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/slab.h>
16 #include <linux/interrupt.h>
17 #include <linux/mutex.h>
18 #include <linux/delay.h>
19 #include <asm/atomic.h>
20
21 #define DM_MSG_PREFIX "table"
22
23 #define MAX_DEPTH 16
24 #define NODE_SIZE L1_CACHE_BYTES
25 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
26 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
27
28 /*
29 * The table has always exactly one reference from either mapped_device->map
30 * or hash_cell->new_map. This reference is not counted in table->holders.
31 * A pair of dm_create_table/dm_destroy_table functions is used for table
32 * creation/destruction.
33 *
34 * Temporary references from the other code increase table->holders. A pair
35 * of dm_table_get/dm_table_put functions is used to manipulate it.
36 *
37 * When the table is about to be destroyed, we wait for table->holders to
38 * drop to zero.
39 */
40
41 struct dm_table {
42 struct mapped_device *md;
43 atomic_t holders;
44 unsigned type;
45
46 /* btree table */
47 unsigned int depth;
48 unsigned int counts[MAX_DEPTH]; /* in nodes */
49 sector_t *index[MAX_DEPTH];
50
51 unsigned int num_targets;
52 unsigned int num_allocated;
53 sector_t *highs;
54 struct dm_target *targets;
55
56 /*
57 * Indicates the rw permissions for the new logical
58 * device. This should be a combination of FMODE_READ
59 * and FMODE_WRITE.
60 */
61 fmode_t mode;
62
63 /* a list of devices used by this table */
64 struct list_head devices;
65
66 /* events get handed up using this callback */
67 void (*event_fn)(void *);
68 void *event_context;
69
70 struct dm_md_mempools *mempools;
71 };
72
73 /*
74 * Similar to ceiling(log_size(n))
75 */
76 static unsigned int int_log(unsigned int n, unsigned int base)
77 {
78 int result = 0;
79
80 while (n > 1) {
81 n = dm_div_up(n, base);
82 result++;
83 }
84
85 return result;
86 }
87
88 /*
89 * Calculate the index of the child node of the n'th node k'th key.
90 */
91 static inline unsigned int get_child(unsigned int n, unsigned int k)
92 {
93 return (n * CHILDREN_PER_NODE) + k;
94 }
95
96 /*
97 * Return the n'th node of level l from table t.
98 */
99 static inline sector_t *get_node(struct dm_table *t,
100 unsigned int l, unsigned int n)
101 {
102 return t->index[l] + (n * KEYS_PER_NODE);
103 }
104
105 /*
106 * Return the highest key that you could lookup from the n'th
107 * node on level l of the btree.
108 */
109 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
110 {
111 for (; l < t->depth - 1; l++)
112 n = get_child(n, CHILDREN_PER_NODE - 1);
113
114 if (n >= t->counts[l])
115 return (sector_t) - 1;
116
117 return get_node(t, l, n)[KEYS_PER_NODE - 1];
118 }
119
120 /*
121 * Fills in a level of the btree based on the highs of the level
122 * below it.
123 */
124 static int setup_btree_index(unsigned int l, struct dm_table *t)
125 {
126 unsigned int n, k;
127 sector_t *node;
128
129 for (n = 0U; n < t->counts[l]; n++) {
130 node = get_node(t, l, n);
131
132 for (k = 0U; k < KEYS_PER_NODE; k++)
133 node[k] = high(t, l + 1, get_child(n, k));
134 }
135
136 return 0;
137 }
138
139 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
140 {
141 unsigned long size;
142 void *addr;
143
144 /*
145 * Check that we're not going to overflow.
146 */
147 if (nmemb > (ULONG_MAX / elem_size))
148 return NULL;
149
150 size = nmemb * elem_size;
151 addr = vmalloc(size);
152 if (addr)
153 memset(addr, 0, size);
154
155 return addr;
156 }
157
158 /*
159 * highs, and targets are managed as dynamic arrays during a
160 * table load.
161 */
162 static int alloc_targets(struct dm_table *t, unsigned int num)
163 {
164 sector_t *n_highs;
165 struct dm_target *n_targets;
166 int n = t->num_targets;
167
168 /*
169 * Allocate both the target array and offset array at once.
170 * Append an empty entry to catch sectors beyond the end of
171 * the device.
172 */
173 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
174 sizeof(sector_t));
175 if (!n_highs)
176 return -ENOMEM;
177
178 n_targets = (struct dm_target *) (n_highs + num);
179
180 if (n) {
181 memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
182 memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
183 }
184
185 memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
186 vfree(t->highs);
187
188 t->num_allocated = num;
189 t->highs = n_highs;
190 t->targets = n_targets;
191
192 return 0;
193 }
194
195 int dm_table_create(struct dm_table **result, fmode_t mode,
196 unsigned num_targets, struct mapped_device *md)
197 {
198 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
199
200 if (!t)
201 return -ENOMEM;
202
203 INIT_LIST_HEAD(&t->devices);
204 atomic_set(&t->holders, 0);
205
206 if (!num_targets)
207 num_targets = KEYS_PER_NODE;
208
209 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
210
211 if (alloc_targets(t, num_targets)) {
212 kfree(t);
213 t = NULL;
214 return -ENOMEM;
215 }
216
217 t->mode = mode;
218 t->md = md;
219 *result = t;
220 return 0;
221 }
222
223 static void free_devices(struct list_head *devices)
224 {
225 struct list_head *tmp, *next;
226
227 list_for_each_safe(tmp, next, devices) {
228 struct dm_dev_internal *dd =
229 list_entry(tmp, struct dm_dev_internal, list);
230 DMWARN("dm_table_destroy: dm_put_device call missing for %s",
231 dd->dm_dev.name);
232 kfree(dd);
233 }
234 }
235
236 void dm_table_destroy(struct dm_table *t)
237 {
238 unsigned int i;
239
240 while (atomic_read(&t->holders))
241 msleep(1);
242 smp_mb();
243
244 /* free the indexes (see dm_table_complete) */
245 if (t->depth >= 2)
246 vfree(t->index[t->depth - 2]);
247
248 /* free the targets */
249 for (i = 0; i < t->num_targets; i++) {
250 struct dm_target *tgt = t->targets + i;
251
252 if (tgt->type->dtr)
253 tgt->type->dtr(tgt);
254
255 dm_put_target_type(tgt->type);
256 }
257
258 vfree(t->highs);
259
260 /* free the device list */
261 if (t->devices.next != &t->devices)
262 free_devices(&t->devices);
263
264 dm_free_md_mempools(t->mempools);
265
266 kfree(t);
267 }
268
269 void dm_table_get(struct dm_table *t)
270 {
271 atomic_inc(&t->holders);
272 }
273
274 void dm_table_put(struct dm_table *t)
275 {
276 if (!t)
277 return;
278
279 smp_mb__before_atomic_dec();
280 atomic_dec(&t->holders);
281 }
282
283 /*
284 * Checks to see if we need to extend highs or targets.
285 */
286 static inline int check_space(struct dm_table *t)
287 {
288 if (t->num_targets >= t->num_allocated)
289 return alloc_targets(t, t->num_allocated * 2);
290
291 return 0;
292 }
293
294 /*
295 * See if we've already got a device in the list.
296 */
297 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
298 {
299 struct dm_dev_internal *dd;
300
301 list_for_each_entry (dd, l, list)
302 if (dd->dm_dev.bdev->bd_dev == dev)
303 return dd;
304
305 return NULL;
306 }
307
308 /*
309 * Open a device so we can use it as a map destination.
310 */
311 static int open_dev(struct dm_dev_internal *d, dev_t dev,
312 struct mapped_device *md)
313 {
314 static char *_claim_ptr = "I belong to device-mapper";
315 struct block_device *bdev;
316
317 int r;
318
319 BUG_ON(d->dm_dev.bdev);
320
321 bdev = open_by_devnum(dev, d->dm_dev.mode);
322 if (IS_ERR(bdev))
323 return PTR_ERR(bdev);
324 r = bd_claim_by_disk(bdev, _claim_ptr, dm_disk(md));
325 if (r)
326 blkdev_put(bdev, d->dm_dev.mode);
327 else
328 d->dm_dev.bdev = bdev;
329 return r;
330 }
331
332 /*
333 * Close a device that we've been using.
334 */
335 static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
336 {
337 if (!d->dm_dev.bdev)
338 return;
339
340 bd_release_from_disk(d->dm_dev.bdev, dm_disk(md));
341 blkdev_put(d->dm_dev.bdev, d->dm_dev.mode);
342 d->dm_dev.bdev = NULL;
343 }
344
345 /*
346 * If possible, this checks an area of a destination device is invalid.
347 */
348 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
349 sector_t start, sector_t len, void *data)
350 {
351 struct queue_limits *limits = data;
352 struct block_device *bdev = dev->bdev;
353 sector_t dev_size =
354 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
355 unsigned short logical_block_size_sectors =
356 limits->logical_block_size >> SECTOR_SHIFT;
357 char b[BDEVNAME_SIZE];
358
359 if (!dev_size)
360 return 0;
361
362 if ((start >= dev_size) || (start + len > dev_size)) {
363 DMWARN("%s: %s too small for target: "
364 "start=%llu, len=%llu, dev_size=%llu",
365 dm_device_name(ti->table->md), bdevname(bdev, b),
366 (unsigned long long)start,
367 (unsigned long long)len,
368 (unsigned long long)dev_size);
369 return 1;
370 }
371
372 if (logical_block_size_sectors <= 1)
373 return 0;
374
375 if (start & (logical_block_size_sectors - 1)) {
376 DMWARN("%s: start=%llu not aligned to h/w "
377 "logical block size %u of %s",
378 dm_device_name(ti->table->md),
379 (unsigned long long)start,
380 limits->logical_block_size, bdevname(bdev, b));
381 return 1;
382 }
383
384 if (len & (logical_block_size_sectors - 1)) {
385 DMWARN("%s: len=%llu not aligned to h/w "
386 "logical block size %u of %s",
387 dm_device_name(ti->table->md),
388 (unsigned long long)len,
389 limits->logical_block_size, bdevname(bdev, b));
390 return 1;
391 }
392
393 return 0;
394 }
395
396 /*
397 * This upgrades the mode on an already open dm_dev, being
398 * careful to leave things as they were if we fail to reopen the
399 * device and not to touch the existing bdev field in case
400 * it is accessed concurrently inside dm_table_any_congested().
401 */
402 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
403 struct mapped_device *md)
404 {
405 int r;
406 struct dm_dev_internal dd_new, dd_old;
407
408 dd_new = dd_old = *dd;
409
410 dd_new.dm_dev.mode |= new_mode;
411 dd_new.dm_dev.bdev = NULL;
412
413 r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md);
414 if (r)
415 return r;
416
417 dd->dm_dev.mode |= new_mode;
418 close_dev(&dd_old, md);
419
420 return 0;
421 }
422
423 /*
424 * Add a device to the list, or just increment the usage count if
425 * it's already present.
426 */
427 static int __table_get_device(struct dm_table *t, struct dm_target *ti,
428 const char *path, sector_t start, sector_t len,
429 fmode_t mode, struct dm_dev **result)
430 {
431 int r;
432 dev_t uninitialized_var(dev);
433 struct dm_dev_internal *dd;
434 unsigned int major, minor;
435
436 BUG_ON(!t);
437
438 if (sscanf(path, "%u:%u", &major, &minor) == 2) {
439 /* Extract the major/minor numbers */
440 dev = MKDEV(major, minor);
441 if (MAJOR(dev) != major || MINOR(dev) != minor)
442 return -EOVERFLOW;
443 } else {
444 /* convert the path to a device */
445 struct block_device *bdev = lookup_bdev(path);
446
447 if (IS_ERR(bdev))
448 return PTR_ERR(bdev);
449 dev = bdev->bd_dev;
450 bdput(bdev);
451 }
452
453 dd = find_device(&t->devices, dev);
454 if (!dd) {
455 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
456 if (!dd)
457 return -ENOMEM;
458
459 dd->dm_dev.mode = mode;
460 dd->dm_dev.bdev = NULL;
461
462 if ((r = open_dev(dd, dev, t->md))) {
463 kfree(dd);
464 return r;
465 }
466
467 format_dev_t(dd->dm_dev.name, dev);
468
469 atomic_set(&dd->count, 0);
470 list_add(&dd->list, &t->devices);
471
472 } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
473 r = upgrade_mode(dd, mode, t->md);
474 if (r)
475 return r;
476 }
477 atomic_inc(&dd->count);
478
479 *result = &dd->dm_dev;
480 return 0;
481 }
482
483 /*
484 * Returns the minimum that is _not_ zero, unless both are zero.
485 */
486 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
487
488 int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
489 sector_t start, sector_t len, void *data)
490 {
491 struct queue_limits *limits = data;
492 struct block_device *bdev = dev->bdev;
493 struct request_queue *q = bdev_get_queue(bdev);
494 char b[BDEVNAME_SIZE];
495
496 if (unlikely(!q)) {
497 DMWARN("%s: Cannot set limits for nonexistent device %s",
498 dm_device_name(ti->table->md), bdevname(bdev, b));
499 return 0;
500 }
501
502 if (blk_stack_limits(limits, &q->limits, start << 9) < 0)
503 DMWARN("%s: target device %s is misaligned: "
504 "physical_block_size=%u, logical_block_size=%u, "
505 "alignment_offset=%u, start=%llu",
506 dm_device_name(ti->table->md), bdevname(bdev, b),
507 q->limits.physical_block_size,
508 q->limits.logical_block_size,
509 q->limits.alignment_offset,
510 (unsigned long long) start << 9);
511
512
513 /*
514 * Check if merge fn is supported.
515 * If not we'll force DM to use PAGE_SIZE or
516 * smaller I/O, just to be safe.
517 */
518
519 if (q->merge_bvec_fn && !ti->type->merge)
520 limits->max_sectors =
521 min_not_zero(limits->max_sectors,
522 (unsigned int) (PAGE_SIZE >> 9));
523 return 0;
524 }
525 EXPORT_SYMBOL_GPL(dm_set_device_limits);
526
527 int dm_get_device(struct dm_target *ti, const char *path, sector_t start,
528 sector_t len, fmode_t mode, struct dm_dev **result)
529 {
530 return __table_get_device(ti->table, ti, path,
531 start, len, mode, result);
532 }
533
534
535 /*
536 * Decrement a devices use count and remove it if necessary.
537 */
538 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
539 {
540 struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
541 dm_dev);
542
543 if (atomic_dec_and_test(&dd->count)) {
544 close_dev(dd, ti->table->md);
545 list_del(&dd->list);
546 kfree(dd);
547 }
548 }
549
550 /*
551 * Checks to see if the target joins onto the end of the table.
552 */
553 static int adjoin(struct dm_table *table, struct dm_target *ti)
554 {
555 struct dm_target *prev;
556
557 if (!table->num_targets)
558 return !ti->begin;
559
560 prev = &table->targets[table->num_targets - 1];
561 return (ti->begin == (prev->begin + prev->len));
562 }
563
564 /*
565 * Used to dynamically allocate the arg array.
566 */
567 static char **realloc_argv(unsigned *array_size, char **old_argv)
568 {
569 char **argv;
570 unsigned new_size;
571
572 new_size = *array_size ? *array_size * 2 : 64;
573 argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
574 if (argv) {
575 memcpy(argv, old_argv, *array_size * sizeof(*argv));
576 *array_size = new_size;
577 }
578
579 kfree(old_argv);
580 return argv;
581 }
582
583 /*
584 * Destructively splits up the argument list to pass to ctr.
585 */
586 int dm_split_args(int *argc, char ***argvp, char *input)
587 {
588 char *start, *end = input, *out, **argv = NULL;
589 unsigned array_size = 0;
590
591 *argc = 0;
592
593 if (!input) {
594 *argvp = NULL;
595 return 0;
596 }
597
598 argv = realloc_argv(&array_size, argv);
599 if (!argv)
600 return -ENOMEM;
601
602 while (1) {
603 start = end;
604
605 /* Skip whitespace */
606 while (*start && isspace(*start))
607 start++;
608
609 if (!*start)
610 break; /* success, we hit the end */
611
612 /* 'out' is used to remove any back-quotes */
613 end = out = start;
614 while (*end) {
615 /* Everything apart from '\0' can be quoted */
616 if (*end == '\\' && *(end + 1)) {
617 *out++ = *(end + 1);
618 end += 2;
619 continue;
620 }
621
622 if (isspace(*end))
623 break; /* end of token */
624
625 *out++ = *end++;
626 }
627
628 /* have we already filled the array ? */
629 if ((*argc + 1) > array_size) {
630 argv = realloc_argv(&array_size, argv);
631 if (!argv)
632 return -ENOMEM;
633 }
634
635 /* we know this is whitespace */
636 if (*end)
637 end++;
638
639 /* terminate the string and put it in the array */
640 *out = '\0';
641 argv[*argc] = start;
642 (*argc)++;
643 }
644
645 *argvp = argv;
646 return 0;
647 }
648
649 /*
650 * Impose necessary and sufficient conditions on a devices's table such
651 * that any incoming bio which respects its logical_block_size can be
652 * processed successfully. If it falls across the boundary between
653 * two or more targets, the size of each piece it gets split into must
654 * be compatible with the logical_block_size of the target processing it.
655 */
656 static int validate_hardware_logical_block_alignment(struct dm_table *table,
657 struct queue_limits *limits)
658 {
659 /*
660 * This function uses arithmetic modulo the logical_block_size
661 * (in units of 512-byte sectors).
662 */
663 unsigned short device_logical_block_size_sects =
664 limits->logical_block_size >> SECTOR_SHIFT;
665
666 /*
667 * Offset of the start of the next table entry, mod logical_block_size.
668 */
669 unsigned short next_target_start = 0;
670
671 /*
672 * Given an aligned bio that extends beyond the end of a
673 * target, how many sectors must the next target handle?
674 */
675 unsigned short remaining = 0;
676
677 struct dm_target *uninitialized_var(ti);
678 struct queue_limits ti_limits;
679 unsigned i = 0;
680
681 /*
682 * Check each entry in the table in turn.
683 */
684 while (i < dm_table_get_num_targets(table)) {
685 ti = dm_table_get_target(table, i++);
686
687 blk_set_default_limits(&ti_limits);
688
689 /* combine all target devices' limits */
690 if (ti->type->iterate_devices)
691 ti->type->iterate_devices(ti, dm_set_device_limits,
692 &ti_limits);
693
694 /*
695 * If the remaining sectors fall entirely within this
696 * table entry are they compatible with its logical_block_size?
697 */
698 if (remaining < ti->len &&
699 remaining & ((ti_limits.logical_block_size >>
700 SECTOR_SHIFT) - 1))
701 break; /* Error */
702
703 next_target_start =
704 (unsigned short) ((next_target_start + ti->len) &
705 (device_logical_block_size_sects - 1));
706 remaining = next_target_start ?
707 device_logical_block_size_sects - next_target_start : 0;
708 }
709
710 if (remaining) {
711 DMWARN("%s: table line %u (start sect %llu len %llu) "
712 "not aligned to h/w logical block size %u",
713 dm_device_name(table->md), i,
714 (unsigned long long) ti->begin,
715 (unsigned long long) ti->len,
716 limits->logical_block_size);
717 return -EINVAL;
718 }
719
720 return 0;
721 }
722
723 int dm_table_add_target(struct dm_table *t, const char *type,
724 sector_t start, sector_t len, char *params)
725 {
726 int r = -EINVAL, argc;
727 char **argv;
728 struct dm_target *tgt;
729
730 if ((r = check_space(t)))
731 return r;
732
733 tgt = t->targets + t->num_targets;
734 memset(tgt, 0, sizeof(*tgt));
735
736 if (!len) {
737 DMERR("%s: zero-length target", dm_device_name(t->md));
738 return -EINVAL;
739 }
740
741 tgt->type = dm_get_target_type(type);
742 if (!tgt->type) {
743 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
744 type);
745 return -EINVAL;
746 }
747
748 tgt->table = t;
749 tgt->begin = start;
750 tgt->len = len;
751 tgt->error = "Unknown error";
752
753 /*
754 * Does this target adjoin the previous one ?
755 */
756 if (!adjoin(t, tgt)) {
757 tgt->error = "Gap in table";
758 r = -EINVAL;
759 goto bad;
760 }
761
762 r = dm_split_args(&argc, &argv, params);
763 if (r) {
764 tgt->error = "couldn't split parameters (insufficient memory)";
765 goto bad;
766 }
767
768 r = tgt->type->ctr(tgt, argc, argv);
769 kfree(argv);
770 if (r)
771 goto bad;
772
773 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
774
775 return 0;
776
777 bad:
778 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
779 dm_put_target_type(tgt->type);
780 return r;
781 }
782
783 int dm_table_set_type(struct dm_table *t)
784 {
785 unsigned i;
786 unsigned bio_based = 0, request_based = 0;
787 struct dm_target *tgt;
788 struct dm_dev_internal *dd;
789 struct list_head *devices;
790
791 for (i = 0; i < t->num_targets; i++) {
792 tgt = t->targets + i;
793 if (dm_target_request_based(tgt))
794 request_based = 1;
795 else
796 bio_based = 1;
797
798 if (bio_based && request_based) {
799 DMWARN("Inconsistent table: different target types"
800 " can't be mixed up");
801 return -EINVAL;
802 }
803 }
804
805 if (bio_based) {
806 /* We must use this table as bio-based */
807 t->type = DM_TYPE_BIO_BASED;
808 return 0;
809 }
810
811 BUG_ON(!request_based); /* No targets in this table */
812
813 /* Non-request-stackable devices can't be used for request-based dm */
814 devices = dm_table_get_devices(t);
815 list_for_each_entry(dd, devices, list) {
816 if (!blk_queue_stackable(bdev_get_queue(dd->dm_dev.bdev))) {
817 DMWARN("table load rejected: including"
818 " non-request-stackable devices");
819 return -EINVAL;
820 }
821 }
822
823 /*
824 * Request-based dm supports only tables that have a single target now.
825 * To support multiple targets, request splitting support is needed,
826 * and that needs lots of changes in the block-layer.
827 * (e.g. request completion process for partial completion.)
828 */
829 if (t->num_targets > 1) {
830 DMWARN("Request-based dm doesn't support multiple targets yet");
831 return -EINVAL;
832 }
833
834 t->type = DM_TYPE_REQUEST_BASED;
835
836 return 0;
837 }
838
839 unsigned dm_table_get_type(struct dm_table *t)
840 {
841 return t->type;
842 }
843
844 bool dm_table_request_based(struct dm_table *t)
845 {
846 return dm_table_get_type(t) == DM_TYPE_REQUEST_BASED;
847 }
848
849 int dm_table_alloc_md_mempools(struct dm_table *t)
850 {
851 unsigned type = dm_table_get_type(t);
852
853 if (unlikely(type == DM_TYPE_NONE)) {
854 DMWARN("no table type is set, can't allocate mempools");
855 return -EINVAL;
856 }
857
858 t->mempools = dm_alloc_md_mempools(type);
859 if (!t->mempools)
860 return -ENOMEM;
861
862 return 0;
863 }
864
865 void dm_table_free_md_mempools(struct dm_table *t)
866 {
867 dm_free_md_mempools(t->mempools);
868 t->mempools = NULL;
869 }
870
871 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
872 {
873 return t->mempools;
874 }
875
876 static int setup_indexes(struct dm_table *t)
877 {
878 int i;
879 unsigned int total = 0;
880 sector_t *indexes;
881
882 /* allocate the space for *all* the indexes */
883 for (i = t->depth - 2; i >= 0; i--) {
884 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
885 total += t->counts[i];
886 }
887
888 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
889 if (!indexes)
890 return -ENOMEM;
891
892 /* set up internal nodes, bottom-up */
893 for (i = t->depth - 2; i >= 0; i--) {
894 t->index[i] = indexes;
895 indexes += (KEYS_PER_NODE * t->counts[i]);
896 setup_btree_index(i, t);
897 }
898
899 return 0;
900 }
901
902 /*
903 * Builds the btree to index the map.
904 */
905 int dm_table_complete(struct dm_table *t)
906 {
907 int r = 0;
908 unsigned int leaf_nodes;
909
910 /* how many indexes will the btree have ? */
911 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
912 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
913
914 /* leaf layer has already been set up */
915 t->counts[t->depth - 1] = leaf_nodes;
916 t->index[t->depth - 1] = t->highs;
917
918 if (t->depth >= 2)
919 r = setup_indexes(t);
920
921 return r;
922 }
923
924 static DEFINE_MUTEX(_event_lock);
925 void dm_table_event_callback(struct dm_table *t,
926 void (*fn)(void *), void *context)
927 {
928 mutex_lock(&_event_lock);
929 t->event_fn = fn;
930 t->event_context = context;
931 mutex_unlock(&_event_lock);
932 }
933
934 void dm_table_event(struct dm_table *t)
935 {
936 /*
937 * You can no longer call dm_table_event() from interrupt
938 * context, use a bottom half instead.
939 */
940 BUG_ON(in_interrupt());
941
942 mutex_lock(&_event_lock);
943 if (t->event_fn)
944 t->event_fn(t->event_context);
945 mutex_unlock(&_event_lock);
946 }
947
948 sector_t dm_table_get_size(struct dm_table *t)
949 {
950 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
951 }
952
953 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
954 {
955 if (index >= t->num_targets)
956 return NULL;
957
958 return t->targets + index;
959 }
960
961 /*
962 * Search the btree for the correct target.
963 *
964 * Caller should check returned pointer with dm_target_is_valid()
965 * to trap I/O beyond end of device.
966 */
967 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
968 {
969 unsigned int l, n = 0, k = 0;
970 sector_t *node;
971
972 for (l = 0; l < t->depth; l++) {
973 n = get_child(n, k);
974 node = get_node(t, l, n);
975
976 for (k = 0; k < KEYS_PER_NODE; k++)
977 if (node[k] >= sector)
978 break;
979 }
980
981 return &t->targets[(KEYS_PER_NODE * n) + k];
982 }
983
984 /*
985 * Establish the new table's queue_limits and validate them.
986 */
987 int dm_calculate_queue_limits(struct dm_table *table,
988 struct queue_limits *limits)
989 {
990 struct dm_target *uninitialized_var(ti);
991 struct queue_limits ti_limits;
992 unsigned i = 0;
993
994 blk_set_default_limits(limits);
995
996 while (i < dm_table_get_num_targets(table)) {
997 blk_set_default_limits(&ti_limits);
998
999 ti = dm_table_get_target(table, i++);
1000
1001 if (!ti->type->iterate_devices)
1002 goto combine_limits;
1003
1004 /*
1005 * Combine queue limits of all the devices this target uses.
1006 */
1007 ti->type->iterate_devices(ti, dm_set_device_limits,
1008 &ti_limits);
1009
1010 /* Set I/O hints portion of queue limits */
1011 if (ti->type->io_hints)
1012 ti->type->io_hints(ti, &ti_limits);
1013
1014 /*
1015 * Check each device area is consistent with the target's
1016 * overall queue limits.
1017 */
1018 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1019 &ti_limits))
1020 return -EINVAL;
1021
1022 combine_limits:
1023 /*
1024 * Merge this target's queue limits into the overall limits
1025 * for the table.
1026 */
1027 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1028 DMWARN("%s: target device "
1029 "(start sect %llu len %llu) "
1030 "is misaligned",
1031 dm_device_name(table->md),
1032 (unsigned long long) ti->begin,
1033 (unsigned long long) ti->len);
1034 }
1035
1036 return validate_hardware_logical_block_alignment(table, limits);
1037 }
1038
1039 /*
1040 * Set the integrity profile for this device if all devices used have
1041 * matching profiles.
1042 */
1043 static void dm_table_set_integrity(struct dm_table *t)
1044 {
1045 struct list_head *devices = dm_table_get_devices(t);
1046 struct dm_dev_internal *prev = NULL, *dd = NULL;
1047
1048 if (!blk_get_integrity(dm_disk(t->md)))
1049 return;
1050
1051 list_for_each_entry(dd, devices, list) {
1052 if (prev &&
1053 blk_integrity_compare(prev->dm_dev.bdev->bd_disk,
1054 dd->dm_dev.bdev->bd_disk) < 0) {
1055 DMWARN("%s: integrity not set: %s and %s mismatch",
1056 dm_device_name(t->md),
1057 prev->dm_dev.bdev->bd_disk->disk_name,
1058 dd->dm_dev.bdev->bd_disk->disk_name);
1059 goto no_integrity;
1060 }
1061 prev = dd;
1062 }
1063
1064 if (!prev || !bdev_get_integrity(prev->dm_dev.bdev))
1065 goto no_integrity;
1066
1067 blk_integrity_register(dm_disk(t->md),
1068 bdev_get_integrity(prev->dm_dev.bdev));
1069
1070 return;
1071
1072 no_integrity:
1073 blk_integrity_register(dm_disk(t->md), NULL);
1074
1075 return;
1076 }
1077
1078 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1079 struct queue_limits *limits)
1080 {
1081 /*
1082 * Each target device in the table has a data area that should normally
1083 * be aligned such that the DM device's alignment_offset is 0.
1084 * FIXME: Propagate alignment_offsets up the stack and warn of
1085 * sub-optimal or inconsistent settings.
1086 */
1087 limits->alignment_offset = 0;
1088 limits->misaligned = 0;
1089
1090 /*
1091 * Copy table's limits to the DM device's request_queue
1092 */
1093 q->limits = *limits;
1094
1095 if (limits->no_cluster)
1096 queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q);
1097 else
1098 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, q);
1099
1100 dm_table_set_integrity(t);
1101
1102 /*
1103 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1104 * visible to other CPUs because, once the flag is set, incoming bios
1105 * are processed by request-based dm, which refers to the queue
1106 * settings.
1107 * Until the flag set, bios are passed to bio-based dm and queued to
1108 * md->deferred where queue settings are not needed yet.
1109 * Those bios are passed to request-based dm at the resume time.
1110 */
1111 smp_mb();
1112 if (dm_table_request_based(t))
1113 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1114 }
1115
1116 unsigned int dm_table_get_num_targets(struct dm_table *t)
1117 {
1118 return t->num_targets;
1119 }
1120
1121 struct list_head *dm_table_get_devices(struct dm_table *t)
1122 {
1123 return &t->devices;
1124 }
1125
1126 fmode_t dm_table_get_mode(struct dm_table *t)
1127 {
1128 return t->mode;
1129 }
1130
1131 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
1132 {
1133 int i = t->num_targets;
1134 struct dm_target *ti = t->targets;
1135
1136 while (i--) {
1137 if (postsuspend) {
1138 if (ti->type->postsuspend)
1139 ti->type->postsuspend(ti);
1140 } else if (ti->type->presuspend)
1141 ti->type->presuspend(ti);
1142
1143 ti++;
1144 }
1145 }
1146
1147 void dm_table_presuspend_targets(struct dm_table *t)
1148 {
1149 if (!t)
1150 return;
1151
1152 suspend_targets(t, 0);
1153 }
1154
1155 void dm_table_postsuspend_targets(struct dm_table *t)
1156 {
1157 if (!t)
1158 return;
1159
1160 suspend_targets(t, 1);
1161 }
1162
1163 int dm_table_resume_targets(struct dm_table *t)
1164 {
1165 int i, r = 0;
1166
1167 for (i = 0; i < t->num_targets; i++) {
1168 struct dm_target *ti = t->targets + i;
1169
1170 if (!ti->type->preresume)
1171 continue;
1172
1173 r = ti->type->preresume(ti);
1174 if (r)
1175 return r;
1176 }
1177
1178 for (i = 0; i < t->num_targets; i++) {
1179 struct dm_target *ti = t->targets + i;
1180
1181 if (ti->type->resume)
1182 ti->type->resume(ti);
1183 }
1184
1185 return 0;
1186 }
1187
1188 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1189 {
1190 struct dm_dev_internal *dd;
1191 struct list_head *devices = dm_table_get_devices(t);
1192 int r = 0;
1193
1194 list_for_each_entry(dd, devices, list) {
1195 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
1196 char b[BDEVNAME_SIZE];
1197
1198 if (likely(q))
1199 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1200 else
1201 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1202 dm_device_name(t->md),
1203 bdevname(dd->dm_dev.bdev, b));
1204 }
1205
1206 return r;
1207 }
1208
1209 int dm_table_any_busy_target(struct dm_table *t)
1210 {
1211 unsigned i;
1212 struct dm_target *ti;
1213
1214 for (i = 0; i < t->num_targets; i++) {
1215 ti = t->targets + i;
1216 if (ti->type->busy && ti->type->busy(ti))
1217 return 1;
1218 }
1219
1220 return 0;
1221 }
1222
1223 void dm_table_unplug_all(struct dm_table *t)
1224 {
1225 struct dm_dev_internal *dd;
1226 struct list_head *devices = dm_table_get_devices(t);
1227
1228 list_for_each_entry(dd, devices, list) {
1229 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
1230 char b[BDEVNAME_SIZE];
1231
1232 if (likely(q))
1233 blk_unplug(q);
1234 else
1235 DMWARN_LIMIT("%s: Cannot unplug nonexistent device %s",
1236 dm_device_name(t->md),
1237 bdevname(dd->dm_dev.bdev, b));
1238 }
1239 }
1240
1241 struct mapped_device *dm_table_get_md(struct dm_table *t)
1242 {
1243 dm_get(t->md);
1244
1245 return t->md;
1246 }
1247
1248 EXPORT_SYMBOL(dm_vcalloc);
1249 EXPORT_SYMBOL(dm_get_device);
1250 EXPORT_SYMBOL(dm_put_device);
1251 EXPORT_SYMBOL(dm_table_event);
1252 EXPORT_SYMBOL(dm_table_get_size);
1253 EXPORT_SYMBOL(dm_table_get_mode);
1254 EXPORT_SYMBOL(dm_table_get_md);
1255 EXPORT_SYMBOL(dm_table_put);
1256 EXPORT_SYMBOL(dm_table_get);
1257 EXPORT_SYMBOL(dm_table_unplug_all);
1258
|
This page was automatically generated by the
LXR engine.
|