EXPORT_SYMBOL(dm_table_event);
1353
1354sector_t dm_table_get_size(struct dm_table *t)
1355{
1356 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1357}
1358EXPORT_SYMBOL(dm_table_get_size);
1359
1360struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1361{
1362 if (index >= t->num_targets)
1363 return NULL;
1364
1365 return t->targets + index;
1366}
1367
1368/*
1369 * Search the btree for the correct target.
1370 *
1371 * Caller should check returned pointer with dm_target_is_valid()
1372 * to trap I/O beyond end of device.
1373 */
1374struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1375{
1376 unsigned int l, n = 0, k = 0;
1377 sector_t *node;
1378
1379 for (l = 0; l < t->depth; l++) {
1380 n = get_child(n, k);
1381 node = get_node(t, l, n);
1382
1383 for (k = 0; k < KEYS_PER_NODE; k++)
1384 if (node[k] >= sector)
1385 break;
1386 }
1387
1388 return &t->targets[(KEYS_PER_NODE * n) + k];
1389}
1390
1391static int count_device(struct dm_target *ti, struct dm_dev *dev,
1392 sector_t start, sector_t len, void *data)
1393{
1394 unsigned *num_devices = data;
1395
1396 (*num_devices)++;
1397
1398 return 0;
1399}
1400
1401/*
1402 * Check whether a table has no data devices attached using each
1403 * target's iterate_devices method.
1404 * Returns false if the result is unknown because a target doesn't
1405 * support iterate_devices.
1406 */
1407bool dm_table_has_no_data_devices(struct dm_table *table)
1408{
1409 struct dm_target *ti;
1410 unsigned i, num_devices;
1411
1412 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1413 ti = dm_table_get_target(table, i);
1414
1415 if (!ti->type->iterate_devices)
1416 return false;
1417
1418 num_devices = 0;
1419 ti->type->iterate_devices(ti, count_device, &num_devices);
1420 if (num_devices)
1421 return false;
1422 }
1423
1424 return true;
1425}
1426
1427static int device_is_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1428 sector_t start, sector_t len, void *data)
1429{
1430 struct request_queue *q = bdev_get_queue(dev->bdev);
1431 enum blk_zoned_model *zoned_model = data;
1432
1433 return q && blk_queue_zoned_model(q) == *zoned_model;
1434}
1435
1436static bool dm_table_supports_zoned_model(struct dm_table *t,
1437 enum blk_zoned_model zoned_model)
1438{
1439 struct dm_target *ti;
1440 unsigned i;
1441
1442 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1443 ti = dm_table_get_target(t, i);
1444
1445 if (zoned_model == BLK_ZONED_HM &&
1446 !dm_target_supports_zoned_hm(ti->type))
1447 return false;
1448
1449 if (!ti->type->iterate_devices ||
1450 !ti->type->iterate_devices(ti, device_is_zoned_model, &zoned_model))
1451 return false;
1452 }
1453
1454 return true;
1455}
1456
1457static int device_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1458 sector_t start, sector_t len, void *data)
1459{
1460 struct request_queue *q = bdev_get_queue(dev->bdev);
1461 unsigned int *zone_sectors = data;
1462
1463 return q && blk_queue_zone_sectors(q) == *zone_sectors;
1464}
1465
1466static bool dm_table_matches_zone_sectors(struct dm_table *t,
1467 unsigned int zone_sectors)
1468{
1469 struct dm_target *ti;
1470 unsigned i;
1471
1472 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1473 ti = dm_table_get_target(t, i);
1474
1475 if (!ti->type->iterate_devices ||
1476 !ti->type->iterate_devices(ti, device_matches_zone_sectors, &zone_sectors))
1477 return false;
1478 }
1479
1480 return true;
1481}
1482
1483static int validate_hardware_zoned_model(struct dm_table *table,
1484 enum blk_zoned_model zoned_model,
1485 unsigned int zone_sectors)
1486{
1487 if (zoned_model == BLK_ZONED_NONE)
1488 return 0;
1489
1490 if (!dm_table_supports_zoned_model(table, zoned_model)) {
1491 DMERR("%s: zoned model is not consistent across all devices",
1492 dm_device_name(table->md));
1493 return -EINVAL;
1494 }
1495
1496 /* Check zone size validity and compatibility */
1497 if (!zone_sectors || !is_power_of_2(zone_sectors))
1498 return -EINVAL;
1499
1500 if (!dm_table_matches_zone_sectors(table, zone_sectors)) {
1501 DMERR("%s: zone sectors is not consistent across all devices",
1502 dm_device_name(table->md));
1503 return -EINVAL;
1504 }
1505
1506 return 0;
1507}
1508
1509/*
1510 * Establish the new table's queue_limits and validate them.
1511 */
1512int dm_calculate_queue_limits(struct dm_table *table,
1513 struct queue_limits *limits)
1514{
1515 struct dm_target *ti;
1516 struct queue_limits ti_limits;
1517 unsigned i;
1518 enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1519 unsigned int zone_sectors = 0;
1520
1521 blk_set_stacking_limits(limits);
1522
1523 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1524 blk_set_stacking_limits(&ti_limits);
1525
1526 ti = dm_table_get_target(table, i);
1527
1528 if (!ti->type->iterate_devices)
1529 goto combine_limits;
1530
1531 /*
1532 * Combine queue limits of all the devices this target uses.
1533 */
1534 ti->type->iterate_devices(ti, dm_set_device_limits,
1535 &ti_limits);
1536
1537 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1538 /*
1539 * After stacking all limits, validate all devices
1540 * in table support this zoned model and zone sectors.
1541 */
1542 zoned_model = ti_limits.zoned;
1543 zone_sectors = ti_limits.chunk_sectors;
1544 }
1545
1546 /* Set I/O hints portion of queue limits */
1547 if (ti->type->io_hints)
1548 ti->type->io_hints(ti, &ti_limits);
1549
1550 /*
1551 * Check each device area is consistent with the target's
1552 * overall queue limits.
1553 */
1554 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1555 &ti_limits))
1556 return -EINVAL;
1557
1558combine_limits:
1559 /*
1560 * Merge this target's queue limits into the overall limits
1561 * for the table.
1562 */
1563 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1564 DMWARN("%s: adding target device "
1565 "(start sect %llu len %llu) "
1566 "caused an alignment inconsistency",
1567 dm_device_name(table->md),
1568 (unsigned long long) ti->begin,
1569 (unsigned long long) ti->len);
1570
1571 /*
1572 * FIXME: this should likely be moved to blk_stack_limits(), would
1573 * also eliminate limits->zoned stacking hack in dm_set_device_limits()
1574 */
1575 if (limits->zoned == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1576 /*
1577 * By default, the stacked limits zoned model is set to
1578 * BLK_ZONED_NONE in blk_set_stacking_limits(). Update
1579 * this model using the first target model reported
1580 * that is not BLK_ZONED_NONE. This will be either the
1581 * first target device zoned model or the model reported
1582 * by the target .io_hints.
1583 */
1584 limits->zoned = ti_limits.zoned;
1585 }
1586 }
1587
1588 /*
1589 * Verify that the zoned model and zone sectors, as determined before
1590 * any .io_hints override, are the same across all devices in the table.
1591 * - this is especially relevant if .io_hints is emulating a disk-managed
1592 * zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1593 * BUT...
1594 */
1595 if (limits->zoned != BLK_ZONED_NONE) {
1596 /*
1597 * ...IF the above limits stacking determined a zoned model
1598 * validate that all of the table's devices conform to it.
1599 */
1600 zoned_model = limits->zoned;
1601 zone_sectors = limits->chunk_sectors;
1602 }
1603 if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1604 return -EINVAL;
1605
1606 return validate_hardware_logical_block_alignment(table, limits);
1607}
1608
1609/*
1610 * Verify that all devices have an integrity profile that matches the
1611 * DM device's registered integrity profile. If the profiles don't
1612 * match then unregister the DM device's integrity profile.
1613 */
1614static void dm_table_verify_integrity(struct dm_table *t)
1615{
1616 struct gendisk *template_disk = NULL;
1617
1618 if (t->integrity_added)
1619 return;
1620
1621 if (t->integrity_supported) {
1622 /*
1623 * Verify that the original integrity profile
1624 * matches all the devices in this table.
1625 */
1626 template_disk = dm_table_get_integrity_disk(t);
1627 if (template_disk &&
1628 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1629 return;
1630 }
1631
1632 if (integrity_profile_exists(dm_disk(t->md))) {
1633 DMWARN("%s: unable to establish an integrity profile",
1634 dm_device_name(t->md));
1635 blk_integrity_unregister(dm_disk(t->md));
1636 }
1637}
1638
1639static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1640 sector_t start, sector_t len, void *data)
1641{
1642 unsigned long flush = (unsigned long) data;
1643 struct request_queue *q = bdev_get_queue(dev->bdev);
1644
1645 return q && (q->queue_flags & flush);
1646}
1647
1648static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1649{
1650 struct dm_target *ti;
1651 unsigned i;
1652
1653 /*
1654 * Require at least one underlying device to support flushes.
1655 * t->devices includes internal dm devices such as mirror logs
1656 * so we need to use iterate_devices here, which targets
1657 * supporting flushes must provide.
1658 */
1659 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1660 ti = dm_table_get_target(t, i);
1661
1662 if (!ti->num_flush_bios)
1663 continue;
1664
1665 if (ti->flush_supported)
1666 return true;
1667
1668 if (ti->type->iterate_devices &&
1669 ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1670 return true;
1671 }
1672
1673 return false;
1674}
1675
1676static int device_dax_write_cache_enabled(struct dm_target *ti,
1677 struct dm_dev *dev, sector_t start,
1678 sector_t len, void *data)
1679{
1680 struct dax_device *dax_dev = dev->dax_dev;
1681
1682 if (!dax_dev)
1683 return false;
1684
1685 if (dax_write_cache_enabled(dax_dev))
1686 return true;
1687 return false;
1688}
1689
1690static int dm_table_supports_dax_write_cache(struct dm_table *t)
1691{
1692 struct dm_target *ti;
1693 unsigned i;
1694
1695 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1696 ti = dm_table_get_target(t, i);
1697
1698 if (ti->type->iterate_devices &&
1699 ti->type->iterate_devices(ti,
1700 device_dax_write_cache_enabled, NULL))
1701 return true;
1702 }
1703
1704 return false;
1705}
1706
1707static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1708 sector_t start, sector_t len, void *data)
1709{
1710 struct request_queue *q = bdev_get_queue(dev->bdev);
1711
1712 return q && blk_queue_nonrot(q);
1713}
1714
1715static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1716 sector_t start, sector_t len, void *data)
1717{
1718 struct request_queue *q = bdev_get_queue(dev->bdev);
1719
1720 return q && !blk_queue_add_random(q);
1721}
1722
1723static int queue_supports_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1724 sector_t start, sector_t len, void *data)
1725{
1726 struct request_queue *q = bdev_get_queue(dev->bdev);
1727
1728 return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1729}
1730
1731static bool dm_table_all_devices_attribute(struct dm_table *t,
1732 iterate_devices_callout_fn func)
1733{
1734 struct dm_target *ti;
1735 unsigned i;
1736
1737 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1738 ti = dm_table_get_target(t, i);
1739
1740 if (!ti->type->iterate_devices ||
1741 !ti->type->iterate_devices(ti, func, NULL))
1742 return false;
1743 }
1744
1745 return true;
1746}
1747
1748static int device_no_partial_completion(struct dm_target *ti, struct dm_dev *dev,
1749 sector_t start, sector_t len, void *data)
1750{
1751 char b[BDEVNAME_SIZE];
1752
1753 /* For now, NVMe devices are the only devices of this class */
1754 return (strncmp(bdevname(dev->bdev, b), "nvme", 4) == 0);
1755}
1756
1757static bool dm_table_does_not_support_partial_completion(struct dm_table *t)
1758{
1759 return dm_table_all_devices_attribute(t, device_no_partial_completion);
1760}
1761
1762static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1763 sector_t start, sector_t len, void *data)
1764{
1765 struct request_queue *q = bdev_get_queue(dev->bdev);
1766
1767 return q && !q->limits.max_write_same_sectors;
1768}
1769
1770static bool dm_table_supports_write_same(struct dm_table *t)
1771{
1772 struct dm_target *ti;
1773 unsigned i;
1774
1775 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1776 ti = dm_table_get_target(t, i);
1777
1778 if (!ti->num_write_same_bios)
1779 return false;
1780
1781 if (!ti->type->iterate_devices ||
1782 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1783 return false;
1784 }
1785
1786 return true;
1787}
1788
1789static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1790 sector_t start, sector_t len, void *data)
1791{
1792 struct request_queue *q = bdev_get_queue(dev->bdev);
1793
1794 return q && !q->limits.max_write_zeroes_sectors;
1795}
1796
1797static bool dm_table_supports_write_zeroes(struct dm_table *t)
1798{
1799 struct dm_target *ti;
1800 unsigned i = 0;
1801
1802 while (i < dm_table_get_num_targets(t)) {
1803 ti = dm_table_get_target(t, i++);
1804
1805 if (!ti->num_write_zeroes_bios)
1806 return false;
1807
1808 if (!ti->type->iterate_devices ||
1809 ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1810 return false;
1811 }
1812
1813 return true;
1814}
1815
1816static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1817 sector_t start, sector_t len, void *data)
1818{
1819 struct request_queue *q = bdev_get_queue(dev->bdev);
1820
1821 return q && !blk_queue_discard(q);
1822}
1823
1824static bool dm_table_supports_discards(struct dm_table *t)
1825{
1826 struct dm_target *ti;
1827 unsigned i;
1828
1829 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1830 ti = dm_table_get_target(t, i);
1831
1832 if (!ti->num_discard_bios)
1833 return false;
1834
1835 /*
1836 * Either the target provides discard support (as implied by setting
1837 * 'discards_supported') or it relies on _all_ data devices having
1838 * discard support.
1839 */
1840 if (!ti->discards_supported &&
1841 (!ti->type->iterate_devices ||
1842 ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1843 return false;
1844 }
1845
1846 return true;
1847}
1848
1849static int device_not_secure_erase_capable(struct dm_target *ti,
1850 struct dm_dev *dev, sector_t start,
1851 sector_t len, void *data)
1852{
1853 struct request_queue *q = bdev_get_queue(dev->bdev);
1854
1855 return q && !blk_queue_secure_erase(q);
1856}
1857
1858static bool dm_table_supports_secure_erase(struct dm_table *t)
1859{
1860 struct dm_target *ti;
1861 unsigned int i;
1862
1863 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1864 ti = dm_table_get_target(t, i);
1865
1866 if (!ti->num_secure_erase_bios)
1867 return false;
1868
1869 if (!ti->type->iterate_devices ||
1870 ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1871 return false;
1872 }
1873
1874 return true;
1875}
1876
1877void 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 if (dm_table_supports_dax_write_cache(t))
1911 dax_write_cache(t->md->dax_dev, true);
1912
1913 /* Ensure that all underlying devices are non-rotational. */
1914 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1915 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
1916 else
1917 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
1918
1919 if (!dm_table_supports_write_same(t))
1920 q->limits.max_write_same_sectors = 0;
1921 if (!dm_table_supports_write_zeroes(t))
1922 q->limits.max_write_zeroes_sectors = 0;
1923
1924 if (dm_table_all_devices_attribute(t, queue_supports_sg_merge))
1925 blk_queue_flag_clear(QUEUE_FLAG_NO_SG_MERGE, q);
1926 else
1927 blk_queue_flag_set(QUEUE_FLAG_NO_SG_MERGE, q);
1928
1929 dm_table_verify_integrity(t);
1930
1931 /*
1932 * Determine whether or not this queue's I/O timings contribute
1933 * to the entropy pool, Only request-based targets use this.
1934 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1935 * have it set.
1936 */
1937 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1938 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
1939}
1940
1941unsigned int dm_table_get_num_targets(struct dm_table *t)
1942{
1943 return t->num_targets;
1944}
1945
1946struct list_head *dm_table_get_devices(struct dm_table *t)
1947{
1948 return &t->devices;
1949}
1950
1951fmode_t dm_table_get_mode(struct dm_table *t)
1952{
1953 return t->mode;
1954}
1955EXPORT_SYMBOL(dm_table_get_mode);
1956
1957enum suspend_mode {
1958 PRESUSPEND,
1959 PRESUSPEND_UNDO,
1960 POSTSUSPEND,
1961};
1962
1963static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1964{
1965 int i = t->num_targets;
1966 struct dm_target *ti = t->targets;
1967
1968 lockdep_assert_held(&t->md->suspend_lock);
1969
1970 while (i--) {
1971 switch (mode) {
1972 case PRESUSPEND:
1973 if (ti->type->presuspend)
1974 ti->type->presuspend(ti);
1975 break;
1976 case PRESUSPEND_UNDO:
1977 if (ti->type->presuspend_undo)
1978 ti->type->presuspend_undo(ti);
1979 break;
1980 case POSTSUSPEND:
1981 if (ti->type->postsuspend)
1982 ti->type->postsuspend(ti);
1983 break;
1984 }
1985 ti++;
1986 }
1987}
1988
1989void dm_table_presuspend_targets(struct dm_table *t)
1990{
1991 if (!t)
1992 return;
1993
1994 suspend_targets(t, PRESUSPEND);
1995}
1996
1997void dm_table_presuspend_undo_targets(struct dm_table *t)
1998{
1999 if (!t)
2000 return;
2001
2002 suspend_targets(t, PRESUSPEND_UNDO);
2003}
2004
2005void dm_table_postsuspend_targets(struct dm_table *t)
2006{
2007 if (!t)
2008 return;
2009
2010 suspend_targets(t, POSTSUSPEND);
2011}
2012
2013int dm_table_resume_targets(struct dm_table *t)
2014{
2015 int i, r = 0;
2016
2017 lockdep_assert_held(&t->md->suspend_lock);
2018
2019 for (i = 0; i < t->num_targets; i++) {
2020 struct dm_target *ti = t->targets + i;
2021
2022 if (!ti->type->preresume)
2023 continue;
2024
2025 r = ti->type->preresume(ti);
2026 if (r) {
2027 DMERR("%s: %s: preresume failed, error = %d",
2028 dm_device_name(t->md), ti->type->name, r);
2029 return r;
2030 }
2031 }
2032
2033 for (i = 0; i < t->num_targets; i++) {
2034 struct dm_target *ti = t->targets + i;
2035
2036 if (ti->type->resume)
2037 ti->type->resume(ti);
2038 }
2039
2040 return 0;
2041}
2042
2043void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
2044{
2045 list_add(&cb->list, &t->target_callbacks);
2046}
2047EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
2048
2049int dm_table_any_congested(struct dm_table *t, int bdi_bits)
2050{
2051 struct dm_dev_internal *dd;
2052 struct list_head *devices = dm_table_get_devices(t);
2053 struct dm_target_callbacks *cb;
2054 int r = 0;
2055
2056 list_for_each_entry(dd, devices, list) {
2057 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
2058 char b[BDEVNAME_SIZE];
2059
2060 if (likely(q))
2061 r |= bdi_congested(q->backing_dev_info, bdi_bits);
2062 else
2063 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
2064 dm_device_name(t->md),
2065 bdevname(dd->dm_dev->bdev, b));
2066 }
2067
2068 list_for_each_entry(cb, &t->target_callbacks, list)
2069 if (cb->congested_fn)
2070 r |= cb->congested_fn(cb, bdi_bits);
2071
2072 return r;
2073}
2074
2075struct mapped_device *dm_table_get_md(struct dm_table *t)
2076{
2077 return t->md;
2078}
2079EXPORT_SYMBOL(dm_table_get_md);
2080
2081void dm_table_run_md_queue_async(struct dm_table *t)
2082{
2083 struct mapped_device *md;
2084 struct request_queue *queue;
2085 unsigned long flags;
2086
2087 if (!dm_table_request_based(t))
2088 return;
2089
2090 md = dm_table_get_md(t);
2091 queue = dm_get_md_queue(md);
2092 if (queue) {
2093 if (queue->mq_ops)
2094 blk_mq_run_hw_queues(queue, true);
2095 else {
2096 spin_lock_irqsave(queue->queue_lock, flags);
2097 blk_run_queue_async(queue);
2098 spin_unlock_irqrestore(queue->queue_lock, flags);
2099 }
2100 }
2101}
2102EXPORT_SYMBOL(dm_table_run_md_queue_async);
2103