1 /*
2  * Copyright (C) 2011 Red Hat, Inc.
3  *
4  * This file is released under the GPL.
5  */
6 
7 #include "dm-space-map.h"
8 #include "dm-space-map-common.h"
9 #include "dm-space-map-metadata.h"
10 
11 #include <linux/list.h>
12 #include <linux/slab.h>
13 #include <linux/device-mapper.h>
14 
15 #define DM_MSG_PREFIX "space map metadata"
16 
17 /*----------------------------------------------------------------*/
18 
19 /*
20  * Space map interface.
21  *
22  * The low level disk format is written using the standard btree and
23  * transaction manager.  This means that performing disk operations may
24  * cause us to recurse into the space map in order to allocate new blocks.
25  * For this reason we have a pool of pre-allocated blocks large enough to
26  * service any metadata_ll_disk operation.
27  */
28 
29 /*
30  * FIXME: we should calculate this based on the size of the device.
31  * Only the metadata space map needs this functionality.
32  */
33 #define MAX_RECURSIVE_ALLOCATIONS 1024
34 
35 enum block_op_type {
36 	BOP_INC,
37 	BOP_DEC
38 };
39 
40 struct block_op {
41 	enum block_op_type type;
42 	dm_block_t block;
43 };
44 
45 struct sm_metadata {
46 	struct dm_space_map sm;
47 
48 	struct ll_disk ll;
49 	struct ll_disk old_ll;
50 
51 	dm_block_t begin;
52 
53 	unsigned recursion_count;
54 	unsigned allocated_this_transaction;
55 	unsigned nr_uncommitted;
56 	struct block_op uncommitted[MAX_RECURSIVE_ALLOCATIONS];
57 };
58 
59 static int add_bop(struct sm_metadata *smm, enum block_op_type type, dm_block_t b)
60 {
61 	struct block_op *op;
62 
63 	if (smm->nr_uncommitted == MAX_RECURSIVE_ALLOCATIONS) {
64 		DMERR("too many recursive allocations");
65 		return -ENOMEM;
66 	}
67 
68 	op = smm->uncommitted + smm->nr_uncommitted++;
69 	op->type = type;
70 	op->block = b;
71 
72 	return 0;
73 }
74 
75 static int commit_bop(struct sm_metadata *smm, struct block_op *op)
76 {
77 	int r = 0;
78 	enum allocation_event ev;
79 
80 	switch (op->type) {
81 	case BOP_INC:
82 		r = sm_ll_inc(&smm->ll, op->block, &ev);
83 		break;
84 
85 	case BOP_DEC:
86 		r = sm_ll_dec(&smm->ll, op->block, &ev);
87 		break;
88 	}
89 
90 	return r;
91 }
92 
93 static void in(struct sm_metadata *smm)
94 {
95 	smm->recursion_count++;
96 }
97 
98 static int out(struct sm_metadata *smm)
99 {
100 	int r = 0;
101 
102 	/*
103 	 * If we're not recursing then very bad things are happening.
104 	 */
105 	if (!smm->recursion_count) {
106 		DMERR("lost track of recursion depth");
107 		return -ENOMEM;
108 	}
109 
110 	if (smm->recursion_count == 1 && smm->nr_uncommitted) {
111 		while (smm->nr_uncommitted && !r) {
112 			smm->nr_uncommitted--;
113 			r = commit_bop(smm, smm->uncommitted +
114 				       smm->nr_uncommitted);
115 			if (r)
116 				break;
117 		}
118 	}
119 
120 	smm->recursion_count--;
121 
122 	return r;
123 }
124 
125 /*
126  * When using the out() function above, we often want to combine an error
127  * code for the operation run in the recursive context with that from
128  * out().
129  */
130 static int combine_errors(int r1, int r2)
131 {
132 	return r1 ? r1 : r2;
133 }
134 
135 static int recursing(struct sm_metadata *smm)
136 {
137 	return smm->recursion_count;
138 }
139 
140 static void sm_metadata_destroy(struct dm_space_map *sm)
141 {
142 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
143 
144 	kfree(smm);
145 }
146 
147 static int sm_metadata_extend(struct dm_space_map *sm, dm_block_t extra_blocks)
148 {
149 	DMERR("doesn't support extend");
150 	return -EINVAL;
151 }
152 
153 static int sm_metadata_get_nr_blocks(struct dm_space_map *sm, dm_block_t *count)
154 {
155 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
156 
157 	*count = smm->ll.nr_blocks;
158 
159 	return 0;
160 }
161 
162 static int sm_metadata_get_nr_free(struct dm_space_map *sm, dm_block_t *count)
163 {
164 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
165 
166 	*count = smm->old_ll.nr_blocks - smm->old_ll.nr_allocated -
167 		 smm->allocated_this_transaction;
168 
169 	return 0;
170 }
171 
172 static int sm_metadata_get_count(struct dm_space_map *sm, dm_block_t b,
173 				 uint32_t *result)
174 {
175 	int r, i;
176 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
177 	unsigned adjustment = 0;
178 
179 	/*
180 	 * We may have some uncommitted adjustments to add.  This list
181 	 * should always be really short.
182 	 */
183 	for (i = 0; i < smm->nr_uncommitted; i++) {
184 		struct block_op *op = smm->uncommitted + i;
185 
186 		if (op->block != b)
187 			continue;
188 
189 		switch (op->type) {
190 		case BOP_INC:
191 			adjustment++;
192 			break;
193 
194 		case BOP_DEC:
195 			adjustment--;
196 			break;
197 		}
198 	}
199 
200 	r = sm_ll_lookup(&smm->ll, b, result);
201 	if (r)
202 		return r;
203 
204 	*result += adjustment;
205 
206 	return 0;
207 }
208 
209 static int sm_metadata_count_is_more_than_one(struct dm_space_map *sm,
210 					      dm_block_t b, int *result)
211 {
212 	int r, i, adjustment = 0;
213 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
214 	uint32_t rc;
215 
216 	/*
217 	 * We may have some uncommitted adjustments to add.  This list
218 	 * should always be really short.
219 	 */
220 	for (i = 0; i < smm->nr_uncommitted; i++) {
221 		struct block_op *op = smm->uncommitted + i;
222 
223 		if (op->block != b)
224 			continue;
225 
226 		switch (op->type) {
227 		case BOP_INC:
228 			adjustment++;
229 			break;
230 
231 		case BOP_DEC:
232 			adjustment--;
233 			break;
234 		}
235 	}
236 
237 	if (adjustment > 1) {
238 		*result = 1;
239 		return 0;
240 	}
241 
242 	r = sm_ll_lookup_bitmap(&smm->ll, b, &rc);
243 	if (r)
244 		return r;
245 
246 	if (rc == 3)
247 		/*
248 		 * We err on the side of caution, and always return true.
249 		 */
250 		*result = 1;
251 	else
252 		*result = rc + adjustment > 1;
253 
254 	return 0;
255 }
256 
257 static int sm_metadata_set_count(struct dm_space_map *sm, dm_block_t b,
258 				 uint32_t count)
259 {
260 	int r, r2;
261 	enum allocation_event ev;
262 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
263 
264 	if (smm->recursion_count) {
265 		DMERR("cannot recurse set_count()");
266 		return -EINVAL;
267 	}
268 
269 	in(smm);
270 	r = sm_ll_insert(&smm->ll, b, count, &ev);
271 	r2 = out(smm);
272 
273 	return combine_errors(r, r2);
274 }
275 
276 static int sm_metadata_inc_block(struct dm_space_map *sm, dm_block_t b)
277 {
278 	int r, r2 = 0;
279 	enum allocation_event ev;
280 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
281 
282 	if (recursing(smm))
283 		r = add_bop(smm, BOP_INC, b);
284 	else {
285 		in(smm);
286 		r = sm_ll_inc(&smm->ll, b, &ev);
287 		r2 = out(smm);
288 	}
289 
290 	return combine_errors(r, r2);
291 }
292 
293 static int sm_metadata_dec_block(struct dm_space_map *sm, dm_block_t b)
294 {
295 	int r, r2 = 0;
296 	enum allocation_event ev;
297 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
298 
299 	if (recursing(smm))
300 		r = add_bop(smm, BOP_DEC, b);
301 	else {
302 		in(smm);
303 		r = sm_ll_dec(&smm->ll, b, &ev);
304 		r2 = out(smm);
305 	}
306 
307 	return combine_errors(r, r2);
308 }
309 
310 static int sm_metadata_new_block_(struct dm_space_map *sm, dm_block_t *b)
311 {
312 	int r, r2 = 0;
313 	enum allocation_event ev;
314 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
315 
316 	r = sm_ll_find_free_block(&smm->old_ll, smm->begin, smm->old_ll.nr_blocks, b);
317 	if (r)
318 		return r;
319 
320 	smm->begin = *b + 1;
321 
322 	if (recursing(smm))
323 		r = add_bop(smm, BOP_INC, *b);
324 	else {
325 		in(smm);
326 		r = sm_ll_inc(&smm->ll, *b, &ev);
327 		r2 = out(smm);
328 	}
329 
330 	if (!r)
331 		smm->allocated_this_transaction++;
332 
333 	return combine_errors(r, r2);
334 }
335 
336 static int sm_metadata_new_block(struct dm_space_map *sm, dm_block_t *b)
337 {
338 	int r = sm_metadata_new_block_(sm, b);
339 	if (r)
340 		DMERR("unable to allocate new metadata block");
341 	return r;
342 }
343 
344 static int sm_metadata_commit(struct dm_space_map *sm)
345 {
346 	int r;
347 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
348 
349 	r = sm_ll_commit(&smm->ll);
350 	if (r)
351 		return r;
352 
353 	memcpy(&smm->old_ll, &smm->ll, sizeof(smm->old_ll));
354 	smm->begin = 0;
355 	smm->allocated_this_transaction = 0;
356 
357 	return 0;
358 }
359 
360 static int sm_metadata_root_size(struct dm_space_map *sm, size_t *result)
361 {
362 	*result = sizeof(struct disk_sm_root);
363 
364 	return 0;
365 }
366 
367 static int sm_metadata_copy_root(struct dm_space_map *sm, void *where_le, size_t max)
368 {
369 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
370 	struct disk_sm_root root_le;
371 
372 	root_le.nr_blocks = cpu_to_le64(smm->ll.nr_blocks);
373 	root_le.nr_allocated = cpu_to_le64(smm->ll.nr_allocated);
374 	root_le.bitmap_root = cpu_to_le64(smm->ll.bitmap_root);
375 	root_le.ref_count_root = cpu_to_le64(smm->ll.ref_count_root);
376 
377 	if (max < sizeof(root_le))
378 		return -ENOSPC;
379 
380 	memcpy(where_le, &root_le, sizeof(root_le));
381 
382 	return 0;
383 }
384 
385 static struct dm_space_map ops = {
386 	.destroy = sm_metadata_destroy,
387 	.extend = sm_metadata_extend,
388 	.get_nr_blocks = sm_metadata_get_nr_blocks,
389 	.get_nr_free = sm_metadata_get_nr_free,
390 	.get_count = sm_metadata_get_count,
391 	.count_is_more_than_one = sm_metadata_count_is_more_than_one,
392 	.set_count = sm_metadata_set_count,
393 	.inc_block = sm_metadata_inc_block,
394 	.dec_block = sm_metadata_dec_block,
395 	.new_block = sm_metadata_new_block,
396 	.commit = sm_metadata_commit,
397 	.root_size = sm_metadata_root_size,
398 	.copy_root = sm_metadata_copy_root
399 };
400 
401 /*----------------------------------------------------------------*/
402 
403 /*
404  * When a new space map is created that manages its own space.  We use
405  * this tiny bootstrap allocator.
406  */
407 static void sm_bootstrap_destroy(struct dm_space_map *sm)
408 {
409 }
410 
411 static int sm_bootstrap_extend(struct dm_space_map *sm, dm_block_t extra_blocks)
412 {
413 	DMERR("boostrap doesn't support extend");
414 
415 	return -EINVAL;
416 }
417 
418 static int sm_bootstrap_get_nr_blocks(struct dm_space_map *sm, dm_block_t *count)
419 {
420 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
421 
422 	return smm->ll.nr_blocks;
423 }
424 
425 static int sm_bootstrap_get_nr_free(struct dm_space_map *sm, dm_block_t *count)
426 {
427 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
428 
429 	*count = smm->ll.nr_blocks - smm->begin;
430 
431 	return 0;
432 }
433 
434 static int sm_bootstrap_get_count(struct dm_space_map *sm, dm_block_t b,
435 				  uint32_t *result)
436 {
437 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
438 
439 	return b < smm->begin ? 1 : 0;
440 }
441 
442 static int sm_bootstrap_count_is_more_than_one(struct dm_space_map *sm,
443 					       dm_block_t b, int *result)
444 {
445 	*result = 0;
446 
447 	return 0;
448 }
449 
450 static int sm_bootstrap_set_count(struct dm_space_map *sm, dm_block_t b,
451 				  uint32_t count)
452 {
453 	DMERR("boostrap doesn't support set_count");
454 
455 	return -EINVAL;
456 }
457 
458 static int sm_bootstrap_new_block(struct dm_space_map *sm, dm_block_t *b)
459 {
460 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
461 
462 	/*
463 	 * We know the entire device is unused.
464 	 */
465 	if (smm->begin == smm->ll.nr_blocks)
466 		return -ENOSPC;
467 
468 	*b = smm->begin++;
469 
470 	return 0;
471 }
472 
473 static int sm_bootstrap_inc_block(struct dm_space_map *sm, dm_block_t b)
474 {
475 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
476 
477 	return add_bop(smm, BOP_INC, b);
478 }
479 
480 static int sm_bootstrap_dec_block(struct dm_space_map *sm, dm_block_t b)
481 {
482 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
483 
484 	return add_bop(smm, BOP_DEC, b);
485 }
486 
487 static int sm_bootstrap_commit(struct dm_space_map *sm)
488 {
489 	return 0;
490 }
491 
492 static int sm_bootstrap_root_size(struct dm_space_map *sm, size_t *result)
493 {
494 	DMERR("boostrap doesn't support root_size");
495 
496 	return -EINVAL;
497 }
498 
499 static int sm_bootstrap_copy_root(struct dm_space_map *sm, void *where,
500 				  size_t max)
501 {
502 	DMERR("boostrap doesn't support copy_root");
503 
504 	return -EINVAL;
505 }
506 
507 static struct dm_space_map bootstrap_ops = {
508 	.destroy = sm_bootstrap_destroy,
509 	.extend = sm_bootstrap_extend,
510 	.get_nr_blocks = sm_bootstrap_get_nr_blocks,
511 	.get_nr_free = sm_bootstrap_get_nr_free,
512 	.get_count = sm_bootstrap_get_count,
513 	.count_is_more_than_one = sm_bootstrap_count_is_more_than_one,
514 	.set_count = sm_bootstrap_set_count,
515 	.inc_block = sm_bootstrap_inc_block,
516 	.dec_block = sm_bootstrap_dec_block,
517 	.new_block = sm_bootstrap_new_block,
518 	.commit = sm_bootstrap_commit,
519 	.root_size = sm_bootstrap_root_size,
520 	.copy_root = sm_bootstrap_copy_root
521 };
522 
523 /*----------------------------------------------------------------*/
524 
525 struct dm_space_map *dm_sm_metadata_init(void)
526 {
527 	struct sm_metadata *smm;
528 
529 	smm = kmalloc(sizeof(*smm), GFP_KERNEL);
530 	if (!smm)
531 		return ERR_PTR(-ENOMEM);
532 
533 	memcpy(&smm->sm, &ops, sizeof(smm->sm));
534 
535 	return &smm->sm;
536 }
537 
538 int dm_sm_metadata_create(struct dm_space_map *sm,
539 			  struct dm_transaction_manager *tm,
540 			  dm_block_t nr_blocks,
541 			  dm_block_t superblock)
542 {
543 	int r;
544 	dm_block_t i;
545 	enum allocation_event ev;
546 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
547 
548 	smm->begin = superblock + 1;
549 	smm->recursion_count = 0;
550 	smm->allocated_this_transaction = 0;
551 	smm->nr_uncommitted = 0;
552 
553 	memcpy(&smm->sm, &bootstrap_ops, sizeof(smm->sm));
554 
555 	r = sm_ll_new_metadata(&smm->ll, tm);
556 	if (r)
557 		return r;
558 
559 	r = sm_ll_extend(&smm->ll, nr_blocks);
560 	if (r)
561 		return r;
562 
563 	memcpy(&smm->sm, &ops, sizeof(smm->sm));
564 
565 	/*
566 	 * Now we need to update the newly created data structures with the
567 	 * allocated blocks that they were built from.
568 	 */
569 	for (i = superblock; !r && i < smm->begin; i++)
570 		r = sm_ll_inc(&smm->ll, i, &ev);
571 
572 	if (r)
573 		return r;
574 
575 	return sm_metadata_commit(sm);
576 }
577 
578 int dm_sm_metadata_open(struct dm_space_map *sm,
579 			struct dm_transaction_manager *tm,
580 			void *root_le, size_t len)
581 {
582 	int r;
583 	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
584 
585 	r = sm_ll_open_metadata(&smm->ll, tm, root_le, len);
586 	if (r)
587 		return r;
588 
589 	smm->begin = 0;
590 	smm->recursion_count = 0;
591 	smm->allocated_this_transaction = 0;
592 	smm->nr_uncommitted = 0;
593 
594 	memcpy(&smm->old_ll, &smm->ll, sizeof(smm->old_ll));
595 	return 0;
596 }
597