drivers/md/persistent-data/dm-space-map-metadata.c

   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  * An edge triggered threshold.
  21  */
  22 struct threshold {
  23         bool threshold_set;
  24         bool value_set;
  25         dm_block_t threshold;
  26         dm_block_t current_value;
  27         dm_sm_threshold_fn fn;
  28         void *context;
  29 };
  30
  31 static void threshold_init(struct threshold *t)
  32 {
  33         t->threshold_set = false;
  34         t->value_set = false;
  35 }
  36
  37 static void set_threshold(struct threshold *t, dm_block_t value,
  38                           dm_sm_threshold_fn fn, void *context)
  39 {
  40         t->threshold_set = true;
  41         t->threshold = value;
  42         t->fn = fn;
  43         t->context = context;
  44 }
  45
  46 static bool below_threshold(struct threshold *t, dm_block_t value)
  47 {
  48         return t->threshold_set && value <= t->threshold;
  49 }
  50
  51 static bool threshold_already_triggered(struct threshold *t)
  52 {
  53         return t->value_set && below_threshold(t, t->current_value);
  54 }
  55
  56 static void check_threshold(struct threshold *t, dm_block_t value)
  57 {
  58         if (below_threshold(t, value) &&
  59             !threshold_already_triggered(t))
  60                 t->fn(t->context);
  61
  62         t->value_set = true;
  63         t->current_value = value;
  64 }
  65
  66 /*----------------------------------------------------------------*/
  67
  68 /*
  69  * Space map interface.
  70  *
  71  * The low level disk format is written using the standard btree and
  72  * transaction manager.  This means that performing disk operations may
  73  * cause us to recurse into the space map in order to allocate new blocks.
  74  * For this reason we have a pool of pre-allocated blocks large enough to
  75  * service any metadata_ll_disk operation.
  76  */
  77
  78 /*
  79  * FIXME: we should calculate this based on the size of the device.
  80  * Only the metadata space map needs this functionality.
  81  */
  82 #define MAX_RECURSIVE_ALLOCATIONS 1024
  83
  84 enum block_op_type {
  85         BOP_INC,
  86         BOP_DEC
  87 };
  88
  89 struct block_op {
  90         enum block_op_type type;
  91         dm_block_t block;
  92 };
  93
  94 struct sm_metadata {
  95         struct dm_space_map sm;
  96
  97         struct ll_disk ll;
  98         struct ll_disk old_ll;
  99
 100         dm_block_t begin;
 101
 102         unsigned recursion_count;
 103         unsigned allocated_this_transaction;
 104         unsigned nr_uncommitted;
 105         struct block_op uncommitted[MAX_RECURSIVE_ALLOCATIONS];
 106
 107         struct threshold threshold;
 108 };
 109
 110 static int add_bop(struct sm_metadata *smm, enum block_op_type type, dm_block_t b)
 111 {
 112         struct block_op *op;
 113
 114         if (smm->nr_uncommitted == MAX_RECURSIVE_ALLOCATIONS) {
 115                 DMERR("too many recursive allocations");
 116                 return -ENOMEM;
 117         }
 118
 119         op = smm->uncommitted + smm->nr_uncommitted++;
 120         op->type = type;
 121         op->block = b;
 122
 123         return 0;
 124 }
 125
 126 static int commit_bop(struct sm_metadata *smm, struct block_op *op)
 127 {
 128         int r = 0;
 129         enum allocation_event ev;
 130
 131         switch (op->type) {
 132         case BOP_INC:
 133                 r = sm_ll_inc(&smm->ll, op->block, &ev);
 134                 break;
 135
 136         case BOP_DEC:
 137                 r = sm_ll_dec(&smm->ll, op->block, &ev);
 138                 break;
 139         }
 140
 141         return r;
 142 }
 143
 144 static void in(struct sm_metadata *smm)
 145 {
 146         smm->recursion_count++;
 147 }
 148
 149 static int out(struct sm_metadata *smm)
 150 {
 151         int r = 0;
 152
 153         /*
 154          * If we're not recursing then very bad things are happening.
 155          */
 156         if (!smm->recursion_count) {
 157                 DMERR("lost track of recursion depth");
 158                 return -ENOMEM;
 159         }
 160
 161         if (smm->recursion_count == 1 && smm->nr_uncommitted) {
 162                 while (smm->nr_uncommitted && !r) {
 163                         smm->nr_uncommitted--;
 164                         r = commit_bop(smm, smm->uncommitted +
 165                                        smm->nr_uncommitted);
 166                         if (r)
 167                                 break;
 168                 }
 169         }
 170
 171         smm->recursion_count--;
 172
 173         return r;
 174 }
 175
 176 /*
 177  * When using the out() function above, we often want to combine an error
 178  * code for the operation run in the recursive context with that from
 179  * out().
 180  */
 181 static int combine_errors(int r1, int r2)
 182 {
 183         return r1 ? r1 : r2;
 184 }
 185
 186 static int recursing(struct sm_metadata *smm)
 187 {
 188         return smm->recursion_count;
 189 }
 190
 191 static void sm_metadata_destroy(struct dm_space_map *sm)
 192 {
 193         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 194
 195         kfree(smm);
 196 }
 197
 198 static int sm_metadata_get_nr_blocks(struct dm_space_map *sm, dm_block_t *count)
 199 {
 200         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 201
 202         *count = smm->ll.nr_blocks;
 203
 204         return 0;
 205 }
 206
 207 static int sm_metadata_get_nr_free(struct dm_space_map *sm, dm_block_t *count)
 208 {
 209         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 210
 211         *count = smm->old_ll.nr_blocks - smm->old_ll.nr_allocated -
 212                  smm->allocated_this_transaction;
 213
 214         return 0;
 215 }
 216
 217 static int sm_metadata_get_count(struct dm_space_map *sm, dm_block_t b,
 218                                  uint32_t *result)
 219 {
 220         int r, i;
 221         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 222         unsigned adjustment = 0;
 223
 224         /*
 225          * We may have some uncommitted adjustments to add.  This list
 226          * should always be really short.
 227          */
 228         for (i = 0; i < smm->nr_uncommitted; i++) {
 229                 struct block_op *op = smm->uncommitted + i;
 230
 231                 if (op->block != b)
 232                         continue;
 233
 234                 switch (op->type) {
 235                 case BOP_INC:
 236                         adjustment++;
 237                         break;
 238
 239                 case BOP_DEC:
 240                         adjustment--;
 241                         break;
 242                 }
 243         }
 244
 245         r = sm_ll_lookup(&smm->ll, b, result);
 246         if (r)
 247                 return r;
 248
 249         *result += adjustment;
 250
 251         return 0;
 252 }
 253
 254 static int sm_metadata_count_is_more_than_one(struct dm_space_map *sm,
 255                                               dm_block_t b, int *result)
 256 {
 257         int r, i, adjustment = 0;
 258         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 259         uint32_t rc;
 260
 261         /*
 262          * We may have some uncommitted adjustments to add.  This list
 263          * should always be really short.
 264          */
 265         for (i = 0; i < smm->nr_uncommitted; i++) {
 266                 struct block_op *op = smm->uncommitted + i;
 267
 268                 if (op->block != b)
 269                         continue;
 270
 271                 switch (op->type) {
 272                 case BOP_INC:
 273                         adjustment++;
 274                         break;
 275
 276                 case BOP_DEC:
 277                         adjustment--;
 278                         break;
 279                 }
 280         }
 281
 282         if (adjustment > 1) {
 283                 *result = 1;
 284                 return 0;
 285         }
 286
 287         r = sm_ll_lookup_bitmap(&smm->ll, b, &rc);
 288         if (r)
 289                 return r;
 290
 291         if (rc == 3)
 292                 /*
 293                  * We err on the side of caution, and always return true.
 294                  */
 295                 *result = 1;
 296         else
 297                 *result = rc + adjustment > 1;
 298
 299         return 0;
 300 }
 301
 302 static int sm_metadata_set_count(struct dm_space_map *sm, dm_block_t b,
 303                                  uint32_t count)
 304 {
 305         int r, r2;
 306         enum allocation_event ev;
 307         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 308
 309         if (smm->recursion_count) {
 310                 DMERR("cannot recurse set_count()");
 311                 return -EINVAL;
 312         }
 313
 314         in(smm);
 315         r = sm_ll_insert(&smm->ll, b, count, &ev);
 316         r2 = out(smm);
 317
 318         return combine_errors(r, r2);
 319 }
 320
 321 static int sm_metadata_inc_block(struct dm_space_map *sm, dm_block_t b)
 322 {
 323         int r, r2 = 0;
 324         enum allocation_event ev;
 325         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 326
 327         if (recursing(smm))
 328                 r = add_bop(smm, BOP_INC, b);
 329         else {
 330                 in(smm);
 331                 r = sm_ll_inc(&smm->ll, b, &ev);
 332                 r2 = out(smm);
 333         }
 334
 335         return combine_errors(r, r2);
 336 }
 337
 338 static int sm_metadata_dec_block(struct dm_space_map *sm, dm_block_t b)
 339 {
 340         int r, r2 = 0;
 341         enum allocation_event ev;
 342         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 343
 344         if (recursing(smm))
 345                 r = add_bop(smm, BOP_DEC, b);
 346         else {
 347                 in(smm);
 348                 r = sm_ll_dec(&smm->ll, b, &ev);
 349                 r2 = out(smm);
 350         }
 351
 352         return combine_errors(r, r2);
 353 }
 354
 355 static int sm_metadata_new_block_(struct dm_space_map *sm, dm_block_t *b)
 356 {
 357         int r, r2 = 0;
 358         enum allocation_event ev;
 359         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 360
 361         r = sm_ll_find_free_block(&smm->old_ll, smm->begin, smm->old_ll.nr_blocks, b);
 362         if (r)
 363                 return r;
 364
 365         smm->begin = *b + 1;
 366
 367         if (recursing(smm))
 368                 r = add_bop(smm, BOP_INC, *b);
 369         else {
 370                 in(smm);
 371                 r = sm_ll_inc(&smm->ll, *b, &ev);
 372                 r2 = out(smm);
 373         }
 374
 375         if (!r)
 376                 smm->allocated_this_transaction++;
 377
 378         return combine_errors(r, r2);
 379 }
 380
 381 static int sm_metadata_new_block(struct dm_space_map *sm, dm_block_t *b)
 382 {
 383         dm_block_t count;
 384         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 385
 386         int r = sm_metadata_new_block_(sm, b);
 387         if (r)
 388                 DMERR("unable to allocate new metadata block");
 389
 390         r = sm_metadata_get_nr_free(sm, &count);
 391         if (r)
 392                 DMERR("couldn't get free block count");
 393
 394         check_threshold(&smm->threshold, count);
 395
 396         return r;
 397 }
 398
 399 static int sm_metadata_commit(struct dm_space_map *sm)
 400 {
 401         int r;
 402         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 403
 404         r = sm_ll_commit(&smm->ll);
 405         if (r)
 406                 return r;
 407
 408         memcpy(&smm->old_ll, &smm->ll, sizeof(smm->old_ll));
 409         smm->begin = 0;
 410         smm->allocated_this_transaction = 0;
 411
 412         return 0;
 413 }
 414
 415 static int sm_metadata_register_threshold_callback(struct dm_space_map *sm,
 416                                                    dm_block_t threshold,
 417                                                    dm_sm_threshold_fn fn,
 418                                                    void *context)
 419 {
 420         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 421
 422         set_threshold(&smm->threshold, threshold, fn, context);
 423
 424         return 0;
 425 }
 426
 427 static int sm_metadata_root_size(struct dm_space_map *sm, size_t *result)
 428 {
 429         *result = sizeof(struct disk_sm_root);
 430
 431         return 0;
 432 }
 433
 434 static int sm_metadata_copy_root(struct dm_space_map *sm, void *where_le, size_t max)
 435 {
 436         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 437         struct disk_sm_root root_le;
 438
 439         root_le.nr_blocks = cpu_to_le64(smm->ll.nr_blocks);
 440         root_le.nr_allocated = cpu_to_le64(smm->ll.nr_allocated);
 441         root_le.bitmap_root = cpu_to_le64(smm->ll.bitmap_root);
 442         root_le.ref_count_root = cpu_to_le64(smm->ll.ref_count_root);
 443
 444         if (max < sizeof(root_le))
 445                 return -ENOSPC;
 446
 447         memcpy(where_le, &root_le, sizeof(root_le));
 448
 449         return 0;
 450 }
 451
 452 static int sm_metadata_extend(struct dm_space_map *sm, dm_block_t extra_blocks);
 453
 454 static struct dm_space_map ops = {
 455         .destroy = sm_metadata_destroy,
 456         .extend = sm_metadata_extend,
 457         .get_nr_blocks = sm_metadata_get_nr_blocks,
 458         .get_nr_free = sm_metadata_get_nr_free,
 459         .get_count = sm_metadata_get_count,
 460         .count_is_more_than_one = sm_metadata_count_is_more_than_one,
 461         .set_count = sm_metadata_set_count,
 462         .inc_block = sm_metadata_inc_block,
 463         .dec_block = sm_metadata_dec_block,
 464         .new_block = sm_metadata_new_block,
 465         .commit = sm_metadata_commit,
 466         .root_size = sm_metadata_root_size,
 467         .copy_root = sm_metadata_copy_root,
 468         .register_threshold_callback = sm_metadata_register_threshold_callback
 469 };
 470
 471 /*----------------------------------------------------------------*/
 472
 473 /*
 474  * When a new space map is created that manages its own space.  We use
 475  * this tiny bootstrap allocator.
 476  */
 477 static void sm_bootstrap_destroy(struct dm_space_map *sm)
 478 {
 479 }
 480
 481 static int sm_bootstrap_extend(struct dm_space_map *sm, dm_block_t extra_blocks)
 482 {
 483         DMERR("bootstrap doesn't support extend");
 484
 485         return -EINVAL;
 486 }
 487
 488 static int sm_bootstrap_get_nr_blocks(struct dm_space_map *sm, dm_block_t *count)
 489 {
 490         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 491
 492         return smm->ll.nr_blocks;
 493 }
 494
 495 static int sm_bootstrap_get_nr_free(struct dm_space_map *sm, dm_block_t *count)
 496 {
 497         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 498
 499         *count = smm->ll.nr_blocks - smm->begin;
 500
 501         return 0;
 502 }
 503
 504 static int sm_bootstrap_get_count(struct dm_space_map *sm, dm_block_t b,
 505                                   uint32_t *result)
 506 {
 507         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 508
 509         return b < smm->begin ? 1 : 0;
 510 }
 511
 512 static int sm_bootstrap_count_is_more_than_one(struct dm_space_map *sm,
 513                                                dm_block_t b, int *result)
 514 {
 515         *result = 0;
 516
 517         return 0;
 518 }
 519
 520 static int sm_bootstrap_set_count(struct dm_space_map *sm, dm_block_t b,
 521                                   uint32_t count)
 522 {
 523         DMERR("bootstrap doesn't support set_count");
 524
 525         return -EINVAL;
 526 }
 527
 528 static int sm_bootstrap_new_block(struct dm_space_map *sm, dm_block_t *b)
 529 {
 530         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 531
 532         /*
 533          * We know the entire device is unused.
 534          */
 535         if (smm->begin == smm->ll.nr_blocks)
 536                 return -ENOSPC;
 537
 538         *b = smm->begin++;
 539
 540         return 0;
 541 }
 542
 543 static int sm_bootstrap_inc_block(struct dm_space_map *sm, dm_block_t b)
 544 {
 545         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 546
 547         return add_bop(smm, BOP_INC, b);
 548 }
 549
 550 static int sm_bootstrap_dec_block(struct dm_space_map *sm, dm_block_t b)
 551 {
 552         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 553
 554         return add_bop(smm, BOP_DEC, b);
 555 }
 556
 557 static int sm_bootstrap_commit(struct dm_space_map *sm)
 558 {
 559         return 0;
 560 }
 561
 562 static int sm_bootstrap_root_size(struct dm_space_map *sm, size_t *result)
 563 {
 564         DMERR("bootstrap doesn't support root_size");
 565
 566         return -EINVAL;
 567 }
 568
 569 static int sm_bootstrap_copy_root(struct dm_space_map *sm, void *where,
 570                                   size_t max)
 571 {
 572         DMERR("bootstrap doesn't support copy_root");
 573
 574         return -EINVAL;
 575 }
 576
 577 static struct dm_space_map bootstrap_ops = {
 578         .destroy = sm_bootstrap_destroy,
 579         .extend = sm_bootstrap_extend,
 580         .get_nr_blocks = sm_bootstrap_get_nr_blocks,
 581         .get_nr_free = sm_bootstrap_get_nr_free,
 582         .get_count = sm_bootstrap_get_count,
 583         .count_is_more_than_one = sm_bootstrap_count_is_more_than_one,
 584         .set_count = sm_bootstrap_set_count,
 585         .inc_block = sm_bootstrap_inc_block,
 586         .dec_block = sm_bootstrap_dec_block,
 587         .new_block = sm_bootstrap_new_block,
 588         .commit = sm_bootstrap_commit,
 589         .root_size = sm_bootstrap_root_size,
 590         .copy_root = sm_bootstrap_copy_root,
 591         .register_threshold_callback = NULL
 592 };
 593
 594 /*----------------------------------------------------------------*/
 595
 596 static int sm_metadata_extend(struct dm_space_map *sm, dm_block_t extra_blocks)
 597 {
 598         int r, i;
 599         enum allocation_event ev;
 600         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 601         dm_block_t old_len = smm->ll.nr_blocks;
 602
 603         /*
 604          * Flick into a mode where all blocks get allocated in the new area.
 605          */
 606         smm->begin = old_len;
 607         memcpy(&smm->sm, &bootstrap_ops, sizeof(smm->sm));
 608
 609         /*
 610          * Extend.
 611          */
 612         r = sm_ll_extend(&smm->ll, extra_blocks);
 613
 614         /*
 615          * Switch back to normal behaviour.
 616          */
 617         memcpy(&smm->sm, &ops, sizeof(smm->sm));
 618         for (i = old_len; !r && i < smm->begin; i++)
 619                 r = sm_ll_inc(&smm->ll, i, &ev);
 620
 621         return r;
 622 }
 623
 624 /*----------------------------------------------------------------*/
 625
 626 struct dm_space_map *dm_sm_metadata_init(void)
 627 {
 628         struct sm_metadata *smm;
 629
 630         smm = kmalloc(sizeof(*smm), GFP_KERNEL);
 631         if (!smm)
 632                 return ERR_PTR(-ENOMEM);
 633
 634         memcpy(&smm->sm, &ops, sizeof(smm->sm));
 635
 636         return &smm->sm;
 637 }
 638
 639 int dm_sm_metadata_create(struct dm_space_map *sm,
 640                           struct dm_transaction_manager *tm,
 641                           dm_block_t nr_blocks,
 642                           dm_block_t superblock)
 643 {
 644         int r;
 645         dm_block_t i;
 646         enum allocation_event ev;
 647         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 648
 649         smm->begin = superblock + 1;
 650         smm->recursion_count = 0;
 651         smm->allocated_this_transaction = 0;
 652         smm->nr_uncommitted = 0;
 653         threshold_init(&smm->threshold);
 654
 655         memcpy(&smm->sm, &bootstrap_ops, sizeof(smm->sm));
 656
 657         r = sm_ll_new_metadata(&smm->ll, tm);
 658         if (r)
 659                 return r;
 660
 661         r = sm_ll_extend(&smm->ll, nr_blocks);
 662         if (r)
 663                 return r;
 664
 665         memcpy(&smm->sm, &ops, sizeof(smm->sm));
 666
 667         /*
 668          * Now we need to update the newly created data structures with the
 669          * allocated blocks that they were built from.
 670          */
 671         for (i = superblock; !r && i < smm->begin; i++)
 672                 r = sm_ll_inc(&smm->ll, i, &ev);
 673
 674         if (r)
 675                 return r;
 676
 677         return sm_metadata_commit(sm);
 678 }
 679
 680 int dm_sm_metadata_open(struct dm_space_map *sm,
 681                         struct dm_transaction_manager *tm,
 682                         void *root_le, size_t len)
 683 {
 684         int r;
 685         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
 686
 687         r = sm_ll_open_metadata(&smm->ll, tm, root_le, len);
 688         if (r)
 689                 return r;
 690
 691         smm->begin = 0;
 692         smm->recursion_count = 0;
 693         smm->allocated_this_transaction = 0;
 694         smm->nr_uncommitted = 0;
 695         threshold_init(&smm->threshold);
 696
 697         memcpy(&smm->old_ll, &smm->ll, sizeof(smm->old_ll));
 698         return 0;
 699 }