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 struct target_type *immutable_target_type;
58 unsigned integrity_supported:1;
59 unsigned singleton:1;
60
61 /*
62 * Indicates the rw permissions for the new logical
63 * device. This should be a combination of FMODE_READ
64 * and FMODE_WRITE.
65 */
66 fmode_t mode;
67
68 /* a list of devices used by this table */
69 struct list_head devices;
70
71 /* events get handed up using this callback */
72 void (*event_fn)(void *);
73 void *event_context;
74
75 struct dm_md_mempools *mempools;
76
77 struct list_head target_callbacks;
78};
79
80/*
81 * Similar to ceiling(log_size(n))
82 */
83static unsigned int int_log(unsigned int n, unsigned int base)
84{
85 int result = 0;
86
87 while (n > 1) {
88 n = dm_div_up(n, base);
89 result++;
90 }
91
92 return result;
93}
94
95/*
96 * Calculate the index of the child node of the n'th node k'th key.
97 */
98static inline unsigned int get_child(unsigned int n, unsigned int k)
99{
100 return (n * CHILDREN_PER_NODE) + k;
101}
102
103/*
104 * Return the n'th node of level l from table t.
105 */
106static inline sector_t *get_node(struct dm_table *t,
107 unsigned int l, unsigned int n)
108{
109 return t->index[l] + (n * KEYS_PER_NODE);
110}
111
112/*
113 * Return the highest key that you could lookup from the n'th
114 * node on level l of the btree.
115 */
116static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
117{
118 for (; l < t->depth - 1; l++)
119 n = get_child(n, CHILDREN_PER_NODE - 1);
120
121 if (n >= t->counts[l])
122 return (sector_t) - 1;
123
124 return get_node(t, l, n)[KEYS_PER_NODE - 1];
125}
126
127/*
128 * Fills in a level of the btree based on the highs of the level
129 * below it.
130 */
131static int setup_btree_index(unsigned int l, struct dm_table *t)
132{
133 unsigned int n, k;
134 sector_t *node;
135
136 for (n = 0U; n < t->counts[l]; n++) {
137 node = get_node(t, l, n);
138
139 for (k = 0U; k < KEYS_PER_NODE; k++)
140 node[k] = high(t, l + 1, get_child(n, k));
141 }
142
143 return 0;
144}
145
146void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
147{
148 unsigned long size;
149 void *addr;
150
151 /*
152 * Check that we're not going to overflow.
153 */
154 if (nmemb > (ULONG_MAX / elem_size))
155 return NULL;
156
157 size = nmemb * elem_size;
158 addr = vzalloc(size);
159
160 return addr;
161}
162EXPORT_SYMBOL(dm_vcalloc);
163
164/*
165 * highs, and targets are managed as dynamic arrays during a
166 * table load.
167 */
168static int alloc_targets(struct dm_table *t, unsigned int num)
169{
170 sector_t *n_highs;
171 struct dm_target *n_targets;
172 int n = t->num_targets;
173
174 /*
175 * Allocate both the target array and offset array at once.
176 * Append an empty entry to catch sectors beyond the end of
177 * the device.
178 */
179 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
180 sizeof(sector_t));
181 if (!n_highs)
182 return -ENOMEM;
183
184 n_targets = (struct dm_target *) (n_highs + num);
185
186 if (n) {
187 memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
188 memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
189 }
190
191 memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
192 vfree(t->highs);
193
194 t->num_allocated = num;
195 t->highs = n_highs;
196 t->targets = n_targets;
197
198 return 0;
199}
200
201int dm_table_create(struct dm_table **result, fmode_t mode,
202 unsigned num_targets, struct mapped_device *md)
203{
204 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
205
206 if (!t)
207 return -ENOMEM;
208
209 INIT_LIST_HEAD(&t->devices);
210 INIT_LIST_HEAD(&t->target_callbacks);
211 atomic_set(&t->holders, 0);
212
213 if (!num_targets)
214 num_targets = KEYS_PER_NODE;
215
216 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
217
218 if (alloc_targets(t, num_targets)) {
219 kfree(t);
220 t = NULL;
221 return -ENOMEM;
222 }
223
224 t->mode = mode;
225 t->md = md;
226 *result = t;
227 return 0;
228}
229
230static void free_devices(struct list_head *devices)
231{
232 struct list_head *tmp, *next;
233
234 list_for_each_safe(tmp, next, devices) {
235 struct dm_dev_internal *dd =
236 list_entry(tmp, struct dm_dev_internal, list);
237 DMWARN("dm_table_destroy: dm_put_device call missing for %s",
238 dd->dm_dev.name);
239 kfree(dd);
240 }
241}
242
243void dm_table_destroy(struct dm_table *t)
244{
245 unsigned int i;
246
247 if (!t)
248 return;
249
250 while (atomic_read(&t->holders))
251 msleep(1);
252 smp_mb();
253
254 /* free the indexes */
255 if (t->depth >= 2)
256 vfree(t->index[t->depth - 2]);
257
258 /* free the targets */
259 for (i = 0; i < t->num_targets; i++) {
260 struct dm_target *tgt = t->targets + i;
261
262 if (tgt->type->dtr)
263 tgt->type->dtr(tgt);
264
265 dm_put_target_type(tgt->type);
266 }
267
268 vfree(t->highs);
269
270 /* free the device list */
271 free_devices(&t->devices);
272
273 dm_free_md_mempools(t->mempools);
274
275 kfree(t);
276}
277
278void dm_table_get(struct dm_table *t)
279{
280 atomic_inc(&t->holders);
281}
282EXPORT_SYMBOL(dm_table_get);
283
284void dm_table_put(struct dm_table *t)
285{
286 if (!t)
287 return;
288
289 smp_mb__before_atomic_dec();
290 atomic_dec(&t->holders);
291}
292EXPORT_SYMBOL(dm_table_put);
293
294/*
295 * Checks to see if we need to extend highs or targets.
296 */
297static inline int check_space(struct dm_table *t)
298{
299 if (t->num_targets >= t->num_allocated)
300 return alloc_targets(t, t->num_allocated * 2);
301
302 return 0;
303}
304
305/*
306 * See if we've already got a device in the list.
307 */
308static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
309{
310 struct dm_dev_internal *dd;
311
312 list_for_each_entry (dd, l, list)
313 if (dd->dm_dev.bdev->bd_dev == dev)
314 return dd;
315
316 return NULL;
317}
318
319/*
320 * Open a device so we can use it as a map destination.
321 */
322static int open_dev(struct dm_dev_internal *d, dev_t dev,
323 struct mapped_device *md)
324{
325 static char *_claim_ptr = "I belong to device-mapper";
326 struct block_device *bdev;
327
328 int r;
329
330 BUG_ON(d->dm_dev.bdev);
331
332 bdev = blkdev_get_by_dev(dev, d->dm_dev.mode | FMODE_EXCL, _claim_ptr);
333 if (IS_ERR(bdev))
334 return PTR_ERR(bdev);
335
336 r = bd_link_disk_holder(bdev, dm_disk(md));
337 if (r) {
338 blkdev_put(bdev, d->dm_dev.mode | FMODE_EXCL);
339 return r;
340 }
341
342 d->dm_dev.bdev = bdev;
343 return 0;
344}
345
346/*
347 * Close a device that we've been using.
348 */
349static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
350{
351 if (!d->dm_dev.bdev)
352 return;
353
354 bd_unlink_disk_holder(d->dm_dev.bdev, dm_disk(md));
355 blkdev_put(d->dm_dev.bdev, d->dm_dev.mode | FMODE_EXCL);
356 d->dm_dev.bdev = NULL;
357}
358
359/*
360 * If possible, this checks an area of a destination device is invalid.
361 */
362static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
363 sector_t start, sector_t len, void *data)
364{
365 struct request_queue *q;
366 struct queue_limits *limits = data;
367 struct block_device *bdev = dev->bdev;
368 sector_t dev_size =
369 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
370 unsigned short logical_block_size_sectors =
371 limits->logical_block_size >> SECTOR_SHIFT;
372 char b[BDEVNAME_SIZE];
373
374 /*
375 * Some devices exist without request functions,
376 * such as loop devices not yet bound to backing files.
377 * Forbid the use of such devices.
378 */
379 q = bdev_get_queue(bdev);
380 if (!q || !q->make_request_fn) {
381 DMWARN("%s: %s is not yet initialised: "
382 "start=%llu, len=%llu, dev_size=%llu",
383 dm_device_name(ti->table->md), bdevname(bdev, b),
384 (unsigned long long)start,
385 (unsigned long long)len,
386 (unsigned long long)dev_size);
387 return 1;
388 }
389
390 if (!dev_size)
391 return 0;
392
393 if ((start >= dev_size) || (start + len > dev_size)) {
394 DMWARN("%s: %s too small for target: "
395 "start=%llu, len=%llu, dev_size=%llu",
396 dm_device_name(ti->table->md), bdevname(bdev, b),
397 (unsigned long long)start,
398 (unsigned long long)len,
399 (unsigned long long)dev_size);
400 return 1;
401 }
402
403 if (logical_block_size_sectors <= 1)
404 return 0;
405
406 if (start & (logical_block_size_sectors - 1)) {
407 DMWARN("%s: start=%llu not aligned to h/w "
408 "logical block size %u of %s",
409 dm_device_name(ti->table->md),
410 (unsigned long long)start,
411 limits->logical_block_size, bdevname(bdev, b));
412 return 1;
413 }
414
415 if (len & (logical_block_size_sectors - 1)) {
416 DMWARN("%s: len=%llu not aligned to h/w "
417 "logical block size %u of %s",
418 dm_device_name(ti->table->md),
419 (unsigned long long)len,
420 limits->logical_block_size, bdevname(bdev, b));
421 return 1;
422 }
423
424 return 0;
425}
426
427/*
428 * This upgrades the mode on an already open dm_dev, being
429 * careful to leave things as they were if we fail to reopen the
430 * device and not to touch the existing bdev field in case
431 * it is accessed concurrently inside dm_table_any_congested().
432 */
433static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
434 struct mapped_device *md)
435{
436 int r;
437 struct dm_dev_internal dd_new, dd_old;
438
439 dd_new = dd_old = *dd;
440
441 dd_new.dm_dev.mode |= new_mode;
442 dd_new.dm_dev.bdev = NULL;
443
444 r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md);
445 if (r)
446 return r;
447
448 dd->dm_dev.mode |= new_mode;
449 close_dev(&dd_old, md);
450
451 return 0;
452}
453
454/*
455 * Add a device to the list, or just increment the usage count if
456 * it's already present.
457 */
458int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
459 struct dm_dev **result)
460{
461 int r;
462 dev_t uninitialized_var(dev);
463 struct dm_dev_internal *dd;
464 unsigned int major, minor;
465 struct dm_table *t = ti->table;
466 char dummy;
467
468 BUG_ON(!t);
469
470 if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
471 /* Extract the major/minor numbers */
472 dev = MKDEV(major, minor);
473 if (MAJOR(dev) != major || MINOR(dev) != minor)
474 return -EOVERFLOW;
475 } else {
476 /* convert the path to a device */
477 struct block_device *bdev = lookup_bdev(path);
478
479 if (IS_ERR(bdev))
480 return PTR_ERR(bdev);
481 dev = bdev->bd_dev;
482 bdput(bdev);
483 }
484
485 dd = find_device(&t->devices, dev);
486 if (!dd) {
487 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
488 if (!dd)
489 return -ENOMEM;
490
491 dd->dm_dev.mode = mode;
492 dd->dm_dev.bdev = NULL;
493
494 if ((r = open_dev(dd, dev, t->md))) {
495 kfree(dd);
496 return r;
497 }
498
499 format_dev_t(dd->dm_dev.name, dev);
500
501 atomic_set(&dd->count, 0);
502 list_add(&dd->list, &t->devices);
503
504 } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
505 r = upgrade_mode(dd, mode, t->md);
506 if (r)
507 return r;
508 }
509 atomic_inc(&dd->count);
510
511 *result = &dd->dm_dev;
512 return 0;
513}
514EXPORT_SYMBOL(dm_get_device);
515
516int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
517 sector_t start, sector_t len, void *data)
518{
519 struct queue_limits *limits = data;
520 struct block_device *bdev = dev->bdev;
521 struct request_queue *q = bdev_get_queue(bdev);
522 char b[BDEVNAME_SIZE];
523
524 if (unlikely(!q)) {
525 DMWARN("%s: Cannot set limits for nonexistent device %s",
526 dm_device_name(ti->table->md), bdevname(bdev, b));
527 return 0;
528 }
529
530 if (bdev_stack_limits(limits, bdev, start) < 0)
531 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
532 "physical_block_size=%u, logical_block_size=%u, "
533 "alignment_offset=%u, start=%llu",
534 dm_device_name(ti->table->md), bdevname(bdev, b),
535 q->limits.physical_block_size,
536 q->limits.logical_block_size,
537 q->limits.alignment_offset,
538 (unsigned long long) start << SECTOR_SHIFT);
539
540 /*
541 * Check if merge fn is supported.
542 * If not we'll force DM to use PAGE_SIZE or
543 * smaller I/O, just to be safe.
544 */
545 if (dm_queue_merge_is_compulsory(q) && !ti->type->merge)
546 blk_limits_max_hw_sectors(limits,
547 (unsigned int) (PAGE_SIZE >> 9));
548 return 0;
549}
550EXPORT_SYMBOL_GPL(dm_set_device_limits);
551
552/*
553 * Decrement a device's use count and remove it if necessary.
554 */
555void dm_put_device(struct dm_target *ti, struct dm_dev *d)
556{
557 struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
558 dm_dev);
559
560 if (atomic_dec_and_test(&dd->count)) {
561 close_dev(dd, ti->table->md);
562 list_del(&dd->list);
563 kfree(dd);
564 }
565}
566EXPORT_SYMBOL(dm_put_device);
567
568/*
569 * Checks to see if the target joins onto the end of the table.
570 */
571static int adjoin(struct dm_table *table, struct dm_target *ti)
572{
573 struct dm_target *prev;
574
575 if (!table->num_targets)
576 return !ti->begin;
577
578 prev = &table->targets[table->num_targets - 1];
579 return (ti->begin == (prev->begin + prev->len));
580}
581
582/*
583 * Used to dynamically allocate the arg array.
584 */
585static char **realloc_argv(unsigned *array_size, char **old_argv)
586{
587 char **argv;
588 unsigned new_size;
589
590 new_size = *array_size ? *array_size * 2 : 64;
591 argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
592 if (argv) {
593 memcpy(argv, old_argv, *array_size * sizeof(*argv));
594 *array_size = new_size;
595 }
596
597 kfree(old_argv);
598 return argv;
599}
600
601/*
602 * Destructively splits up the argument list to pass to ctr.
603 */
604int dm_split_args(int *argc, char ***argvp, char *input)
605{
606 char *start, *end = input, *out, **argv = NULL;
607 unsigned array_size = 0;
608
609 *argc = 0;
610
611 if (!input) {
612 *argvp = NULL;
613 return 0;
614 }
615
616 argv = realloc_argv(&array_size, argv);
617 if (!argv)
618 return -ENOMEM;
619
620 while (1) {
621 /* Skip whitespace */
622 start = skip_spaces(end);
623
624 if (!*start)
625 break; /* success, we hit the end */
626
627 /* 'out' is used to remove any back-quotes */
628 end = out = start;
629 while (*end) {
630 /* Everything apart from '\0' can be quoted */
631 if (*end == '\\' && *(end + 1)) {
632 *out++ = *(end + 1);
633 end += 2;
634 continue;
635 }
636
637 if (isspace(*end))
638 break; /* end of token */
639
640 *out++ = *end++;
641 }
642
643 /* have we already filled the array ? */
644 if ((*argc + 1) > array_size) {
645 argv = realloc_argv(&array_size, argv);
646 if (!argv)
647 return -ENOMEM;
648 }
649
650 /* we know this is whitespace */
651 if (*end)
652 end++;
653
654 /* terminate the string and put it in the array */
655 *out = '\0';
656 argv[*argc] = start;
657 (*argc)++;
658 }
659
660 *argvp = argv;
661 return 0;
662}
663
664/*
665 * Impose necessary and sufficient conditions on a devices's table such
666 * that any incoming bio which respects its logical_block_size can be
667 * processed successfully. If it falls across the boundary between
668 * two or more targets, the size of each piece it gets split into must
669 * be compatible with the logical_block_size of the target processing it.
670 */
671static int validate_hardware_logical_block_alignment(struct dm_table *table,
672 struct queue_limits *limits)
673{
674 /*
675 * This function uses arithmetic modulo the logical_block_size
676 * (in units of 512-byte sectors).
677 */
678 unsigned short device_logical_block_size_sects =
679 limits->logical_block_size >> SECTOR_SHIFT;
680
681 /*
682 * Offset of the start of the next table entry, mod logical_block_size.
683 */
684 unsigned short next_target_start = 0;
685
686 /*
687 * Given an aligned bio that extends beyond the end of a
688 * target, how many sectors must the next target handle?
689 */
690 unsigned short remaining = 0;
691
692 struct dm_target *uninitialized_var(ti);
693 struct queue_limits ti_limits;
694 unsigned i = 0;
695
696 /*
697 * Check each entry in the table in turn.
698 */
699 while (i < dm_table_get_num_targets(table)) {
700 ti = dm_table_get_target(table, i++);
701
702 blk_set_stacking_limits(&ti_limits);
703
704 /* combine all target devices' limits */
705 if (ti->type->iterate_devices)
706 ti->type->iterate_devices(ti, dm_set_device_limits,
707 &ti_limits);
708
709 /*
710 * If the remaining sectors fall entirely within this
711 * table entry are they compatible with its logical_block_size?
712 */
713 if (remaining < ti->len &&
714 remaining & ((ti_limits.logical_block_size >>
715 SECTOR_SHIFT) - 1))
716 break; /* Error */
717
718 next_target_start =
719 (unsigned short) ((next_target_start + ti->len) &
720 (device_logical_block_size_sects - 1));
721 remaining = next_target_start ?
722 device_logical_block_size_sects - next_target_start : 0;
723 }
724
725 if (remaining) {
726 DMWARN("%s: table line %u (start sect %llu len %llu) "
727 "not aligned to h/w logical block size %u",
728 dm_device_name(table->md), i,
729 (unsigned long long) ti->begin,
730 (unsigned long long) ti->len,
731 limits->logical_block_size);
732 return -EINVAL;
733 }
734
735 return 0;
736}
737
738int dm_table_add_target(struct dm_table *t, const char *type,
739 sector_t start, sector_t len, char *params)
740{
741 int r = -EINVAL, argc;
742 char **argv;
743 struct dm_target *tgt;
744
745 if (t->singleton) {
746 DMERR("%s: target type %s must appear alone in table",
747 dm_device_name(t->md), t->targets->type->name);
748 return -EINVAL;
749 }
750
751 if ((r = check_space(t)))
752 return r;
753
754 tgt = t->targets + t->num_targets;
755 memset(tgt, 0, sizeof(*tgt));
756
757 if (!len) {
758 DMERR("%s: zero-length target", dm_device_name(t->md));
759 return -EINVAL;
760 }
761
762 tgt->type = dm_get_target_type(type);
763 if (!tgt->type) {
764 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
765 type);
766 return -EINVAL;
767 }
768
769 if (dm_target_needs_singleton(tgt->type)) {
770 if (t->num_targets) {
771 DMERR("%s: target type %s must appear alone in table",
772 dm_device_name(t->md), type);
773 return -EINVAL;
774 }
775 t->singleton = 1;
776 }
777
778 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
779 DMERR("%s: target type %s may not be included in read-only tables",
780 dm_device_name(t->md), type);
781 return -EINVAL;
782 }
783
784 if (t->immutable_target_type) {
785 if (t->immutable_target_type != tgt->type) {
786 DMERR("%s: immutable target type %s cannot be mixed with other target types",
787 dm_device_name(t->md), t->immutable_target_type->name);
788 return -EINVAL;
789 }
790 } else if (dm_target_is_immutable(tgt->type)) {
791 if (t->num_targets) {
792 DMERR("%s: immutable target type %s cannot be mixed with other target types",
793 dm_device_name(t->md), tgt->type->name);
794 return -EINVAL;
795 }
796 t->immutable_target_type = tgt->type;
797 }
798
799 tgt->table = t;
800 tgt->begin = start;
801 tgt->len = len;
802 tgt->error = "Unknown error";
803
804 /*
805 * Does this target adjoin the previous one ?
806 */
807 if (!adjoin(t, tgt)) {
808 tgt->error = "Gap in table";
809 r = -EINVAL;
810 goto bad;
811 }
812
813 r = dm_split_args(&argc, &argv, params);
814 if (r) {
815 tgt->error = "couldn't split parameters (insufficient memory)";
816 goto bad;
817 }
818
819 r = tgt->type->ctr(tgt, argc, argv);
820 kfree(argv);
821 if (r)
822 goto bad;
823
824 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
825
826 if (!tgt->num_discard_requests && tgt->discards_supported)
827 DMWARN("%s: %s: ignoring discards_supported because num_discard_requests is zero.",
828 dm_device_name(t->md), type);
829
830 return 0;
831
832 bad:
833 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
834 dm_put_target_type(tgt->type);
835 return r;
836}
837
838/*
839 * Target argument parsing helpers.
840 */
841static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
842 unsigned *value, char **error, unsigned grouped)
843{
844 const char *arg_str = dm_shift_arg(arg_set);
845 char dummy;
846
847 if (!arg_str ||
848 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
849 (*value < arg->min) ||
850 (*value > arg->max) ||
851 (grouped && arg_set->argc < *value)) {
852 *error = arg->error;
853 return -EINVAL;
854 }
855
856 return 0;
857}
858
859int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
860 unsigned *value, char **error)
861{
862 return validate_next_arg(arg, arg_set, value, error, 0);
863}
864EXPORT_SYMBOL(dm_read_arg);
865
866int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
867 unsigned *value, char **error)
868{
869 return validate_next_arg(arg, arg_set, value, error, 1);
870}
871EXPORT_SYMBOL(dm_read_arg_group);
872
873const char *dm_shift_arg(struct dm_arg_set *as)
874{
875 char *r;
876
877 if (as->argc) {
878 as->argc--;
879 r = *as->argv;
880 as->argv++;
881 return r;
882 }
883
884 return NULL;
885}
886EXPORT_SYMBOL(dm_shift_arg);
887
888void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
889{
890 BUG_ON(as->argc < num_args);
891 as->argc -= num_args;
892 as->argv += num_args;
893}
894EXPORT_SYMBOL(dm_consume_args);
895
896static int dm_table_set_type(struct dm_table *t)
897{
898 unsigned i;
899 unsigned bio_based = 0, request_based = 0;
900 struct dm_target *tgt;
901 struct dm_dev_internal *dd;
902 struct list_head *devices;
903
904 for (i = 0; i < t->num_targets; i++) {
905 tgt = t->targets + i;
906 if (dm_target_request_based(tgt))
907 request_based = 1;
908 else
909 bio_based = 1;
910
911 if (bio_based && request_based) {
912 DMWARN("Inconsistent table: different target types"
913 " can't be mixed up");
914 return -EINVAL;
915 }
916 }
917
918 if (bio_based) {
919 /* We must use this table as bio-based */
920 t->type = DM_TYPE_BIO_BASED;
921 return 0;
922 }
923
924 BUG_ON(!request_based); /* No targets in this table */
925
926 /* Non-request-stackable devices can't be used for request-based dm */
927 devices = dm_table_get_devices(t);
928 list_for_each_entry(dd, devices, list) {
929 if (!blk_queue_stackable(bdev_get_queue(dd->dm_dev.bdev))) {
930 DMWARN("table load rejected: including"
931 " non-request-stackable devices");
932 return -EINVAL;
933 }
934 }
935
936 /*
937 * Request-based dm supports only tables that have a single target now.
938 * To support multiple targets, request splitting support is needed,
939 * and that needs lots of changes in the block-layer.
940 * (e.g. request completion process for partial completion.)
941 */
942 if (t->num_targets > 1) {
943 DMWARN("Request-based dm doesn't support multiple targets yet");
944 return -EINVAL;
945 }
946
947 t->type = DM_TYPE_REQUEST_BASED;
948
949 return 0;
950}
951
952unsigned dm_table_get_type(struct dm_table *t)
953{
954 return t->type;
955}
956
957struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
958{
959 return t->immutable_target_type;
960}
961
962bool dm_table_request_based(struct dm_table *t)
963{
964 return dm_table_get_type(t) == DM_TYPE_REQUEST_BASED;
965}
966
967int dm_table_alloc_md_mempools(struct dm_table *t)
968{
969 unsigned type = dm_table_get_type(t);
970
971 if (unlikely(type == DM_TYPE_NONE)) {
972 DMWARN("no table type is set, can't allocate mempools");
973 return -EINVAL;
974 }
975
976 t->mempools = dm_alloc_md_mempools(type, t->integrity_supported);
977 if (!t->mempools)
978 return -ENOMEM;
979
980 return 0;
981}
982
983void dm_table_free_md_mempools(struct dm_table *t)
984{
985 dm_free_md_mempools(t->mempools);
986 t->mempools = NULL;
987}
988
989struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
990{
991 return t->mempools;
992}
993
994static int setup_indexes(struct dm_table *t)
995{
996 int i;
997 unsigned int total = 0;
998 sector_t *indexes;
999
1000 /* allocate the space for *all* the indexes */
1001 for (i = t->depth - 2; i >= 0; i--) {
1002 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1003 total += t->counts[i];
1004 }
1005
1006 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1007 if (!indexes)
1008 return -ENOMEM;
1009
1010 /* set up internal nodes, bottom-up */
1011 for (i = t->depth - 2; i >= 0; i--) {
1012 t->index[i] = indexes;
1013 indexes += (KEYS_PER_NODE * t->counts[i]);
1014 setup_btree_index(i, t);
1015 }
1016
1017 return 0;
1018}
1019
1020/*
1021 * Builds the btree to index the map.
1022 */
1023static int dm_table_build_index(struct dm_table *t)
1024{
1025 int r = 0;
1026 unsigned int leaf_nodes;
1027
1028 /* how many indexes will the btree have ? */
1029 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1030 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1031
1032 /* leaf layer has already been set up */
1033 t->counts[t->depth - 1] = leaf_nodes;
1034 t->index[t->depth - 1] = t->highs;
1035
1036 if (t->depth >= 2)
1037 r = setup_indexes(t);
1038
1039 return r;
1040}
1041
1042/*
1043 * Get a disk whose integrity profile reflects the table's profile.
1044 * If %match_all is true, all devices' profiles must match.
1045 * If %match_all is false, all devices must at least have an
1046 * allocated integrity profile; but uninitialized is ok.
1047 * Returns NULL if integrity support was inconsistent or unavailable.
1048 */
1049static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t,
1050 bool match_all)
1051{
1052 struct list_head *devices = dm_table_get_devices(t);
1053 struct dm_dev_internal *dd = NULL;
1054 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1055
1056 list_for_each_entry(dd, devices, list) {
1057 template_disk = dd->dm_dev.bdev->bd_disk;
1058 if (!blk_get_integrity(template_disk))
1059 goto no_integrity;
1060 if (!match_all && !blk_integrity_is_initialized(template_disk))
1061 continue; /* skip uninitialized profiles */
1062 else if (prev_disk &&
1063 blk_integrity_compare(prev_disk, template_disk) < 0)
1064 goto no_integrity;
1065 prev_disk = template_disk;
1066 }
1067
1068 return template_disk;
1069
1070no_integrity:
1071 if (prev_disk)
1072 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1073 dm_device_name(t->md),
1074 prev_disk->disk_name,
1075 template_disk->disk_name);
1076 return NULL;
1077}
1078
1079/*
1080 * Register the mapped device for blk_integrity support if
1081 * the underlying devices have an integrity profile. But all devices
1082 * may not have matching profiles (checking all devices isn't reliable
1083 * during table load because this table may use other DM device(s) which
1084 * must be resumed before they will have an initialized integity profile).
1085 * Stacked DM devices force a 2 stage integrity profile validation:
1086 * 1 - during load, validate all initialized integrity profiles match
1087 * 2 - during resume, validate all integrity profiles match
1088 */
1089static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md)
1090{
1091 struct gendisk *template_disk = NULL;
1092
1093 template_disk = dm_table_get_integrity_disk(t, false);
1094 if (!template_disk)
1095 return 0;
1096
1097 if (!blk_integrity_is_initialized(dm_disk(md))) {
1098 t->integrity_supported = 1;
1099 return blk_integrity_register(dm_disk(md), NULL);
1100 }
1101
1102 /*
1103 * If DM device already has an initalized integrity
1104 * profile the new profile should not conflict.
1105 */
1106 if (blk_integrity_is_initialized(template_disk) &&
1107 blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1108 DMWARN("%s: conflict with existing integrity profile: "
1109 "%s profile mismatch",
1110 dm_device_name(t->md),
1111 template_disk->disk_name);
1112 return 1;
1113 }
1114
1115 /* Preserve existing initialized integrity profile */
1116 t->integrity_supported = 1;
1117 return 0;
1118}
1119
1120/*
1121 * Prepares the table for use by building the indices,
1122 * setting the type, and allocating mempools.
1123 */
1124int dm_table_complete(struct dm_table *t)
1125{
1126 int r;
1127
1128 r = dm_table_set_type(t);
1129 if (r) {
1130 DMERR("unable to set table type");
1131 return r;
1132 }
1133
1134 r = dm_table_build_index(t);
1135 if (r) {
1136 DMERR("unable to build btrees");
1137 return r;
1138 }
1139
1140 r = dm_table_prealloc_integrity(t, t->md);
1141 if (r) {
1142 DMERR("could not register integrity profile.");
1143 return r;
1144 }
1145
1146 r = dm_table_alloc_md_mempools(t);
1147 if (r)
1148 DMERR("unable to allocate mempools");
1149
1150 return r;
1151}
1152
1153static DEFINE_MUTEX(_event_lock);
1154void dm_table_event_callback(struct dm_table *t,
1155 void (*fn)(void *), void *context)
1156{
1157 mutex_lock(&_event_lock);
1158 t->event_fn = fn;
1159 t->event_context = context;
1160 mutex_unlock(&_event_lock);
1161}
1162
1163void dm_table_event(struct dm_table *t)
1164{
1165 /*
1166 * You can no longer call dm_table_event() from interrupt
1167 * context, use a bottom half instead.
1168 */
1169 BUG_ON(in_interrupt());
1170
1171 mutex_lock(&_event_lock);
1172 if (t->event_fn)
1173 t->event_fn(t->event_context);
1174 mutex_unlock(&_event_lock);
1175}
1176EXPORT_SYMBOL(dm_table_event);
1177
1178sector_t dm_table_get_size(struct dm_table *t)
1179{
1180 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1181}
1182EXPORT_SYMBOL(dm_table_get_size);
1183
1184struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1185{
1186 if (index >= t->num_targets)
1187 return NULL;
1188
1189 return t->targets + index;
1190}
1191
1192/*
1193 * Search the btree for the correct target.
1194 *
1195 * Caller should check returned pointer with dm_target_is_valid()
1196 * to trap I/O beyond end of device.
1197 */
1198struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1199{
1200 unsigned int l, n = 0, k = 0;
1201 sector_t *node;
1202
1203 for (l = 0; l < t->depth; l++) {
1204 n = get_child(n, k);
1205 node = get_node(t, l, n);
1206
1207 for (k = 0; k < KEYS_PER_NODE; k++)
1208 if (node[k] >= sector)
1209 break;
1210 }
1211
1212 return &t->targets[(KEYS_PER_NODE * n) + k];
1213}
1214
1215/*
1216 * Establish the new table's queue_limits and validate them.
1217 */
1218int dm_calculate_queue_limits(struct dm_table *table,
1219 struct queue_limits *limits)
1220{
1221 struct dm_target *uninitialized_var(ti);
1222 struct queue_limits ti_limits;
1223 unsigned i = 0;
1224
1225 blk_set_stacking_limits(limits);
1226
1227 while (i < dm_table_get_num_targets(table)) {
1228 blk_set_stacking_limits(&ti_limits);
1229
1230 ti = dm_table_get_target(table, i++);
1231
1232 if (!ti->type->iterate_devices)
1233 goto combine_limits;
1234
1235 /*
1236 * Combine queue limits of all the devices this target uses.
1237 */
1238 ti->type->iterate_devices(ti, dm_set_device_limits,
1239 &ti_limits);
1240
1241 /* Set I/O hints portion of queue limits */
1242 if (ti->type->io_hints)
1243 ti->type->io_hints(ti, &ti_limits);
1244
1245 /*
1246 * Check each device area is consistent with the target's
1247 * overall queue limits.
1248 */
1249 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1250 &ti_limits))
1251 return -EINVAL;
1252
1253combine_limits:
1254 /*
1255 * Merge this target's queue limits into the overall limits
1256 * for the table.
1257 */
1258 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1259 DMWARN("%s: adding target device "
1260 "(start sect %llu len %llu) "
1261 "caused an alignment inconsistency",
1262 dm_device_name(table->md),
1263 (unsigned long long) ti->begin,
1264 (unsigned long long) ti->len);
1265 }
1266
1267 return validate_hardware_logical_block_alignment(table, limits);
1268}
1269
1270/*
1271 * Set the integrity profile for this device if all devices used have
1272 * matching profiles. We're quite deep in the resume path but still
1273 * don't know if all devices (particularly DM devices this device
1274 * may be stacked on) have matching profiles. Even if the profiles
1275 * don't match we have no way to fail (to resume) at this point.
1276 */
1277static void dm_table_set_integrity(struct dm_table *t)
1278{
1279 struct gendisk *template_disk = NULL;
1280
1281 if (!blk_get_integrity(dm_disk(t->md)))
1282 return;
1283
1284 template_disk = dm_table_get_integrity_disk(t, true);
1285 if (template_disk)
1286 blk_integrity_register(dm_disk(t->md),
1287 blk_get_integrity(template_disk));
1288 else if (blk_integrity_is_initialized(dm_disk(t->md)))
1289 DMWARN("%s: device no longer has a valid integrity profile",
1290 dm_device_name(t->md));
1291 else
1292 DMWARN("%s: unable to establish an integrity profile",
1293 dm_device_name(t->md));
1294}
1295
1296static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1297 sector_t start, sector_t len, void *data)
1298{
1299 unsigned flush = (*(unsigned *)data);
1300 struct request_queue *q = bdev_get_queue(dev->bdev);
1301
1302 return q && (q->flush_flags & flush);
1303}
1304
1305static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1306{
1307 struct dm_target *ti;
1308 unsigned i = 0;
1309
1310 /*
1311 * Require at least one underlying device to support flushes.
1312 * t->devices includes internal dm devices such as mirror logs
1313 * so we need to use iterate_devices here, which targets
1314 * supporting flushes must provide.
1315 */
1316 while (i < dm_table_get_num_targets(t)) {
1317 ti = dm_table_get_target(t, i++);
1318
1319 if (!ti->num_flush_requests)
1320 continue;
1321
1322 if (ti->type->iterate_devices &&
1323 ti->type->iterate_devices(ti, device_flush_capable, &flush))
1324 return 1;
1325 }
1326
1327 return 0;
1328}
1329
1330static bool dm_table_discard_zeroes_data(struct dm_table *t)
1331{
1332 struct dm_target *ti;
1333 unsigned i = 0;
1334
1335 /* Ensure that all targets supports discard_zeroes_data. */
1336 while (i < dm_table_get_num_targets(t)) {
1337 ti = dm_table_get_target(t, i++);
1338
1339 if (ti->discard_zeroes_data_unsupported)
1340 return 0;
1341 }
1342
1343 return 1;
1344}
1345
1346static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1347 sector_t start, sector_t len, void *data)
1348{
1349 struct request_queue *q = bdev_get_queue(dev->bdev);
1350
1351 return q && blk_queue_nonrot(q);
1352}
1353
1354static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1355 sector_t start, sector_t len, void *data)
1356{
1357 struct request_queue *q = bdev_get_queue(dev->bdev);
1358
1359 return q && !blk_queue_add_random(q);
1360}
1361
1362static bool dm_table_all_devices_attribute(struct dm_table *t,
1363 iterate_devices_callout_fn func)
1364{
1365 struct dm_target *ti;
1366 unsigned i = 0;
1367
1368 while (i < dm_table_get_num_targets(t)) {
1369 ti = dm_table_get_target(t, i++);
1370
1371 if (!ti->type->iterate_devices ||
1372 !ti->type->iterate_devices(ti, func, NULL))
1373 return 0;
1374 }
1375
1376 return 1;
1377}
1378
1379void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1380 struct queue_limits *limits)
1381{
1382 unsigned flush = 0;
1383
1384 /*
1385 * Copy table's limits to the DM device's request_queue
1386 */
1387 q->limits = *limits;
1388
1389 if (!dm_table_supports_discards(t))
1390 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1391 else
1392 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1393
1394 if (dm_table_supports_flush(t, REQ_FLUSH)) {
1395 flush |= REQ_FLUSH;
1396 if (dm_table_supports_flush(t, REQ_FUA))
1397 flush |= REQ_FUA;
1398 }
1399 blk_queue_flush(q, flush);
1400
1401 if (!dm_table_discard_zeroes_data(t))
1402 q->limits.discard_zeroes_data = 0;
1403
1404 /* Ensure that all underlying devices are non-rotational. */
1405 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1406 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1407 else
1408 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1409
1410 dm_table_set_integrity(t);
1411
1412 /*
1413 * Determine whether or not this queue's I/O timings contribute
1414 * to the entropy pool, Only request-based targets use this.
1415 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1416 * have it set.
1417 */
1418 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1419 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1420
1421 /*
1422 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1423 * visible to other CPUs because, once the flag is set, incoming bios
1424 * are processed by request-based dm, which refers to the queue
1425 * settings.
1426 * Until the flag set, bios are passed to bio-based dm and queued to
1427 * md->deferred where queue settings are not needed yet.
1428 * Those bios are passed to request-based dm at the resume time.
1429 */
1430 smp_mb();
1431 if (dm_table_request_based(t))
1432 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1433}
1434
1435unsigned int dm_table_get_num_targets(struct dm_table *t)
1436{
1437 return t->num_targets;
1438}
1439
1440struct list_head *dm_table_get_devices(struct dm_table *t)
1441{
1442 return &t->devices;
1443}
1444
1445fmode_t dm_table_get_mode(struct dm_table *t)
1446{
1447 return t->mode;
1448}
1449EXPORT_SYMBOL(dm_table_get_mode);
1450
1451static void suspend_targets(struct dm_table *t, unsigned postsuspend)
1452{
1453 int i = t->num_targets;
1454 struct dm_target *ti = t->targets;
1455
1456 while (i--) {
1457 if (postsuspend) {
1458 if (ti->type->postsuspend)
1459 ti->type->postsuspend(ti);
1460 } else if (ti->type->presuspend)
1461 ti->type->presuspend(ti);
1462
1463 ti++;
1464 }
1465}
1466
1467void dm_table_presuspend_targets(struct dm_table *t)
1468{
1469 if (!t)
1470 return;
1471
1472 suspend_targets(t, 0);
1473}
1474
1475void dm_table_postsuspend_targets(struct dm_table *t)
1476{
1477 if (!t)
1478 return;
1479
1480 suspend_targets(t, 1);
1481}
1482
1483int dm_table_resume_targets(struct dm_table *t)
1484{
1485 int i, r = 0;
1486
1487 for (i = 0; i < t->num_targets; i++) {
1488 struct dm_target *ti = t->targets + i;
1489
1490 if (!ti->type->preresume)
1491 continue;
1492
1493 r = ti->type->preresume(ti);
1494 if (r)
1495 return r;
1496 }
1497
1498 for (i = 0; i < t->num_targets; i++) {
1499 struct dm_target *ti = t->targets + i;
1500
1501 if (ti->type->resume)
1502 ti->type->resume(ti);
1503 }
1504
1505 return 0;
1506}
1507
1508void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1509{
1510 list_add(&cb->list, &t->target_callbacks);
1511}
1512EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1513
1514int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1515{
1516 struct dm_dev_internal *dd;
1517 struct list_head *devices = dm_table_get_devices(t);
1518 struct dm_target_callbacks *cb;
1519 int r = 0;
1520
1521 list_for_each_entry(dd, devices, list) {
1522 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
1523 char b[BDEVNAME_SIZE];
1524
1525 if (likely(q))
1526 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1527 else
1528 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1529 dm_device_name(t->md),
1530 bdevname(dd->dm_dev.bdev, b));
1531 }
1532
1533 list_for_each_entry(cb, &t->target_callbacks, list)
1534 if (cb->congested_fn)
1535 r |= cb->congested_fn(cb, bdi_bits);
1536
1537 return r;
1538}
1539
1540int dm_table_any_busy_target(struct dm_table *t)
1541{
1542 unsigned i;
1543 struct dm_target *ti;
1544
1545 for (i = 0; i < t->num_targets; i++) {
1546 ti = t->targets + i;
1547 if (ti->type->busy && ti->type->busy(ti))
1548 return 1;
1549 }
1550
1551 return 0;
1552}
1553
1554struct mapped_device *dm_table_get_md(struct dm_table *t)
1555{
1556 return t->md;
1557}
1558EXPORT_SYMBOL(dm_table_get_md);
1559
1560static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1561 sector_t start, sector_t len, void *data)
1562{
1563 struct request_queue *q = bdev_get_queue(dev->bdev);
1564
1565 return q && blk_queue_discard(q);
1566}
1567
1568bool dm_table_supports_discards(struct dm_table *t)
1569{
1570 struct dm_target *ti;
1571 unsigned i = 0;
1572
1573 /*
1574 * Unless any target used by the table set discards_supported,
1575 * require at least one underlying device to support discards.
1576 * t->devices includes internal dm devices such as mirror logs
1577 * so we need to use iterate_devices here, which targets
1578 * supporting discard selectively must provide.
1579 */
1580 while (i < dm_table_get_num_targets(t)) {
1581 ti = dm_table_get_target(t, i++);
1582
1583 if (!ti->num_discard_requests)
1584 continue;
1585
1586 if (ti->discards_supported)
1587 return 1;
1588
1589 if (ti->type->iterate_devices &&
1590 ti->type->iterate_devices(ti, device_discard_capable, NULL))
1591 return 1;
1592 }
1593
1594 return 0;
1595}
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 unsigned integrity_added:1;
51
52 /*
53 * Indicates the rw permissions for the new logical
54 * device. This should be a combination of FMODE_READ
55 * and FMODE_WRITE.
56 */
57 fmode_t mode;
58
59 /* a list of devices used by this table */
60 struct list_head devices;
61
62 /* events get handed up using this callback */
63 void (*event_fn)(void *);
64 void *event_context;
65
66 struct dm_md_mempools *mempools;
67
68 struct list_head target_callbacks;
69};
70
71/*
72 * Similar to ceiling(log_size(n))
73 */
74static unsigned int int_log(unsigned int n, unsigned int base)
75{
76 int result = 0;
77
78 while (n > 1) {
79 n = dm_div_up(n, base);
80 result++;
81 }
82
83 return result;
84}
85
86/*
87 * Calculate the index of the child node of the n'th node k'th key.
88 */
89static inline unsigned int get_child(unsigned int n, unsigned int k)
90{
91 return (n * CHILDREN_PER_NODE) + k;
92}
93
94/*
95 * Return the n'th node of level l from table t.
96 */
97static inline sector_t *get_node(struct dm_table *t,
98 unsigned int l, unsigned int n)
99{
100 return t->index[l] + (n * KEYS_PER_NODE);
101}
102
103/*
104 * Return the highest key that you could lookup from the n'th
105 * node on level l of the btree.
106 */
107static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
108{
109 for (; l < t->depth - 1; l++)
110 n = get_child(n, CHILDREN_PER_NODE - 1);
111
112 if (n >= t->counts[l])
113 return (sector_t) - 1;
114
115 return get_node(t, l, n)[KEYS_PER_NODE - 1];
116}
117
118/*
119 * Fills in a level of the btree based on the highs of the level
120 * below it.
121 */
122static int setup_btree_index(unsigned int l, struct dm_table *t)
123{
124 unsigned int n, k;
125 sector_t *node;
126
127 for (n = 0U; n < t->counts[l]; n++) {
128 node = get_node(t, l, n);
129
130 for (k = 0U; k < KEYS_PER_NODE; k++)
131 node[k] = high(t, l + 1, get_child(n, k));
132 }
133
134 return 0;
135}
136
137void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
138{
139 unsigned long size;
140 void *addr;
141
142 /*
143 * Check that we're not going to overflow.
144 */
145 if (nmemb > (ULONG_MAX / elem_size))
146 return NULL;
147
148 size = nmemb * elem_size;
149 addr = vzalloc(size);
150
151 return addr;
152}
153EXPORT_SYMBOL(dm_vcalloc);
154
155/*
156 * highs, and targets are managed as dynamic arrays during a
157 * table load.
158 */
159static int alloc_targets(struct dm_table *t, unsigned int num)
160{
161 sector_t *n_highs;
162 struct dm_target *n_targets;
163
164 /*
165 * Allocate both the target array and offset array at once.
166 */
167 n_highs = (sector_t *) dm_vcalloc(num, sizeof(struct dm_target) +
168 sizeof(sector_t));
169 if (!n_highs)
170 return -ENOMEM;
171
172 n_targets = (struct dm_target *) (n_highs + num);
173
174 memset(n_highs, -1, sizeof(*n_highs) * num);
175 vfree(t->highs);
176
177 t->num_allocated = num;
178 t->highs = n_highs;
179 t->targets = n_targets;
180
181 return 0;
182}
183
184int dm_table_create(struct dm_table **result, fmode_t mode,
185 unsigned num_targets, struct mapped_device *md)
186{
187 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
188
189 if (!t)
190 return -ENOMEM;
191
192 INIT_LIST_HEAD(&t->devices);
193 INIT_LIST_HEAD(&t->target_callbacks);
194
195 if (!num_targets)
196 num_targets = KEYS_PER_NODE;
197
198 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
199
200 if (!num_targets) {
201 kfree(t);
202 return -ENOMEM;
203 }
204
205 if (alloc_targets(t, num_targets)) {
206 kfree(t);
207 return -ENOMEM;
208 }
209
210 t->type = DM_TYPE_NONE;
211 t->mode = mode;
212 t->md = md;
213 *result = t;
214 return 0;
215}
216
217static void free_devices(struct list_head *devices, struct mapped_device *md)
218{
219 struct list_head *tmp, *next;
220
221 list_for_each_safe(tmp, next, devices) {
222 struct dm_dev_internal *dd =
223 list_entry(tmp, struct dm_dev_internal, list);
224 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
225 dm_device_name(md), dd->dm_dev->name);
226 dm_put_table_device(md, dd->dm_dev);
227 kfree(dd);
228 }
229}
230
231void dm_table_destroy(struct dm_table *t)
232{
233 unsigned int i;
234
235 if (!t)
236 return;
237
238 /* free the indexes */
239 if (t->depth >= 2)
240 vfree(t->index[t->depth - 2]);
241
242 /* free the targets */
243 for (i = 0; i < t->num_targets; i++) {
244 struct dm_target *tgt = t->targets + i;
245
246 if (tgt->type->dtr)
247 tgt->type->dtr(tgt);
248
249 dm_put_target_type(tgt->type);
250 }
251
252 vfree(t->highs);
253
254 /* free the device list */
255 free_devices(&t->devices, t->md);
256
257 dm_free_md_mempools(t->mempools);
258
259 kfree(t);
260}
261
262/*
263 * See if we've already got a device in the list.
264 */
265static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
266{
267 struct dm_dev_internal *dd;
268
269 list_for_each_entry (dd, l, list)
270 if (dd->dm_dev->bdev->bd_dev == dev)
271 return dd;
272
273 return NULL;
274}
275
276/*
277 * If possible, this checks an area of a destination device is invalid.
278 */
279static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
280 sector_t start, sector_t len, void *data)
281{
282 struct request_queue *q;
283 struct queue_limits *limits = data;
284 struct block_device *bdev = dev->bdev;
285 sector_t dev_size =
286 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
287 unsigned short logical_block_size_sectors =
288 limits->logical_block_size >> SECTOR_SHIFT;
289 char b[BDEVNAME_SIZE];
290
291 /*
292 * Some devices exist without request functions,
293 * such as loop devices not yet bound to backing files.
294 * Forbid the use of such devices.
295 */
296 q = bdev_get_queue(bdev);
297 if (!q || !q->make_request_fn) {
298 DMWARN("%s: %s is not yet initialised: "
299 "start=%llu, len=%llu, dev_size=%llu",
300 dm_device_name(ti->table->md), bdevname(bdev, b),
301 (unsigned long long)start,
302 (unsigned long long)len,
303 (unsigned long long)dev_size);
304 return 1;
305 }
306
307 if (!dev_size)
308 return 0;
309
310 if ((start >= dev_size) || (start + len > dev_size)) {
311 DMWARN("%s: %s too small for target: "
312 "start=%llu, len=%llu, dev_size=%llu",
313 dm_device_name(ti->table->md), bdevname(bdev, b),
314 (unsigned long long)start,
315 (unsigned long long)len,
316 (unsigned long long)dev_size);
317 return 1;
318 }
319
320 /*
321 * If the target is mapped to zoned block device(s), check
322 * that the zones are not partially mapped.
323 */
324 if (bdev_zoned_model(bdev) != BLK_ZONED_NONE) {
325 unsigned int zone_sectors = bdev_zone_sectors(bdev);
326
327 if (start & (zone_sectors - 1)) {
328 DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s",
329 dm_device_name(ti->table->md),
330 (unsigned long long)start,
331 zone_sectors, bdevname(bdev, b));
332 return 1;
333 }
334
335 /*
336 * Note: The last zone of a zoned block device may be smaller
337 * than other zones. So for a target mapping the end of a
338 * zoned block device with such a zone, len would not be zone
339 * aligned. We do not allow such last smaller zone to be part
340 * of the mapping here to ensure that mappings with multiple
341 * devices do not end up with a smaller zone in the middle of
342 * the sector range.
343 */
344 if (len & (zone_sectors - 1)) {
345 DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s",
346 dm_device_name(ti->table->md),
347 (unsigned long long)len,
348 zone_sectors, bdevname(bdev, b));
349 return 1;
350 }
351 }
352
353 if (logical_block_size_sectors <= 1)
354 return 0;
355
356 if (start & (logical_block_size_sectors - 1)) {
357 DMWARN("%s: start=%llu not aligned to h/w "
358 "logical block size %u of %s",
359 dm_device_name(ti->table->md),
360 (unsigned long long)start,
361 limits->logical_block_size, bdevname(bdev, b));
362 return 1;
363 }
364
365 if (len & (logical_block_size_sectors - 1)) {
366 DMWARN("%s: len=%llu not aligned to h/w "
367 "logical block size %u of %s",
368 dm_device_name(ti->table->md),
369 (unsigned long long)len,
370 limits->logical_block_size, bdevname(bdev, b));
371 return 1;
372 }
373
374 return 0;
375}
376
377/*
378 * This upgrades the mode on an already open dm_dev, being
379 * careful to leave things as they were if we fail to reopen the
380 * device and not to touch the existing bdev field in case
381 * it is accessed concurrently inside dm_table_any_congested().
382 */
383static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
384 struct mapped_device *md)
385{
386 int r;
387 struct dm_dev *old_dev, *new_dev;
388
389 old_dev = dd->dm_dev;
390
391 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
392 dd->dm_dev->mode | new_mode, &new_dev);
393 if (r)
394 return r;
395
396 dd->dm_dev = new_dev;
397 dm_put_table_device(md, old_dev);
398
399 return 0;
400}
401
402/*
403 * Convert the path to a device
404 */
405dev_t dm_get_dev_t(const char *path)
406{
407 dev_t dev;
408 struct block_device *bdev;
409
410 bdev = lookup_bdev(path);
411 if (IS_ERR(bdev))
412 dev = name_to_dev_t(path);
413 else {
414 dev = bdev->bd_dev;
415 bdput(bdev);
416 }
417
418 return dev;
419}
420EXPORT_SYMBOL_GPL(dm_get_dev_t);
421
422/*
423 * Add a device to the list, or just increment the usage count if
424 * it's already present.
425 */
426int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
427 struct dm_dev **result)
428{
429 int r;
430 dev_t dev;
431 struct dm_dev_internal *dd;
432 struct dm_table *t = ti->table;
433
434 BUG_ON(!t);
435
436 dev = dm_get_dev_t(path);
437 if (!dev)
438 return -ENODEV;
439
440 dd = find_device(&t->devices, dev);
441 if (!dd) {
442 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
443 if (!dd)
444 return -ENOMEM;
445
446 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
447 kfree(dd);
448 return r;
449 }
450
451 refcount_set(&dd->count, 1);
452 list_add(&dd->list, &t->devices);
453 goto out;
454
455 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
456 r = upgrade_mode(dd, mode, t->md);
457 if (r)
458 return r;
459 }
460 refcount_inc(&dd->count);
461out:
462 *result = dd->dm_dev;
463 return 0;
464}
465EXPORT_SYMBOL(dm_get_device);
466
467static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
468 sector_t start, sector_t len, void *data)
469{
470 struct queue_limits *limits = data;
471 struct block_device *bdev = dev->bdev;
472 struct request_queue *q = bdev_get_queue(bdev);
473 char b[BDEVNAME_SIZE];
474
475 if (unlikely(!q)) {
476 DMWARN("%s: Cannot set limits for nonexistent device %s",
477 dm_device_name(ti->table->md), bdevname(bdev, b));
478 return 0;
479 }
480
481 if (bdev_stack_limits(limits, bdev, start) < 0)
482 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
483 "physical_block_size=%u, logical_block_size=%u, "
484 "alignment_offset=%u, start=%llu",
485 dm_device_name(ti->table->md), bdevname(bdev, b),
486 q->limits.physical_block_size,
487 q->limits.logical_block_size,
488 q->limits.alignment_offset,
489 (unsigned long long) start << SECTOR_SHIFT);
490
491 limits->zoned = blk_queue_zoned_model(q);
492
493 return 0;
494}
495
496/*
497 * Decrement a device's use count and remove it if necessary.
498 */
499void dm_put_device(struct dm_target *ti, struct dm_dev *d)
500{
501 int found = 0;
502 struct list_head *devices = &ti->table->devices;
503 struct dm_dev_internal *dd;
504
505 list_for_each_entry(dd, devices, list) {
506 if (dd->dm_dev == d) {
507 found = 1;
508 break;
509 }
510 }
511 if (!found) {
512 DMWARN("%s: device %s not in table devices list",
513 dm_device_name(ti->table->md), d->name);
514 return;
515 }
516 if (refcount_dec_and_test(&dd->count)) {
517 dm_put_table_device(ti->table->md, d);
518 list_del(&dd->list);
519 kfree(dd);
520 }
521}
522EXPORT_SYMBOL(dm_put_device);
523
524/*
525 * Checks to see if the target joins onto the end of the table.
526 */
527static int adjoin(struct dm_table *table, struct dm_target *ti)
528{
529 struct dm_target *prev;
530
531 if (!table->num_targets)
532 return !ti->begin;
533
534 prev = &table->targets[table->num_targets - 1];
535 return (ti->begin == (prev->begin + prev->len));
536}
537
538/*
539 * Used to dynamically allocate the arg array.
540 *
541 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
542 * process messages even if some device is suspended. These messages have a
543 * small fixed number of arguments.
544 *
545 * On the other hand, dm-switch needs to process bulk data using messages and
546 * excessive use of GFP_NOIO could cause trouble.
547 */
548static char **realloc_argv(unsigned *size, char **old_argv)
549{
550 char **argv;
551 unsigned new_size;
552 gfp_t gfp;
553
554 if (*size) {
555 new_size = *size * 2;
556 gfp = GFP_KERNEL;
557 } else {
558 new_size = 8;
559 gfp = GFP_NOIO;
560 }
561 argv = kmalloc_array(new_size, sizeof(*argv), gfp);
562 if (argv && old_argv) {
563 memcpy(argv, old_argv, *size * sizeof(*argv));
564 *size = new_size;
565 }
566
567 kfree(old_argv);
568 return argv;
569}
570
571/*
572 * Destructively splits up the argument list to pass to ctr.
573 */
574int dm_split_args(int *argc, char ***argvp, char *input)
575{
576 char *start, *end = input, *out, **argv = NULL;
577 unsigned array_size = 0;
578
579 *argc = 0;
580
581 if (!input) {
582 *argvp = NULL;
583 return 0;
584 }
585
586 argv = realloc_argv(&array_size, argv);
587 if (!argv)
588 return -ENOMEM;
589
590 while (1) {
591 /* Skip whitespace */
592 start = skip_spaces(end);
593
594 if (!*start)
595 break; /* success, we hit the end */
596
597 /* 'out' is used to remove any back-quotes */
598 end = out = start;
599 while (*end) {
600 /* Everything apart from '\0' can be quoted */
601 if (*end == '\\' && *(end + 1)) {
602 *out++ = *(end + 1);
603 end += 2;
604 continue;
605 }
606
607 if (isspace(*end))
608 break; /* end of token */
609
610 *out++ = *end++;
611 }
612
613 /* have we already filled the array ? */
614 if ((*argc + 1) > array_size) {
615 argv = realloc_argv(&array_size, argv);
616 if (!argv)
617 return -ENOMEM;
618 }
619
620 /* we know this is whitespace */
621 if (*end)
622 end++;
623
624 /* terminate the string and put it in the array */
625 *out = '\0';
626 argv[*argc] = start;
627 (*argc)++;
628 }
629
630 *argvp = argv;
631 return 0;
632}
633
634/*
635 * Impose necessary and sufficient conditions on a devices's table such
636 * that any incoming bio which respects its logical_block_size can be
637 * processed successfully. If it falls across the boundary between
638 * two or more targets, the size of each piece it gets split into must
639 * be compatible with the logical_block_size of the target processing it.
640 */
641static int validate_hardware_logical_block_alignment(struct dm_table *table,
642 struct queue_limits *limits)
643{
644 /*
645 * This function uses arithmetic modulo the logical_block_size
646 * (in units of 512-byte sectors).
647 */
648 unsigned short device_logical_block_size_sects =
649 limits->logical_block_size >> SECTOR_SHIFT;
650
651 /*
652 * Offset of the start of the next table entry, mod logical_block_size.
653 */
654 unsigned short next_target_start = 0;
655
656 /*
657 * Given an aligned bio that extends beyond the end of a
658 * target, how many sectors must the next target handle?
659 */
660 unsigned short remaining = 0;
661
662 struct dm_target *uninitialized_var(ti);
663 struct queue_limits ti_limits;
664 unsigned i;
665
666 /*
667 * Check each entry in the table in turn.
668 */
669 for (i = 0; i < dm_table_get_num_targets(table); i++) {
670 ti = dm_table_get_target(table, i);
671
672 blk_set_stacking_limits(&ti_limits);
673
674 /* combine all target devices' limits */
675 if (ti->type->iterate_devices)
676 ti->type->iterate_devices(ti, dm_set_device_limits,
677 &ti_limits);
678
679 /*
680 * If the remaining sectors fall entirely within this
681 * table entry are they compatible with its logical_block_size?
682 */
683 if (remaining < ti->len &&
684 remaining & ((ti_limits.logical_block_size >>
685 SECTOR_SHIFT) - 1))
686 break; /* Error */
687
688 next_target_start =
689 (unsigned short) ((next_target_start + ti->len) &
690 (device_logical_block_size_sects - 1));
691 remaining = next_target_start ?
692 device_logical_block_size_sects - next_target_start : 0;
693 }
694
695 if (remaining) {
696 DMWARN("%s: table line %u (start sect %llu len %llu) "
697 "not aligned to h/w logical block size %u",
698 dm_device_name(table->md), i,
699 (unsigned long long) ti->begin,
700 (unsigned long long) ti->len,
701 limits->logical_block_size);
702 return -EINVAL;
703 }
704
705 return 0;
706}
707
708int dm_table_add_target(struct dm_table *t, const char *type,
709 sector_t start, sector_t len, char *params)
710{
711 int r = -EINVAL, argc;
712 char **argv;
713 struct dm_target *tgt;
714
715 if (t->singleton) {
716 DMERR("%s: target type %s must appear alone in table",
717 dm_device_name(t->md), t->targets->type->name);
718 return -EINVAL;
719 }
720
721 BUG_ON(t->num_targets >= t->num_allocated);
722
723 tgt = t->targets + t->num_targets;
724 memset(tgt, 0, sizeof(*tgt));
725
726 if (!len) {
727 DMERR("%s: zero-length target", dm_device_name(t->md));
728 return -EINVAL;
729 }
730
731 tgt->type = dm_get_target_type(type);
732 if (!tgt->type) {
733 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
734 return -EINVAL;
735 }
736
737 if (dm_target_needs_singleton(tgt->type)) {
738 if (t->num_targets) {
739 tgt->error = "singleton target type must appear alone in table";
740 goto bad;
741 }
742 t->singleton = true;
743 }
744
745 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
746 tgt->error = "target type may not be included in a read-only table";
747 goto bad;
748 }
749
750 if (t->immutable_target_type) {
751 if (t->immutable_target_type != tgt->type) {
752 tgt->error = "immutable target type cannot be mixed with other target types";
753 goto bad;
754 }
755 } else if (dm_target_is_immutable(tgt->type)) {
756 if (t->num_targets) {
757 tgt->error = "immutable target type cannot be mixed with other target types";
758 goto bad;
759 }
760 t->immutable_target_type = tgt->type;
761 }
762
763 if (dm_target_has_integrity(tgt->type))
764 t->integrity_added = 1;
765
766 tgt->table = t;
767 tgt->begin = start;
768 tgt->len = len;
769 tgt->error = "Unknown error";
770
771 /*
772 * Does this target adjoin the previous one ?
773 */
774 if (!adjoin(t, tgt)) {
775 tgt->error = "Gap in table";
776 goto bad;
777 }
778
779 r = dm_split_args(&argc, &argv, params);
780 if (r) {
781 tgt->error = "couldn't split parameters (insufficient memory)";
782 goto bad;
783 }
784
785 r = tgt->type->ctr(tgt, argc, argv);
786 kfree(argv);
787 if (r)
788 goto bad;
789
790 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
791
792 if (!tgt->num_discard_bios && tgt->discards_supported)
793 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
794 dm_device_name(t->md), type);
795
796 return 0;
797
798 bad:
799 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
800 dm_put_target_type(tgt->type);
801 return r;
802}
803
804/*
805 * Target argument parsing helpers.
806 */
807static int validate_next_arg(const struct dm_arg *arg,
808 struct dm_arg_set *arg_set,
809 unsigned *value, char **error, unsigned grouped)
810{
811 const char *arg_str = dm_shift_arg(arg_set);
812 char dummy;
813
814 if (!arg_str ||
815 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
816 (*value < arg->min) ||
817 (*value > arg->max) ||
818 (grouped && arg_set->argc < *value)) {
819 *error = arg->error;
820 return -EINVAL;
821 }
822
823 return 0;
824}
825
826int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
827 unsigned *value, char **error)
828{
829 return validate_next_arg(arg, arg_set, value, error, 0);
830}
831EXPORT_SYMBOL(dm_read_arg);
832
833int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
834 unsigned *value, char **error)
835{
836 return validate_next_arg(arg, arg_set, value, error, 1);
837}
838EXPORT_SYMBOL(dm_read_arg_group);
839
840const char *dm_shift_arg(struct dm_arg_set *as)
841{
842 char *r;
843
844 if (as->argc) {
845 as->argc--;
846 r = *as->argv;
847 as->argv++;
848 return r;
849 }
850
851 return NULL;
852}
853EXPORT_SYMBOL(dm_shift_arg);
854
855void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
856{
857 BUG_ON(as->argc < num_args);
858 as->argc -= num_args;
859 as->argv += num_args;
860}
861EXPORT_SYMBOL(dm_consume_args);
862
863static bool __table_type_bio_based(enum dm_queue_mode table_type)
864{
865 return (table_type == DM_TYPE_BIO_BASED ||
866 table_type == DM_TYPE_DAX_BIO_BASED ||
867 table_type == DM_TYPE_NVME_BIO_BASED);
868}
869
870static bool __table_type_request_based(enum dm_queue_mode table_type)
871{
872 return table_type == DM_TYPE_REQUEST_BASED;
873}
874
875void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
876{
877 t->type = type;
878}
879EXPORT_SYMBOL_GPL(dm_table_set_type);
880
881/* validate the dax capability of the target device span */
882int device_supports_dax(struct dm_target *ti, struct dm_dev *dev,
883 sector_t start, sector_t len, void *data)
884{
885 int blocksize = *(int *) data;
886
887 return generic_fsdax_supported(dev->dax_dev, dev->bdev, blocksize,
888 start, len);
889}
890
891/* Check devices support synchronous DAX */
892static int device_dax_synchronous(struct dm_target *ti, struct dm_dev *dev,
893 sector_t start, sector_t len, void *data)
894{
895 return dev->dax_dev && dax_synchronous(dev->dax_dev);
896}
897
898bool dm_table_supports_dax(struct dm_table *t,
899 iterate_devices_callout_fn iterate_fn, int *blocksize)
900{
901 struct dm_target *ti;
902 unsigned i;
903
904 /* Ensure that all targets support DAX. */
905 for (i = 0; i < dm_table_get_num_targets(t); i++) {
906 ti = dm_table_get_target(t, i);
907
908 if (!ti->type->direct_access)
909 return false;
910
911 if (!ti->type->iterate_devices ||
912 !ti->type->iterate_devices(ti, iterate_fn, blocksize))
913 return false;
914 }
915
916 return true;
917}
918
919static bool dm_table_does_not_support_partial_completion(struct dm_table *t);
920
921struct verify_rq_based_data {
922 unsigned sq_count;
923 unsigned mq_count;
924};
925
926static int device_is_rq_based(struct dm_target *ti, struct dm_dev *dev,
927 sector_t start, sector_t len, void *data)
928{
929 struct request_queue *q = bdev_get_queue(dev->bdev);
930 struct verify_rq_based_data *v = data;
931
932 if (queue_is_mq(q))
933 v->mq_count++;
934 else
935 v->sq_count++;
936
937 return queue_is_mq(q);
938}
939
940static int dm_table_determine_type(struct dm_table *t)
941{
942 unsigned i;
943 unsigned bio_based = 0, request_based = 0, hybrid = 0;
944 struct verify_rq_based_data v = {.sq_count = 0, .mq_count = 0};
945 struct dm_target *tgt;
946 struct list_head *devices = dm_table_get_devices(t);
947 enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
948 int page_size = PAGE_SIZE;
949
950 if (t->type != DM_TYPE_NONE) {
951 /* target already set the table's type */
952 if (t->type == DM_TYPE_BIO_BASED) {
953 /* possibly upgrade to a variant of bio-based */
954 goto verify_bio_based;
955 }
956 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
957 BUG_ON(t->type == DM_TYPE_NVME_BIO_BASED);
958 goto verify_rq_based;
959 }
960
961 for (i = 0; i < t->num_targets; i++) {
962 tgt = t->targets + i;
963 if (dm_target_hybrid(tgt))
964 hybrid = 1;
965 else if (dm_target_request_based(tgt))
966 request_based = 1;
967 else
968 bio_based = 1;
969
970 if (bio_based && request_based) {
971 DMERR("Inconsistent table: different target types"
972 " can't be mixed up");
973 return -EINVAL;
974 }
975 }
976
977 if (hybrid && !bio_based && !request_based) {
978 /*
979 * The targets can work either way.
980 * Determine the type from the live device.
981 * Default to bio-based if device is new.
982 */
983 if (__table_type_request_based(live_md_type))
984 request_based = 1;
985 else
986 bio_based = 1;
987 }
988
989 if (bio_based) {
990verify_bio_based:
991 /* We must use this table as bio-based */
992 t->type = DM_TYPE_BIO_BASED;
993 if (dm_table_supports_dax(t, device_supports_dax, &page_size) ||
994 (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
995 t->type = DM_TYPE_DAX_BIO_BASED;
996 } else {
997 /* Check if upgrading to NVMe bio-based is valid or required */
998 tgt = dm_table_get_immutable_target(t);
999 if (tgt && !tgt->max_io_len && dm_table_does_not_support_partial_completion(t)) {
1000 t->type = DM_TYPE_NVME_BIO_BASED;
1001 goto verify_rq_based; /* must be stacked directly on NVMe (blk-mq) */
1002 } else if (list_empty(devices) && live_md_type == DM_TYPE_NVME_BIO_BASED) {
1003 t->type = DM_TYPE_NVME_BIO_BASED;
1004 }
1005 }
1006 return 0;
1007 }
1008
1009 BUG_ON(!request_based); /* No targets in this table */
1010
1011 t->type = DM_TYPE_REQUEST_BASED;
1012
1013verify_rq_based:
1014 /*
1015 * Request-based dm supports only tables that have a single target now.
1016 * To support multiple targets, request splitting support is needed,
1017 * and that needs lots of changes in the block-layer.
1018 * (e.g. request completion process for partial completion.)
1019 */
1020 if (t->num_targets > 1) {
1021 DMERR("%s DM doesn't support multiple targets",
1022 t->type == DM_TYPE_NVME_BIO_BASED ? "nvme bio-based" : "request-based");
1023 return -EINVAL;
1024 }
1025
1026 if (list_empty(devices)) {
1027 int srcu_idx;
1028 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
1029
1030 /* inherit live table's type */
1031 if (live_table)
1032 t->type = live_table->type;
1033 dm_put_live_table(t->md, srcu_idx);
1034 return 0;
1035 }
1036
1037 tgt = dm_table_get_immutable_target(t);
1038 if (!tgt) {
1039 DMERR("table load rejected: immutable target is required");
1040 return -EINVAL;
1041 } else if (tgt->max_io_len) {
1042 DMERR("table load rejected: immutable target that splits IO is not supported");
1043 return -EINVAL;
1044 }
1045
1046 /* Non-request-stackable devices can't be used for request-based dm */
1047 if (!tgt->type->iterate_devices ||
1048 !tgt->type->iterate_devices(tgt, device_is_rq_based, &v)) {
1049 DMERR("table load rejected: including non-request-stackable devices");
1050 return -EINVAL;
1051 }
1052 if (v.sq_count > 0) {
1053 DMERR("table load rejected: not all devices are blk-mq request-stackable");
1054 return -EINVAL;
1055 }
1056
1057 return 0;
1058}
1059
1060enum dm_queue_mode dm_table_get_type(struct dm_table *t)
1061{
1062 return t->type;
1063}
1064
1065struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
1066{
1067 return t->immutable_target_type;
1068}
1069
1070struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
1071{
1072 /* Immutable target is implicitly a singleton */
1073 if (t->num_targets > 1 ||
1074 !dm_target_is_immutable(t->targets[0].type))
1075 return NULL;
1076
1077 return t->targets;
1078}
1079
1080struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1081{
1082 struct dm_target *ti;
1083 unsigned i;
1084
1085 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1086 ti = dm_table_get_target(t, i);
1087 if (dm_target_is_wildcard(ti->type))
1088 return ti;
1089 }
1090
1091 return NULL;
1092}
1093
1094bool dm_table_bio_based(struct dm_table *t)
1095{
1096 return __table_type_bio_based(dm_table_get_type(t));
1097}
1098
1099bool dm_table_request_based(struct dm_table *t)
1100{
1101 return __table_type_request_based(dm_table_get_type(t));
1102}
1103
1104static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1105{
1106 enum dm_queue_mode type = dm_table_get_type(t);
1107 unsigned per_io_data_size = 0;
1108 unsigned min_pool_size = 0;
1109 struct dm_target *ti;
1110 unsigned i;
1111
1112 if (unlikely(type == DM_TYPE_NONE)) {
1113 DMWARN("no table type is set, can't allocate mempools");
1114 return -EINVAL;
1115 }
1116
1117 if (__table_type_bio_based(type))
1118 for (i = 0; i < t->num_targets; i++) {
1119 ti = t->targets + i;
1120 per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1121 min_pool_size = max(min_pool_size, ti->num_flush_bios);
1122 }
1123
1124 t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1125 per_io_data_size, min_pool_size);
1126 if (!t->mempools)
1127 return -ENOMEM;
1128
1129 return 0;
1130}
1131
1132void dm_table_free_md_mempools(struct dm_table *t)
1133{
1134 dm_free_md_mempools(t->mempools);
1135 t->mempools = NULL;
1136}
1137
1138struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1139{
1140 return t->mempools;
1141}
1142
1143static int setup_indexes(struct dm_table *t)
1144{
1145 int i;
1146 unsigned int total = 0;
1147 sector_t *indexes;
1148
1149 /* allocate the space for *all* the indexes */
1150 for (i = t->depth - 2; i >= 0; i--) {
1151 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1152 total += t->counts[i];
1153 }
1154
1155 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1156 if (!indexes)
1157 return -ENOMEM;
1158
1159 /* set up internal nodes, bottom-up */
1160 for (i = t->depth - 2; i >= 0; i--) {
1161 t->index[i] = indexes;
1162 indexes += (KEYS_PER_NODE * t->counts[i]);
1163 setup_btree_index(i, t);
1164 }
1165
1166 return 0;
1167}
1168
1169/*
1170 * Builds the btree to index the map.
1171 */
1172static int dm_table_build_index(struct dm_table *t)
1173{
1174 int r = 0;
1175 unsigned int leaf_nodes;
1176
1177 /* how many indexes will the btree have ? */
1178 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1179 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1180
1181 /* leaf layer has already been set up */
1182 t->counts[t->depth - 1] = leaf_nodes;
1183 t->index[t->depth - 1] = t->highs;
1184
1185 if (t->depth >= 2)
1186 r = setup_indexes(t);
1187
1188 return r;
1189}
1190
1191static bool integrity_profile_exists(struct gendisk *disk)
1192{
1193 return !!blk_get_integrity(disk);
1194}
1195
1196/*
1197 * Get a disk whose integrity profile reflects the table's profile.
1198 * Returns NULL if integrity support was inconsistent or unavailable.
1199 */
1200static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1201{
1202 struct list_head *devices = dm_table_get_devices(t);
1203 struct dm_dev_internal *dd = NULL;
1204 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1205 unsigned i;
1206
1207 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1208 struct dm_target *ti = dm_table_get_target(t, i);
1209 if (!dm_target_passes_integrity(ti->type))
1210 goto no_integrity;
1211 }
1212
1213 list_for_each_entry(dd, devices, list) {
1214 template_disk = dd->dm_dev->bdev->bd_disk;
1215 if (!integrity_profile_exists(template_disk))
1216 goto no_integrity;
1217 else if (prev_disk &&
1218 blk_integrity_compare(prev_disk, template_disk) < 0)
1219 goto no_integrity;
1220 prev_disk = template_disk;
1221 }
1222
1223 return template_disk;
1224
1225no_integrity:
1226 if (prev_disk)
1227 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1228 dm_device_name(t->md),
1229 prev_disk->disk_name,
1230 template_disk->disk_name);
1231 return NULL;
1232}
1233
1234/*
1235 * Register the mapped device for blk_integrity support if the
1236 * underlying devices have an integrity profile. But all devices may
1237 * not have matching profiles (checking all devices isn't reliable
1238 * during table load because this table may use other DM device(s) which
1239 * must be resumed before they will have an initialized integity
1240 * profile). Consequently, stacked DM devices force a 2 stage integrity
1241 * profile validation: First pass during table load, final pass during
1242 * resume.
1243 */
1244static int dm_table_register_integrity(struct dm_table *t)
1245{
1246 struct mapped_device *md = t->md;
1247 struct gendisk *template_disk = NULL;
1248
1249 /* If target handles integrity itself do not register it here. */
1250 if (t->integrity_added)
1251 return 0;
1252
1253 template_disk = dm_table_get_integrity_disk(t);
1254 if (!template_disk)
1255 return 0;
1256
1257 if (!integrity_profile_exists(dm_disk(md))) {
1258 t->integrity_supported = true;
1259 /*
1260 * Register integrity profile during table load; we can do
1261 * this because the final profile must match during resume.
1262 */
1263 blk_integrity_register(dm_disk(md),
1264 blk_get_integrity(template_disk));
1265 return 0;
1266 }
1267
1268 /*
1269 * If DM device already has an initialized integrity
1270 * profile the new profile should not conflict.
1271 */
1272 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1273 DMWARN("%s: conflict with existing integrity profile: "
1274 "%s profile mismatch",
1275 dm_device_name(t->md),
1276 template_disk->disk_name);
1277 return 1;
1278 }
1279
1280 /* Preserve existing integrity profile */
1281 t->integrity_supported = true;
1282 return 0;
1283}
1284
1285/*
1286 * Prepares the table for use by building the indices,
1287 * setting the type, and allocating mempools.
1288 */
1289int dm_table_complete(struct dm_table *t)
1290{
1291 int r;
1292
1293 r = dm_table_determine_type(t);
1294 if (r) {
1295 DMERR("unable to determine table type");
1296 return r;
1297 }
1298
1299 r = dm_table_build_index(t);
1300 if (r) {
1301 DMERR("unable to build btrees");
1302 return r;
1303 }
1304
1305 r = dm_table_register_integrity(t);
1306 if (r) {
1307 DMERR("could not register integrity profile.");
1308 return r;
1309 }
1310
1311 r = dm_table_alloc_md_mempools(t, t->md);
1312 if (r)
1313 DMERR("unable to allocate mempools");
1314
1315 return r;
1316}
1317
1318static DEFINE_MUTEX(_event_lock);
1319void dm_table_event_callback(struct dm_table *t,
1320 void (*fn)(void *), void *context)
1321{
1322 mutex_lock(&_event_lock);
1323 t->event_fn = fn;
1324 t->event_context = context;
1325 mutex_unlock(&_event_lock);
1326}
1327
1328void dm_table_event(struct dm_table *t)
1329{
1330 /*
1331 * You can no longer call dm_table_event() from interrupt
1332 * context, use a bottom half instead.
1333 */
1334 BUG_ON(in_interrupt());
1335
1336 mutex_lock(&_event_lock);
1337 if (t->event_fn)
1338 t->event_fn(t->event_context);
1339 mutex_unlock(&_event_lock);
1340}
1341EXPORT_SYMBOL(dm_table_event);
1342
1343inline sector_t dm_table_get_size(struct dm_table *t)
1344{
1345 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1346}
1347EXPORT_SYMBOL(dm_table_get_size);
1348
1349struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1350{
1351 if (index >= t->num_targets)
1352 return NULL;
1353
1354 return t->targets + index;
1355}
1356
1357/*
1358 * Search the btree for the correct target.
1359 *
1360 * Caller should check returned pointer for NULL
1361 * to trap I/O beyond end of device.
1362 */
1363struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1364{
1365 unsigned int l, n = 0, k = 0;
1366 sector_t *node;
1367
1368 if (unlikely(sector >= dm_table_get_size(t)))
1369 return NULL;
1370
1371 for (l = 0; l < t->depth; l++) {
1372 n = get_child(n, k);
1373 node = get_node(t, l, n);
1374
1375 for (k = 0; k < KEYS_PER_NODE; k++)
1376 if (node[k] >= sector)
1377 break;
1378 }
1379
1380 return &t->targets[(KEYS_PER_NODE * n) + k];
1381}
1382
1383static int count_device(struct dm_target *ti, struct dm_dev *dev,
1384 sector_t start, sector_t len, void *data)
1385{
1386 unsigned *num_devices = data;
1387
1388 (*num_devices)++;
1389
1390 return 0;
1391}
1392
1393/*
1394 * Check whether a table has no data devices attached using each
1395 * target's iterate_devices method.
1396 * Returns false if the result is unknown because a target doesn't
1397 * support iterate_devices.
1398 */
1399bool dm_table_has_no_data_devices(struct dm_table *table)
1400{
1401 struct dm_target *ti;
1402 unsigned i, num_devices;
1403
1404 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1405 ti = dm_table_get_target(table, i);
1406
1407 if (!ti->type->iterate_devices)
1408 return false;
1409
1410 num_devices = 0;
1411 ti->type->iterate_devices(ti, count_device, &num_devices);
1412 if (num_devices)
1413 return false;
1414 }
1415
1416 return true;
1417}
1418
1419static int device_is_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1420 sector_t start, sector_t len, void *data)
1421{
1422 struct request_queue *q = bdev_get_queue(dev->bdev);
1423 enum blk_zoned_model *zoned_model = data;
1424
1425 return q && blk_queue_zoned_model(q) == *zoned_model;
1426}
1427
1428static bool dm_table_supports_zoned_model(struct dm_table *t,
1429 enum blk_zoned_model zoned_model)
1430{
1431 struct dm_target *ti;
1432 unsigned i;
1433
1434 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1435 ti = dm_table_get_target(t, i);
1436
1437 if (zoned_model == BLK_ZONED_HM &&
1438 !dm_target_supports_zoned_hm(ti->type))
1439 return false;
1440
1441 if (!ti->type->iterate_devices ||
1442 !ti->type->iterate_devices(ti, device_is_zoned_model, &zoned_model))
1443 return false;
1444 }
1445
1446 return true;
1447}
1448
1449static int device_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1450 sector_t start, sector_t len, void *data)
1451{
1452 struct request_queue *q = bdev_get_queue(dev->bdev);
1453 unsigned int *zone_sectors = data;
1454
1455 return q && blk_queue_zone_sectors(q) == *zone_sectors;
1456}
1457
1458static bool dm_table_matches_zone_sectors(struct dm_table *t,
1459 unsigned int zone_sectors)
1460{
1461 struct dm_target *ti;
1462 unsigned i;
1463
1464 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1465 ti = dm_table_get_target(t, i);
1466
1467 if (!ti->type->iterate_devices ||
1468 !ti->type->iterate_devices(ti, device_matches_zone_sectors, &zone_sectors))
1469 return false;
1470 }
1471
1472 return true;
1473}
1474
1475static int validate_hardware_zoned_model(struct dm_table *table,
1476 enum blk_zoned_model zoned_model,
1477 unsigned int zone_sectors)
1478{
1479 if (zoned_model == BLK_ZONED_NONE)
1480 return 0;
1481
1482 if (!dm_table_supports_zoned_model(table, zoned_model)) {
1483 DMERR("%s: zoned model is not consistent across all devices",
1484 dm_device_name(table->md));
1485 return -EINVAL;
1486 }
1487
1488 /* Check zone size validity and compatibility */
1489 if (!zone_sectors || !is_power_of_2(zone_sectors))
1490 return -EINVAL;
1491
1492 if (!dm_table_matches_zone_sectors(table, zone_sectors)) {
1493 DMERR("%s: zone sectors is not consistent across all devices",
1494 dm_device_name(table->md));
1495 return -EINVAL;
1496 }
1497
1498 return 0;
1499}
1500
1501/*
1502 * Establish the new table's queue_limits and validate them.
1503 */
1504int dm_calculate_queue_limits(struct dm_table *table,
1505 struct queue_limits *limits)
1506{
1507 struct dm_target *ti;
1508 struct queue_limits ti_limits;
1509 unsigned i;
1510 enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1511 unsigned int zone_sectors = 0;
1512
1513 blk_set_stacking_limits(limits);
1514
1515 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1516 blk_set_stacking_limits(&ti_limits);
1517
1518 ti = dm_table_get_target(table, i);
1519
1520 if (!ti->type->iterate_devices)
1521 goto combine_limits;
1522
1523 /*
1524 * Combine queue limits of all the devices this target uses.
1525 */
1526 ti->type->iterate_devices(ti, dm_set_device_limits,
1527 &ti_limits);
1528
1529 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1530 /*
1531 * After stacking all limits, validate all devices
1532 * in table support this zoned model and zone sectors.
1533 */
1534 zoned_model = ti_limits.zoned;
1535 zone_sectors = ti_limits.chunk_sectors;
1536 }
1537
1538 /* Set I/O hints portion of queue limits */
1539 if (ti->type->io_hints)
1540 ti->type->io_hints(ti, &ti_limits);
1541
1542 /*
1543 * Check each device area is consistent with the target's
1544 * overall queue limits.
1545 */
1546 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1547 &ti_limits))
1548 return -EINVAL;
1549
1550combine_limits:
1551 /*
1552 * Merge this target's queue limits into the overall limits
1553 * for the table.
1554 */
1555 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1556 DMWARN("%s: adding target device "
1557 "(start sect %llu len %llu) "
1558 "caused an alignment inconsistency",
1559 dm_device_name(table->md),
1560 (unsigned long long) ti->begin,
1561 (unsigned long long) ti->len);
1562
1563 /*
1564 * FIXME: this should likely be moved to blk_stack_limits(), would
1565 * also eliminate limits->zoned stacking hack in dm_set_device_limits()
1566 */
1567 if (limits->zoned == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1568 /*
1569 * By default, the stacked limits zoned model is set to
1570 * BLK_ZONED_NONE in blk_set_stacking_limits(). Update
1571 * this model using the first target model reported
1572 * that is not BLK_ZONED_NONE. This will be either the
1573 * first target device zoned model or the model reported
1574 * by the target .io_hints.
1575 */
1576 limits->zoned = ti_limits.zoned;
1577 }
1578 }
1579
1580 /*
1581 * Verify that the zoned model and zone sectors, as determined before
1582 * any .io_hints override, are the same across all devices in the table.
1583 * - this is especially relevant if .io_hints is emulating a disk-managed
1584 * zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1585 * BUT...
1586 */
1587 if (limits->zoned != BLK_ZONED_NONE) {
1588 /*
1589 * ...IF the above limits stacking determined a zoned model
1590 * validate that all of the table's devices conform to it.
1591 */
1592 zoned_model = limits->zoned;
1593 zone_sectors = limits->chunk_sectors;
1594 }
1595 if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1596 return -EINVAL;
1597
1598 return validate_hardware_logical_block_alignment(table, limits);
1599}
1600
1601/*
1602 * Verify that all devices have an integrity profile that matches the
1603 * DM device's registered integrity profile. If the profiles don't
1604 * match then unregister the DM device's integrity profile.
1605 */
1606static void dm_table_verify_integrity(struct dm_table *t)
1607{
1608 struct gendisk *template_disk = NULL;
1609
1610 if (t->integrity_added)
1611 return;
1612
1613 if (t->integrity_supported) {
1614 /*
1615 * Verify that the original integrity profile
1616 * matches all the devices in this table.
1617 */
1618 template_disk = dm_table_get_integrity_disk(t);
1619 if (template_disk &&
1620 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1621 return;
1622 }
1623
1624 if (integrity_profile_exists(dm_disk(t->md))) {
1625 DMWARN("%s: unable to establish an integrity profile",
1626 dm_device_name(t->md));
1627 blk_integrity_unregister(dm_disk(t->md));
1628 }
1629}
1630
1631static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1632 sector_t start, sector_t len, void *data)
1633{
1634 unsigned long flush = (unsigned long) data;
1635 struct request_queue *q = bdev_get_queue(dev->bdev);
1636
1637 return q && (q->queue_flags & flush);
1638}
1639
1640static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1641{
1642 struct dm_target *ti;
1643 unsigned i;
1644
1645 /*
1646 * Require at least one underlying device to support flushes.
1647 * t->devices includes internal dm devices such as mirror logs
1648 * so we need to use iterate_devices here, which targets
1649 * supporting flushes must provide.
1650 */
1651 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1652 ti = dm_table_get_target(t, i);
1653
1654 if (!ti->num_flush_bios)
1655 continue;
1656
1657 if (ti->flush_supported)
1658 return true;
1659
1660 if (ti->type->iterate_devices &&
1661 ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1662 return true;
1663 }
1664
1665 return false;
1666}
1667
1668static int device_dax_write_cache_enabled(struct dm_target *ti,
1669 struct dm_dev *dev, sector_t start,
1670 sector_t len, void *data)
1671{
1672 struct dax_device *dax_dev = dev->dax_dev;
1673
1674 if (!dax_dev)
1675 return false;
1676
1677 if (dax_write_cache_enabled(dax_dev))
1678 return true;
1679 return false;
1680}
1681
1682static int dm_table_supports_dax_write_cache(struct dm_table *t)
1683{
1684 struct dm_target *ti;
1685 unsigned i;
1686
1687 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1688 ti = dm_table_get_target(t, i);
1689
1690 if (ti->type->iterate_devices &&
1691 ti->type->iterate_devices(ti,
1692 device_dax_write_cache_enabled, NULL))
1693 return true;
1694 }
1695
1696 return false;
1697}
1698
1699static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1700 sector_t start, sector_t len, void *data)
1701{
1702 struct request_queue *q = bdev_get_queue(dev->bdev);
1703
1704 return q && blk_queue_nonrot(q);
1705}
1706
1707static int device_is_not_random(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_add_random(q);
1713}
1714
1715static bool dm_table_all_devices_attribute(struct dm_table *t,
1716 iterate_devices_callout_fn func)
1717{
1718 struct dm_target *ti;
1719 unsigned i;
1720
1721 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1722 ti = dm_table_get_target(t, i);
1723
1724 if (!ti->type->iterate_devices ||
1725 !ti->type->iterate_devices(ti, func, NULL))
1726 return false;
1727 }
1728
1729 return true;
1730}
1731
1732static int device_no_partial_completion(struct dm_target *ti, struct dm_dev *dev,
1733 sector_t start, sector_t len, void *data)
1734{
1735 char b[BDEVNAME_SIZE];
1736
1737 /* For now, NVMe devices are the only devices of this class */
1738 return (strncmp(bdevname(dev->bdev, b), "nvme", 4) == 0);
1739}
1740
1741static bool dm_table_does_not_support_partial_completion(struct dm_table *t)
1742{
1743 return dm_table_all_devices_attribute(t, device_no_partial_completion);
1744}
1745
1746static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1747 sector_t start, sector_t len, void *data)
1748{
1749 struct request_queue *q = bdev_get_queue(dev->bdev);
1750
1751 return q && !q->limits.max_write_same_sectors;
1752}
1753
1754static bool dm_table_supports_write_same(struct dm_table *t)
1755{
1756 struct dm_target *ti;
1757 unsigned i;
1758
1759 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1760 ti = dm_table_get_target(t, i);
1761
1762 if (!ti->num_write_same_bios)
1763 return false;
1764
1765 if (!ti->type->iterate_devices ||
1766 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1767 return false;
1768 }
1769
1770 return true;
1771}
1772
1773static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1774 sector_t start, sector_t len, void *data)
1775{
1776 struct request_queue *q = bdev_get_queue(dev->bdev);
1777
1778 return q && !q->limits.max_write_zeroes_sectors;
1779}
1780
1781static bool dm_table_supports_write_zeroes(struct dm_table *t)
1782{
1783 struct dm_target *ti;
1784 unsigned i = 0;
1785
1786 while (i < dm_table_get_num_targets(t)) {
1787 ti = dm_table_get_target(t, i++);
1788
1789 if (!ti->num_write_zeroes_bios)
1790 return false;
1791
1792 if (!ti->type->iterate_devices ||
1793 ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1794 return false;
1795 }
1796
1797 return true;
1798}
1799
1800static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1801 sector_t start, sector_t len, void *data)
1802{
1803 struct request_queue *q = bdev_get_queue(dev->bdev);
1804
1805 return q && !blk_queue_discard(q);
1806}
1807
1808static bool dm_table_supports_discards(struct dm_table *t)
1809{
1810 struct dm_target *ti;
1811 unsigned i;
1812
1813 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1814 ti = dm_table_get_target(t, i);
1815
1816 if (!ti->num_discard_bios)
1817 return false;
1818
1819 /*
1820 * Either the target provides discard support (as implied by setting
1821 * 'discards_supported') or it relies on _all_ data devices having
1822 * discard support.
1823 */
1824 if (!ti->discards_supported &&
1825 (!ti->type->iterate_devices ||
1826 ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1827 return false;
1828 }
1829
1830 return true;
1831}
1832
1833static int device_not_secure_erase_capable(struct dm_target *ti,
1834 struct dm_dev *dev, sector_t start,
1835 sector_t len, void *data)
1836{
1837 struct request_queue *q = bdev_get_queue(dev->bdev);
1838
1839 return q && !blk_queue_secure_erase(q);
1840}
1841
1842static bool dm_table_supports_secure_erase(struct dm_table *t)
1843{
1844 struct dm_target *ti;
1845 unsigned int i;
1846
1847 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1848 ti = dm_table_get_target(t, i);
1849
1850 if (!ti->num_secure_erase_bios)
1851 return false;
1852
1853 if (!ti->type->iterate_devices ||
1854 ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1855 return false;
1856 }
1857
1858 return true;
1859}
1860
1861static int device_requires_stable_pages(struct dm_target *ti,
1862 struct dm_dev *dev, sector_t start,
1863 sector_t len, void *data)
1864{
1865 struct request_queue *q = bdev_get_queue(dev->bdev);
1866
1867 return q && bdi_cap_stable_pages_required(q->backing_dev_info);
1868}
1869
1870/*
1871 * If any underlying device requires stable pages, a table must require
1872 * them as well. Only targets that support iterate_devices are considered:
1873 * don't want error, zero, etc to require stable pages.
1874 */
1875static bool dm_table_requires_stable_pages(struct dm_table *t)
1876{
1877 struct dm_target *ti;
1878 unsigned i;
1879
1880 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1881 ti = dm_table_get_target(t, i);
1882
1883 if (ti->type->iterate_devices &&
1884 ti->type->iterate_devices(ti, device_requires_stable_pages, NULL))
1885 return true;
1886 }
1887
1888 return false;
1889}
1890
1891void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1892 struct queue_limits *limits)
1893{
1894 bool wc = false, fua = false;
1895 int page_size = PAGE_SIZE;
1896
1897 /*
1898 * Copy table's limits to the DM device's request_queue
1899 */
1900 q->limits = *limits;
1901
1902 if (!dm_table_supports_discards(t)) {
1903 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
1904 /* Must also clear discard limits... */
1905 q->limits.max_discard_sectors = 0;
1906 q->limits.max_hw_discard_sectors = 0;
1907 q->limits.discard_granularity = 0;
1908 q->limits.discard_alignment = 0;
1909 q->limits.discard_misaligned = 0;
1910 } else
1911 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
1912
1913 if (dm_table_supports_secure_erase(t))
1914 blk_queue_flag_set(QUEUE_FLAG_SECERASE, q);
1915
1916 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1917 wc = true;
1918 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1919 fua = true;
1920 }
1921 blk_queue_write_cache(q, wc, fua);
1922
1923 if (dm_table_supports_dax(t, device_supports_dax, &page_size)) {
1924 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
1925 if (dm_table_supports_dax(t, device_dax_synchronous, NULL))
1926 set_dax_synchronous(t->md->dax_dev);
1927 }
1928 else
1929 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
1930
1931 if (dm_table_supports_dax_write_cache(t))
1932 dax_write_cache(t->md->dax_dev, true);
1933
1934 /* Ensure that all underlying devices are non-rotational. */
1935 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1936 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
1937 else
1938 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
1939
1940 if (!dm_table_supports_write_same(t))
1941 q->limits.max_write_same_sectors = 0;
1942 if (!dm_table_supports_write_zeroes(t))
1943 q->limits.max_write_zeroes_sectors = 0;
1944
1945 dm_table_verify_integrity(t);
1946
1947 /*
1948 * Some devices don't use blk_integrity but still want stable pages
1949 * because they do their own checksumming.
1950 */
1951 if (dm_table_requires_stable_pages(t))
1952 q->backing_dev_info->capabilities |= BDI_CAP_STABLE_WRITES;
1953 else
1954 q->backing_dev_info->capabilities &= ~BDI_CAP_STABLE_WRITES;
1955
1956 /*
1957 * Determine whether or not this queue's I/O timings contribute
1958 * to the entropy pool, Only request-based targets use this.
1959 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1960 * have it set.
1961 */
1962 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1963 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
1964
1965 /*
1966 * For a zoned target, the number of zones should be updated for the
1967 * correct value to be exposed in sysfs queue/nr_zones. For a BIO based
1968 * target, this is all that is needed. For a request based target, the
1969 * queue zone bitmaps must also be updated.
1970 * Use blk_revalidate_disk_zones() to handle this.
1971 */
1972 if (blk_queue_is_zoned(q))
1973 blk_revalidate_disk_zones(t->md->disk);
1974
1975 /* Allow reads to exceed readahead limits */
1976 q->backing_dev_info->io_pages = limits->max_sectors >> (PAGE_SHIFT - 9);
1977}
1978
1979unsigned int dm_table_get_num_targets(struct dm_table *t)
1980{
1981 return t->num_targets;
1982}
1983
1984struct list_head *dm_table_get_devices(struct dm_table *t)
1985{
1986 return &t->devices;
1987}
1988
1989fmode_t dm_table_get_mode(struct dm_table *t)
1990{
1991 return t->mode;
1992}
1993EXPORT_SYMBOL(dm_table_get_mode);
1994
1995enum suspend_mode {
1996 PRESUSPEND,
1997 PRESUSPEND_UNDO,
1998 POSTSUSPEND,
1999};
2000
2001static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
2002{
2003 int i = t->num_targets;
2004 struct dm_target *ti = t->targets;
2005
2006 lockdep_assert_held(&t->md->suspend_lock);
2007
2008 while (i--) {
2009 switch (mode) {
2010 case PRESUSPEND:
2011 if (ti->type->presuspend)
2012 ti->type->presuspend(ti);
2013 break;
2014 case PRESUSPEND_UNDO:
2015 if (ti->type->presuspend_undo)
2016 ti->type->presuspend_undo(ti);
2017 break;
2018 case POSTSUSPEND:
2019 if (ti->type->postsuspend)
2020 ti->type->postsuspend(ti);
2021 break;
2022 }
2023 ti++;
2024 }
2025}
2026
2027void dm_table_presuspend_targets(struct dm_table *t)
2028{
2029 if (!t)
2030 return;
2031
2032 suspend_targets(t, PRESUSPEND);
2033}
2034
2035void dm_table_presuspend_undo_targets(struct dm_table *t)
2036{
2037 if (!t)
2038 return;
2039
2040 suspend_targets(t, PRESUSPEND_UNDO);
2041}
2042
2043void dm_table_postsuspend_targets(struct dm_table *t)
2044{
2045 if (!t)
2046 return;
2047
2048 suspend_targets(t, POSTSUSPEND);
2049}
2050
2051int dm_table_resume_targets(struct dm_table *t)
2052{
2053 int i, r = 0;
2054
2055 lockdep_assert_held(&t->md->suspend_lock);
2056
2057 for (i = 0; i < t->num_targets; i++) {
2058 struct dm_target *ti = t->targets + i;
2059
2060 if (!ti->type->preresume)
2061 continue;
2062
2063 r = ti->type->preresume(ti);
2064 if (r) {
2065 DMERR("%s: %s: preresume failed, error = %d",
2066 dm_device_name(t->md), ti->type->name, r);
2067 return r;
2068 }
2069 }
2070
2071 for (i = 0; i < t->num_targets; i++) {
2072 struct dm_target *ti = t->targets + i;
2073
2074 if (ti->type->resume)
2075 ti->type->resume(ti);
2076 }
2077
2078 return 0;
2079}
2080
2081void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
2082{
2083 list_add(&cb->list, &t->target_callbacks);
2084}
2085EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
2086
2087int dm_table_any_congested(struct dm_table *t, int bdi_bits)
2088{
2089 struct dm_dev_internal *dd;
2090 struct list_head *devices = dm_table_get_devices(t);
2091 struct dm_target_callbacks *cb;
2092 int r = 0;
2093
2094 list_for_each_entry(dd, devices, list) {
2095 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
2096 char b[BDEVNAME_SIZE];
2097
2098 if (likely(q))
2099 r |= bdi_congested(q->backing_dev_info, bdi_bits);
2100 else
2101 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
2102 dm_device_name(t->md),
2103 bdevname(dd->dm_dev->bdev, b));
2104 }
2105
2106 list_for_each_entry(cb, &t->target_callbacks, list)
2107 if (cb->congested_fn)
2108 r |= cb->congested_fn(cb, bdi_bits);
2109
2110 return r;
2111}
2112
2113struct mapped_device *dm_table_get_md(struct dm_table *t)
2114{
2115 return t->md;
2116}
2117EXPORT_SYMBOL(dm_table_get_md);
2118
2119const char *dm_table_device_name(struct dm_table *t)
2120{
2121 return dm_device_name(t->md);
2122}
2123EXPORT_SYMBOL_GPL(dm_table_device_name);
2124
2125void dm_table_run_md_queue_async(struct dm_table *t)
2126{
2127 struct mapped_device *md;
2128 struct request_queue *queue;
2129
2130 if (!dm_table_request_based(t))
2131 return;
2132
2133 md = dm_table_get_md(t);
2134 queue = dm_get_md_queue(md);
2135 if (queue)
2136 blk_mq_run_hw_queues(queue, true);
2137}
2138EXPORT_SYMBOL(dm_table_run_md_queue_async);
2139