extent_map.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #define _XOPEN_SOURCE 600
  19 #define __USE_XOPEN2K
  20 #include <stdio.h>
  21 #include <stdlib.h>
  22 #include <sys/types.h>
  23 #include <sys/stat.h>
  24 #include <fcntl.h>
  25 #include <unistd.h>
  26 #include "kerncompat.h"
  27 #include "extent_map.h"
  28 #include "list.h"
  29
  30 u64 cache_max = 1024 * 1024 * 32;
  31
  32 void extent_map_tree_init(struct extent_map_tree *tree)
  33 {
  34         cache_tree_init(&tree->state);
  35         cache_tree_init(&tree->cache);
  36         INIT_LIST_HEAD(&tree->lru);
  37         tree->cache_size = 0;
  38 }
  39
  40 static struct extent_state *alloc_extent_state(void)
  41 {
  42         struct extent_state *state;
  43
  44         state = malloc(sizeof(*state));
  45         if (!state)
  46                 return NULL;
  47         state->refs = 1;
  48         state->state = 0;
  49         state->private = 0;
  50         return state;
  51 }
  52
  53 static void free_extent_state(struct extent_state *state)
  54 {
  55         state->refs--;
  56         BUG_ON(state->refs < 0);
  57         if (state->refs == 0)
  58                 free(state);
  59 }
  60
  61 void extent_map_tree_cleanup(struct extent_map_tree *tree)
  62 {
  63         struct extent_state *es;
  64         struct extent_buffer *eb;
  65         struct cache_extent *cache;
  66
  67         while(!list_empty(&tree->lru)) {
  68                 eb = list_entry(tree->lru.next, struct extent_buffer, lru);
  69                 if (eb->refs != 1) {
  70                         fprintf(stderr, "extent buffer leak: "
  71                                 "start %Lu len %u\n", eb->start, eb->len);
  72                         eb->refs = 1;
  73                 }
  74                 free_extent_buffer(eb);
  75         }
  76         while (1) {
  77                 cache = find_first_cache_extent(&tree->state, 0);
  78                 if (!cache)
  79                         break;
  80                 es = container_of(cache, struct extent_state, cache_node);
  81                 remove_cache_extent(&tree->state, &es->cache_node);
  82                 free_extent_state(es);
  83         }
  84 }
  85
  86 static inline void update_extent_state(struct extent_state *state)
  87 {
  88         state->cache_node.start = state->start;
  89         state->cache_node.size = state->end + 1 - state->start;
  90 }
  91
  92 /*
  93  * Utility function to look for merge candidates inside a given range.
  94  * Any extents with matching state are merged together into a single
  95  * extent in the tree. Extents with EXTENT_IO in their state field are
  96  * not merged
  97  */
  98 static int merge_state(struct extent_map_tree *tree,
  99                        struct extent_state *state)
 100 {
 101         struct extent_state *other;
 102         struct cache_extent *other_node;
 103
 104         if (state->state & EXTENT_IOBITS)
 105                 return 0;
 106
 107         other_node = prev_cache_extent(&state->cache_node);
 108         if (other_node) {
 109                 other = container_of(other_node, struct extent_state,
 110                                      cache_node);
 111                 if (other->end == state->start - 1 &&
 112                     other->state == state->state) {
 113                         state->start = other->start;
 114                         update_extent_state(state);
 115                         remove_cache_extent(&tree->state, &other->cache_node);
 116                         free_extent_state(other);
 117                 }
 118         }
 119         other_node = next_cache_extent(&state->cache_node);
 120         if (other_node) {
 121                 other = container_of(other_node, struct extent_state,
 122                                      cache_node);
 123                 if (other->start == state->end + 1 &&
 124                     other->state == state->state) {
 125                         other->start = state->start;
 126                         update_extent_state(state);
 127                         remove_cache_extent(&tree->state, &state->cache_node);
 128                         free_extent_state(state);
 129                 }
 130         }
 131         return 0;
 132 }
 133
 134 /*
 135  * insert an extent_state struct into the tree.  'bits' are set on the
 136  * struct before it is inserted.
 137  */
 138 static int insert_state(struct extent_map_tree *tree,
 139                         struct extent_state *state, u64 start, u64 end,
 140                         int bits)
 141 {
 142         int ret;
 143
 144         BUG_ON(end < start);
 145         state->state |= bits;
 146         state->start = start;
 147         state->end = end;
 148         update_extent_state(state);
 149         ret = insert_existing_cache_extent(&tree->state, &state->cache_node);
 150         BUG_ON(ret);
 151         merge_state(tree, state);
 152         return 0;
 153 }
 154
 155 /*
 156  * split a given extent state struct in two, inserting the preallocated
 157  * struct 'prealloc' as the newly created second half.  'split' indicates an
 158  * offset inside 'orig' where it should be split.
 159  */
 160 static int split_state(struct extent_map_tree *tree, struct extent_state *orig,
 161                        struct extent_state *prealloc, u64 split)
 162 {
 163         int ret;
 164         prealloc->start = orig->start;
 165         prealloc->end = split - 1;
 166         prealloc->state = orig->state;
 167         update_extent_state(prealloc);
 168         orig->start = split;
 169         update_extent_state(orig);
 170         ret = insert_existing_cache_extent(&tree->state,
 171                                            &prealloc->cache_node);
 172         BUG_ON(ret);
 173         return 0;
 174 }
 175
 176 /*
 177  * clear some bits on a range in the tree.
 178  */
 179 static int clear_state_bit(struct extent_map_tree *tree,
 180                             struct extent_state *state, int bits)
 181 {
 182         int ret = state->state & bits;
 183
 184         state->state &= ~bits;
 185         if (state->state == 0) {
 186                 remove_cache_extent(&tree->state, &state->cache_node);
 187                 free_extent_state(state);
 188         } else {
 189                 merge_state(tree, state);
 190         }
 191         return ret;
 192 }
 193
 194 /*
 195  * set some bits on a range in the tree.
 196  */
 197 int clear_extent_bits(struct extent_map_tree *tree, u64 start,
 198                       u64 end, int bits, gfp_t mask)
 199 {
 200         struct extent_state *state;
 201         struct extent_state *prealloc = NULL;
 202         struct cache_extent *node;
 203         int err;
 204         int set = 0;
 205
 206 again:
 207         prealloc = alloc_extent_state();
 208         if (!prealloc)
 209                 return -ENOMEM;
 210
 211         /*
 212          * this search will find the extents that end after
 213          * our range starts
 214          */
 215         node = find_first_cache_extent(&tree->state, start);
 216         if (!node)
 217                 goto out;
 218         state = container_of(node, struct extent_state, cache_node);
 219         if (state->start > end)
 220                 goto out;
 221
 222         /*
 223          *     | ---- desired range ---- |
 224          *  | state | or
 225          *  | ------------- state -------------- |
 226          *
 227          * We need to split the extent we found, and may flip
 228          * bits on second half.
 229          *
 230          * If the extent we found extends past our range, we
 231          * just split and search again.  It'll get split again
 232          * the next time though.
 233          *
 234          * If the extent we found is inside our range, we clear
 235          * the desired bit on it.
 236          */
 237         if (state->start < start) {
 238                 err = split_state(tree, state, prealloc, start);
 239                 BUG_ON(err == -EEXIST);
 240                 prealloc = NULL;
 241                 if (err)
 242                         goto out;
 243                 if (state->end <= end) {
 244                         start = state->end + 1;
 245                         set |= clear_state_bit(tree, state, bits);
 246                 } else {
 247                         start = state->start;
 248                 }
 249                 goto search_again;
 250         }
 251         /*
 252          * | ---- desired range ---- |
 253          *                        | state |
 254          * We need to split the extent, and clear the bit
 255          * on the first half
 256          */
 257         if (state->start <= end && state->end > end) {
 258                 err = split_state(tree, state, prealloc, end + 1);
 259                 BUG_ON(err == -EEXIST);
 260
 261                 set |= clear_state_bit(tree, prealloc, bits);
 262                 prealloc = NULL;
 263                 goto out;
 264         }
 265
 266         start = state->end + 1;
 267         set |= clear_state_bit(tree, state, bits);
 268         goto search_again;
 269 out:
 270         if (prealloc)
 271                 free_extent_state(prealloc);
 272         return set;
 273
 274 search_again:
 275         if (start > end)
 276                 goto out;
 277         goto again;
 278 }
 279
 280 /*
 281  * set some bits on a range in the tree.
 282  */
 283 int set_extent_bits(struct extent_map_tree *tree, u64 start,
 284                     u64 end, int bits, gfp_t mask)
 285 {
 286         struct extent_state *state;
 287         struct extent_state *prealloc = NULL;
 288         struct cache_extent *node;
 289         int err = 0;
 290         int set;
 291         u64 last_start;
 292         u64 last_end;
 293 again:
 294         prealloc = alloc_extent_state();
 295         if (!prealloc)
 296                 return -ENOMEM;
 297
 298         /*
 299          * this search will find the extents that end after
 300          * our range starts
 301          */
 302         node = find_first_cache_extent(&tree->state, start);
 303         if (!node) {
 304                 err = insert_state(tree, prealloc, start, end, bits);
 305                 BUG_ON(err == -EEXIST);
 306                 prealloc = NULL;
 307                 goto out;
 308         }
 309
 310         state = container_of(node, struct extent_state, cache_node);
 311         last_start = state->start;
 312         last_end = state->end;
 313
 314         /*
 315          * | ---- desired range ---- |
 316          * | state |
 317          *
 318          * Just lock what we found and keep going
 319          */
 320         if (state->start == start && state->end <= end) {
 321                 set = state->state & bits;
 322                 state->state |= bits;
 323                 start = state->end + 1;
 324                 merge_state(tree, state);
 325                 goto search_again;
 326         }
 327         /*
 328          *     | ---- desired range ---- |
 329          * | state |
 330          *   or
 331          * | ------------- state -------------- |
 332          *
 333          * We need to split the extent we found, and may flip bits on
 334          * second half.
 335          *
 336          * If the extent we found extends past our
 337          * range, we just split and search again.  It'll get split
 338          * again the next time though.
 339          *
 340          * If the extent we found is inside our range, we set the
 341          * desired bit on it.
 342          */
 343         if (state->start < start) {
 344                 set = state->state & bits;
 345                 err = split_state(tree, state, prealloc, start);
 346                 BUG_ON(err == -EEXIST);
 347                 prealloc = NULL;
 348                 if (err)
 349                         goto out;
 350                 if (state->end <= end) {
 351                         state->state |= bits;
 352                         start = state->end + 1;
 353                         merge_state(tree, state);
 354                 } else {
 355                         start = state->start;
 356                 }
 357                 goto search_again;
 358         }
 359         /*
 360          * | ---- desired range ---- |
 361          *     | state | or               | state |
 362          *
 363          * There's a hole, we need to insert something in it and
 364          * ignore the extent we found.
 365          */
 366         if (state->start > start) {
 367                 u64 this_end;
 368                 if (end < last_start)
 369                         this_end = end;
 370                 else
 371                         this_end = last_start -1;
 372                 err = insert_state(tree, prealloc, start, this_end,
 373                                 bits);
 374                 BUG_ON(err == -EEXIST);
 375                 prealloc = NULL;
 376                 if (err)
 377                         goto out;
 378                 start = this_end + 1;
 379                 goto search_again;
 380         }
 381         /*
 382          * | ---- desired range ---- |
 383          * | ---------- state ---------- |
 384          * We need to split the extent, and set the bit
 385          * on the first half
 386          */
 387         set = state->state & bits;
 388         err = split_state(tree, state, prealloc, end + 1);
 389         BUG_ON(err == -EEXIST);
 390
 391         state->state |= bits;
 392         merge_state(tree, prealloc);
 393         prealloc = NULL;
 394 out:
 395         if (prealloc)
 396                 free_extent_state(prealloc);
 397         return err;
 398 search_again:
 399         if (start > end)
 400                 goto out;
 401         goto again;
 402 }
 403
 404 int set_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
 405                      gfp_t mask)
 406 {
 407         return set_extent_bits(tree, start, end, EXTENT_DIRTY, mask);
 408 }
 409
 410 int clear_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
 411                        gfp_t mask)
 412 {
 413         return clear_extent_bits(tree, start, end, EXTENT_DIRTY, mask);
 414 }
 415
 416 int find_first_extent_bit(struct extent_map_tree *tree, u64 start,
 417                           u64 *start_ret, u64 *end_ret, int bits)
 418 {
 419         struct cache_extent *node;
 420         struct extent_state *state;
 421         int ret = 1;
 422
 423         /*
 424          * this search will find all the extents that end after
 425          * our range starts.
 426          */
 427         node = find_first_cache_extent(&tree->state, start);
 428         if (!node)
 429                 goto out;
 430
 431         while(1) {
 432                 state = container_of(node, struct extent_state, cache_node);
 433                 if (state->end >= start && (state->state & bits)) {
 434                         *start_ret = state->start;
 435                         *end_ret = state->end;
 436                         ret = 0;
 437                         break;
 438                 }
 439                 node = next_cache_extent(node);
 440                 if (!node)
 441                         break;
 442         }
 443 out:
 444         return ret;
 445 }
 446
 447 int test_range_bit(struct extent_map_tree *tree, u64 start, u64 end,
 448                    int bits, int filled)
 449 {
 450         struct extent_state *state = NULL;
 451         struct cache_extent *node;
 452         int bitset = 0;
 453
 454         node = find_first_cache_extent(&tree->state, start);
 455         while (node && start <= end) {
 456                 state = container_of(node, struct extent_state, cache_node);
 457
 458                 if (filled && state->start > start) {
 459                         bitset = 0;
 460                         break;
 461                 }
 462                 if (state->start > end)
 463                         break;
 464                 if (state->state & bits) {
 465                         bitset = 1;
 466                         if (!filled)
 467                                 break;
 468                 } else if (filled) {
 469                         bitset = 0;
 470                         break;
 471                 }
 472                 start = state->end + 1;
 473                 if (start > end)
 474                         break;
 475                 node = next_cache_extent(node);
 476         }
 477         return bitset;
 478 }
 479
 480 int set_state_private(struct extent_map_tree *tree, u64 start, u64 private)
 481 {
 482         struct cache_extent *node;
 483         struct extent_state *state;
 484         int ret = 0;
 485
 486         node = find_first_cache_extent(&tree->state, start);
 487         if (!node) {
 488                 ret = -ENOENT;
 489                 goto out;
 490         }
 491         state = container_of(node, struct extent_state, cache_node);
 492         if (state->start != start) {
 493                 ret = -ENOENT;
 494                 goto out;
 495         }
 496         state->private = private;
 497 out:
 498         return ret;
 499 }
 500
 501 int get_state_private(struct extent_map_tree *tree, u64 start, u64 *private)
 502 {
 503         struct cache_extent *node;
 504         struct extent_state *state;
 505         int ret = 0;
 506
 507         node = find_first_cache_extent(&tree->state, start);
 508         if (!node) {
 509                 ret = -ENOENT;
 510                 goto out;
 511         }
 512         state = container_of(node, struct extent_state, cache_node);
 513         if (state->start != start) {
 514                 ret = -ENOENT;
 515                 goto out;
 516         }
 517         *private = state->private;
 518 out:
 519         return ret;
 520 }
 521
 522 static int free_some_buffers(struct extent_map_tree *tree)
 523 {
 524         u32 nrscan = 0;
 525         struct extent_buffer *eb;
 526         struct list_head *node, *next;
 527
 528         if (tree->cache_size < cache_max)
 529                 return 0;
 530         list_for_each_safe(node, next, &tree->lru) {
 531                 eb = list_entry(node, struct extent_buffer, lru);
 532                 if (eb->refs == 1) {
 533                         free_extent_buffer(eb);
 534                         if (tree->cache_size < cache_max)
 535                                 break;
 536                 }
 537                 if (nrscan++ > 64)
 538                         break;
 539         }
 540         return 0;
 541 }
 542
 543 static struct extent_buffer *__alloc_extent_buffer(struct extent_map_tree *tree,
 544                                                    u64 bytenr, u32 blocksize)
 545 {
 546         struct extent_buffer *eb;
 547         int ret;
 548
 549         eb = malloc(sizeof(struct extent_buffer) + blocksize);
 550         if (!eb)
 551                 return NULL;
 552
 553         eb->start = bytenr;
 554         eb->len = blocksize;
 555         eb->refs = 2;
 556         eb->flags = 0;
 557         eb->tree = tree;
 558         eb->fd = -1;
 559         eb->dev_bytenr = (u64)-1;
 560         eb->cache_node.start = bytenr;
 561         eb->cache_node.size = blocksize;
 562
 563         free_some_buffers(tree);
 564         ret = insert_existing_cache_extent(&tree->cache, &eb->cache_node);
 565         if (ret) {
 566                 free(eb);
 567                 return NULL;
 568         }
 569         list_add_tail(&eb->lru, &tree->lru);
 570         tree->cache_size += blocksize;
 571         return eb;
 572 }
 573
 574 void free_extent_buffer(struct extent_buffer *eb)
 575 {
 576         if (!eb)
 577                 return;
 578
 579         eb->refs--;
 580         BUG_ON(eb->refs < 0);
 581         if (eb->refs == 0) {
 582                 struct extent_map_tree *tree = eb->tree;
 583                 BUG_ON(eb->flags & EXTENT_DIRTY);
 584                 list_del_init(&eb->lru);
 585                 remove_cache_extent(&tree->cache, &eb->cache_node);
 586                 BUG_ON(tree->cache_size < eb->len);
 587                 tree->cache_size -= eb->len;
 588                 free(eb);
 589         }
 590 }
 591
 592 struct extent_buffer *find_extent_buffer(struct extent_map_tree *tree,
 593                                          u64 bytenr, u32 blocksize)
 594 {
 595         struct extent_buffer *eb = NULL;
 596         struct cache_extent *cache;
 597
 598         cache = find_cache_extent(&tree->cache, bytenr, blocksize);
 599         if (cache && cache->start == bytenr && cache->size == blocksize) {
 600                 eb = container_of(cache, struct extent_buffer, cache_node);
 601                 list_move_tail(&eb->lru, &tree->lru);
 602                 eb->refs++;
 603         }
 604         return eb;
 605 }
 606
 607 struct extent_buffer *find_first_extent_buffer(struct extent_map_tree *tree,
 608                                                u64 start)
 609 {
 610         struct extent_buffer *eb = NULL;
 611         struct cache_extent *cache;
 612
 613         cache = find_first_cache_extent(&tree->cache, start);
 614         if (cache) {
 615                 eb = container_of(cache, struct extent_buffer, cache_node);
 616                 list_move_tail(&eb->lru, &tree->lru);
 617                 eb->refs++;
 618         }
 619         return eb;
 620 }
 621
 622 struct extent_buffer *alloc_extent_buffer(struct extent_map_tree *tree,
 623                                           u64 bytenr, u32 blocksize)
 624 {
 625         struct extent_buffer *eb;
 626         struct cache_extent *cache;
 627
 628         cache = find_cache_extent(&tree->cache, bytenr, blocksize);
 629         if (cache && cache->start == bytenr && cache->size == blocksize) {
 630                 eb = container_of(cache, struct extent_buffer, cache_node);
 631                 list_move_tail(&eb->lru, &tree->lru);
 632                 eb->refs++;
 633         } else {
 634                 if (cache) {
 635                         eb = container_of(cache, struct extent_buffer,
 636                                           cache_node);
 637                         BUG_ON(eb->refs != 1);
 638                         free_extent_buffer(eb);
 639                 }
 640                 eb = __alloc_extent_buffer(tree, bytenr, blocksize);
 641         }
 642         return eb;
 643 }
 644
 645 int read_extent_from_disk(struct extent_buffer *eb)
 646 {
 647         int ret;
 648         ret = pread(eb->fd, eb->data, eb->len, eb->dev_bytenr);
 649         if (ret < 0)
 650                 goto out;
 651         if (ret != eb->len) {
 652                 ret = -EIO;
 653                 goto out;
 654         }
 655         ret = 0;
 656 out:
 657         return ret;
 658 }
 659
 660 int write_extent_to_disk(struct extent_buffer *eb)
 661 {
 662         int ret;
 663         ret = pwrite(eb->fd, eb->data, eb->len, eb->dev_bytenr);
 664         if (ret < 0)
 665                 goto out;
 666         if (ret != eb->len) {
 667                 ret = -EIO;
 668                 goto out;
 669         }
 670         ret = 0;
 671 out:
 672         return ret;
 673 }
 674
 675 int set_extent_buffer_uptodate(struct extent_buffer *eb)
 676 {
 677         eb->flags |= EXTENT_UPTODATE;
 678         return 0;
 679 }
 680
 681 int extent_buffer_uptodate(struct extent_buffer *eb)
 682 {
 683         if (eb->flags & EXTENT_UPTODATE)
 684                 return 1;
 685         return 0;
 686 }
 687
 688 int set_extent_buffer_dirty(struct extent_buffer *eb)
 689 {
 690         struct extent_map_tree *tree = eb->tree;
 691         if (!(eb->flags & EXTENT_DIRTY)) {
 692                 eb->flags |= EXTENT_DIRTY;
 693                 set_extent_dirty(tree, eb->start, eb->start + eb->len - 1, 0);
 694                 extent_buffer_get(eb);
 695         }
 696         return 0;
 697 }
 698
 699 int clear_extent_buffer_dirty(struct extent_buffer *eb)
 700 {
 701         struct extent_map_tree *tree = eb->tree;
 702         if (eb->flags & EXTENT_DIRTY) {
 703                 eb->flags &= ~EXTENT_DIRTY;
 704                 clear_extent_dirty(tree, eb->start, eb->start + eb->len - 1, 0);
 705                 free_extent_buffer(eb);
 706         }
 707         return 0;
 708 }
 709
 710 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
 711                          unsigned long start, unsigned long len)
 712 {
 713         return memcmp(eb->data + start, ptrv, len);
 714 }
 715
 716 void read_extent_buffer(struct extent_buffer *eb, void *dst,
 717                         unsigned long start, unsigned long len)
 718 {
 719         memcpy(dst, eb->data + start, len);
 720 }
 721
 722 void write_extent_buffer(struct extent_buffer *eb, const void *src,
 723                          unsigned long start, unsigned long len)
 724 {
 725         memcpy(eb->data + start, src, len);
 726 }
 727
 728 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
 729                         unsigned long dst_offset, unsigned long src_offset,
 730                         unsigned long len)
 731 {
 732         memcpy(dst->data + dst_offset, src->data + src_offset, len);
 733 }
 734
 735 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
 736                           unsigned long src_offset, unsigned long len)
 737 {
 738         memcpy(dst->data + dst_offset, dst->data + src_offset, len);
 739 }
 740
 741 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
 742                            unsigned long src_offset, unsigned long len)
 743 {
 744         memmove(dst->data + dst_offset, dst->data + src_offset, len);
 745 }
 746
 747 void memset_extent_buffer(struct extent_buffer *eb, char c,
 748                           unsigned long start, unsigned long len)
 749 {
 750         memset(eb->data + start, c, len);
 751 }