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