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