subr_blist.c revision 319965
1/*- 2 * Copyright (c) 1998 Matthew Dillon. All Rights Reserved. 3 * Redistribution and use in source and binary forms, with or without 4 * modification, are permitted provided that the following conditions 5 * are met: 6 * 1. Redistributions of source code must retain the above copyright 7 * notice, this list of conditions and the following disclaimer. 8 * 2. Redistributions in binary form must reproduce the above copyright 9 * notice, this list of conditions and the following disclaimer in the 10 * documentation and/or other materials provided with the distribution. 11 * 4. Neither the name of the University nor the names of its contributors 12 * may be used to endorse or promote products derived from this software 13 * without specific prior written permission. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS 16 * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 17 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY 19 * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE 21 * GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 22 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 23 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 24 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 25 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 */ 27/* 28 * BLIST.C - Bitmap allocator/deallocator, using a radix tree with hinting 29 * 30 * This module implements a general bitmap allocator/deallocator. The 31 * allocator eats around 2 bits per 'block'. The module does not 32 * try to interpret the meaning of a 'block' other than to return 33 * SWAPBLK_NONE on an allocation failure. 34 * 35 * A radix tree is used to maintain the bitmap. Two radix constants are 36 * involved: One for the bitmaps contained in the leaf nodes (typically 37 * 32), and one for the meta nodes (typically 16). Both meta and leaf 38 * nodes have a hint field. This field gives us a hint as to the largest 39 * free contiguous range of blocks under the node. It may contain a 40 * value that is too high, but will never contain a value that is too 41 * low. When the radix tree is searched, allocation failures in subtrees 42 * update the hint. 43 * 44 * The radix tree also implements two collapsed states for meta nodes: 45 * the ALL-ALLOCATED state and the ALL-FREE state. If a meta node is 46 * in either of these two states, all information contained underneath 47 * the node is considered stale. These states are used to optimize 48 * allocation and freeing operations. 49 * 50 * The hinting greatly increases code efficiency for allocations while 51 * the general radix structure optimizes both allocations and frees. The 52 * radix tree should be able to operate well no matter how much 53 * fragmentation there is and no matter how large a bitmap is used. 54 * 55 * The blist code wires all necessary memory at creation time. Neither 56 * allocations nor frees require interaction with the memory subsystem. 57 * The non-blocking features of the blist code are used in the swap code 58 * (vm/swap_pager.c). 59 * 60 * LAYOUT: The radix tree is laid out recursively using a 61 * linear array. Each meta node is immediately followed (laid out 62 * sequentially in memory) by BLIST_META_RADIX lower level nodes. This 63 * is a recursive structure but one that can be easily scanned through 64 * a very simple 'skip' calculation. In order to support large radixes, 65 * portions of the tree may reside outside our memory allocation. We 66 * handle this with an early-termination optimization (when bighint is 67 * set to -1) on the scan. The memory allocation is only large enough 68 * to cover the number of blocks requested at creation time even if it 69 * must be encompassed in larger root-node radix. 70 * 71 * NOTE: the allocator cannot currently allocate more than 72 * BLIST_BMAP_RADIX blocks per call. It will panic with 'allocation too 73 * large' if you try. This is an area that could use improvement. The 74 * radix is large enough that this restriction does not effect the swap 75 * system, though. Currently only the allocation code is effected by 76 * this algorithmic unfeature. The freeing code can handle arbitrary 77 * ranges. 78 * 79 * This code can be compiled stand-alone for debugging. 80 */ 81 82#include <sys/cdefs.h> 83__FBSDID("$FreeBSD: stable/10/sys/kern/subr_blist.c 319965 2017-06-15 03:58:23Z alc $"); 84 85#ifdef _KERNEL 86 87#include <sys/param.h> 88#include <sys/systm.h> 89#include <sys/lock.h> 90#include <sys/kernel.h> 91#include <sys/blist.h> 92#include <sys/malloc.h> 93#include <sys/proc.h> 94#include <sys/mutex.h> 95 96#else 97 98#ifndef BLIST_NO_DEBUG 99#define BLIST_DEBUG 100#endif 101 102#include <sys/types.h> 103#include <sys/malloc.h> 104#include <stdio.h> 105#include <string.h> 106#include <stdlib.h> 107#include <stdarg.h> 108 109#define malloc(a,b,c) calloc(a, 1) 110#define free(a,b) free(a) 111 112#include <sys/blist.h> 113 114void panic(const char *ctl, ...); 115 116#endif 117 118/* 119 * static support functions 120 */ 121 122static daddr_t blst_leaf_alloc(blmeta_t *scan, daddr_t blk, int count); 123static daddr_t blst_meta_alloc(blmeta_t *scan, daddr_t blk, 124 daddr_t count, daddr_t radix, int skip); 125static void blst_leaf_free(blmeta_t *scan, daddr_t relblk, int count); 126static void blst_meta_free(blmeta_t *scan, daddr_t freeBlk, daddr_t count, 127 daddr_t radix, int skip, daddr_t blk); 128static void blst_copy(blmeta_t *scan, daddr_t blk, daddr_t radix, 129 daddr_t skip, blist_t dest, daddr_t count); 130static int blst_leaf_fill(blmeta_t *scan, daddr_t blk, int count); 131static int blst_meta_fill(blmeta_t *scan, daddr_t allocBlk, daddr_t count, 132 daddr_t radix, int skip, daddr_t blk); 133static daddr_t blst_radix_init(blmeta_t *scan, daddr_t radix, 134 int skip, daddr_t count); 135#ifndef _KERNEL 136static void blst_radix_print(blmeta_t *scan, daddr_t blk, 137 daddr_t radix, int skip, int tab); 138#endif 139 140#ifdef _KERNEL 141static MALLOC_DEFINE(M_SWAP, "SWAP", "Swap space"); 142#endif 143 144/* 145 * blist_create() - create a blist capable of handling up to the specified 146 * number of blocks 147 * 148 * blocks - must be greater than 0 149 * flags - malloc flags 150 * 151 * The smallest blist consists of a single leaf node capable of 152 * managing BLIST_BMAP_RADIX blocks. 153 */ 154 155blist_t 156blist_create(daddr_t blocks, int flags) 157{ 158 blist_t bl; 159 int radix; 160 int skip = 0; 161 162 /* 163 * Calculate radix and skip field used for scanning. 164 */ 165 radix = BLIST_BMAP_RADIX; 166 167 while (radix < blocks) { 168 radix *= BLIST_META_RADIX; 169 skip = (skip + 1) * BLIST_META_RADIX; 170 } 171 172 bl = malloc(sizeof(struct blist), M_SWAP, flags | M_ZERO); 173 174 bl->bl_blocks = blocks; 175 bl->bl_radix = radix; 176 bl->bl_skip = skip; 177 bl->bl_rootblks = 1 + 178 blst_radix_init(NULL, bl->bl_radix, bl->bl_skip, blocks); 179 bl->bl_root = malloc(sizeof(blmeta_t) * bl->bl_rootblks, M_SWAP, flags); 180 181#if defined(BLIST_DEBUG) 182 printf( 183 "BLIST representing %lld blocks (%lld MB of swap)" 184 ", requiring %lldK of ram\n", 185 (long long)bl->bl_blocks, 186 (long long)bl->bl_blocks * 4 / 1024, 187 (long long)(bl->bl_rootblks * sizeof(blmeta_t) + 1023) / 1024 188 ); 189 printf("BLIST raw radix tree contains %lld records\n", 190 (long long)bl->bl_rootblks); 191#endif 192 blst_radix_init(bl->bl_root, bl->bl_radix, bl->bl_skip, blocks); 193 194 return(bl); 195} 196 197void 198blist_destroy(blist_t bl) 199{ 200 free(bl->bl_root, M_SWAP); 201 free(bl, M_SWAP); 202} 203 204/* 205 * blist_alloc() - reserve space in the block bitmap. Return the base 206 * of a contiguous region or SWAPBLK_NONE if space could 207 * not be allocated. 208 */ 209 210daddr_t 211blist_alloc(blist_t bl, daddr_t count) 212{ 213 daddr_t blk = SWAPBLK_NONE; 214 215 if (bl) { 216 if (bl->bl_radix == BLIST_BMAP_RADIX) 217 blk = blst_leaf_alloc(bl->bl_root, 0, count); 218 else 219 blk = blst_meta_alloc(bl->bl_root, 0, count, bl->bl_radix, bl->bl_skip); 220 if (blk != SWAPBLK_NONE) 221 bl->bl_free -= count; 222 } 223 return(blk); 224} 225 226/* 227 * blist_free() - free up space in the block bitmap. Return the base 228 * of a contiguous region. Panic if an inconsistancy is 229 * found. 230 */ 231 232void 233blist_free(blist_t bl, daddr_t blkno, daddr_t count) 234{ 235 if (bl) { 236 if (bl->bl_radix == BLIST_BMAP_RADIX) 237 blst_leaf_free(bl->bl_root, blkno, count); 238 else 239 blst_meta_free(bl->bl_root, blkno, count, bl->bl_radix, bl->bl_skip, 0); 240 bl->bl_free += count; 241 } 242} 243 244/* 245 * blist_fill() - mark a region in the block bitmap as off-limits 246 * to the allocator (i.e. allocate it), ignoring any 247 * existing allocations. Return the number of blocks 248 * actually filled that were free before the call. 249 */ 250 251int 252blist_fill(blist_t bl, daddr_t blkno, daddr_t count) 253{ 254 int filled; 255 256 if (bl) { 257 if (bl->bl_radix == BLIST_BMAP_RADIX) 258 filled = blst_leaf_fill(bl->bl_root, blkno, count); 259 else 260 filled = blst_meta_fill(bl->bl_root, blkno, count, 261 bl->bl_radix, bl->bl_skip, 0); 262 bl->bl_free -= filled; 263 return filled; 264 } else 265 return 0; 266} 267 268/* 269 * blist_resize() - resize an existing radix tree to handle the 270 * specified number of blocks. This will reallocate 271 * the tree and transfer the previous bitmap to the new 272 * one. When extending the tree you can specify whether 273 * the new blocks are to left allocated or freed. 274 */ 275 276void 277blist_resize(blist_t *pbl, daddr_t count, int freenew, int flags) 278{ 279 blist_t newbl = blist_create(count, flags); 280 blist_t save = *pbl; 281 282 *pbl = newbl; 283 if (count > save->bl_blocks) 284 count = save->bl_blocks; 285 blst_copy(save->bl_root, 0, save->bl_radix, save->bl_skip, newbl, count); 286 287 /* 288 * If resizing upwards, should we free the new space or not? 289 */ 290 if (freenew && count < newbl->bl_blocks) { 291 blist_free(newbl, count, newbl->bl_blocks - count); 292 } 293 blist_destroy(save); 294} 295 296#ifdef BLIST_DEBUG 297 298/* 299 * blist_print() - dump radix tree 300 */ 301 302void 303blist_print(blist_t bl) 304{ 305 printf("BLIST {\n"); 306 blst_radix_print(bl->bl_root, 0, bl->bl_radix, bl->bl_skip, 4); 307 printf("}\n"); 308} 309 310#endif 311 312/************************************************************************ 313 * ALLOCATION SUPPORT FUNCTIONS * 314 ************************************************************************ 315 * 316 * These support functions do all the actual work. They may seem 317 * rather longish, but that's because I've commented them up. The 318 * actual code is straight forward. 319 * 320 */ 321 322/* 323 * blist_leaf_alloc() - allocate at a leaf in the radix tree (a bitmap). 324 * 325 * This is the core of the allocator and is optimized for the 1 block 326 * and the BLIST_BMAP_RADIX block allocation cases. Other cases are 327 * somewhat slower. The 1 block allocation case is log2 and extremely 328 * quick. 329 */ 330 331static daddr_t 332blst_leaf_alloc( 333 blmeta_t *scan, 334 daddr_t blk, 335 int count 336) { 337 u_daddr_t orig = scan->u.bmu_bitmap; 338 339 if (orig == 0) { 340 /* 341 * Optimize bitmap all-allocated case. Also, count = 1 342 * case assumes at least 1 bit is free in the bitmap, so 343 * we have to take care of this case here. 344 */ 345 scan->bm_bighint = 0; 346 return(SWAPBLK_NONE); 347 } 348 if (count == 1) { 349 /* 350 * Optimized code to allocate one bit out of the bitmap 351 */ 352 u_daddr_t mask; 353 int j = BLIST_BMAP_RADIX/2; 354 int r = 0; 355 356 mask = (u_daddr_t)-1 >> (BLIST_BMAP_RADIX/2); 357 358 while (j) { 359 if ((orig & mask) == 0) { 360 r += j; 361 orig >>= j; 362 } 363 j >>= 1; 364 mask >>= j; 365 } 366 scan->u.bmu_bitmap &= ~((u_daddr_t)1 << r); 367 return(blk + r); 368 } 369 if (count <= BLIST_BMAP_RADIX) { 370 /* 371 * non-optimized code to allocate N bits out of the bitmap. 372 * The more bits, the faster the code runs. It will run 373 * the slowest allocating 2 bits, but since there aren't any 374 * memory ops in the core loop (or shouldn't be, anyway), 375 * you probably won't notice the difference. 376 */ 377 int j; 378 int n = BLIST_BMAP_RADIX - count; 379 u_daddr_t mask; 380 381 mask = (u_daddr_t)-1 >> n; 382 383 for (j = 0; j <= n; ++j) { 384 if ((orig & mask) == mask) { 385 scan->u.bmu_bitmap &= ~mask; 386 return(blk + j); 387 } 388 mask = (mask << 1); 389 } 390 } 391 /* 392 * We couldn't allocate count in this subtree, update bighint. 393 */ 394 scan->bm_bighint = count - 1; 395 return(SWAPBLK_NONE); 396} 397 398/* 399 * blist_meta_alloc() - allocate at a meta in the radix tree. 400 * 401 * Attempt to allocate at a meta node. If we can't, we update 402 * bighint and return a failure. Updating bighint optimize future 403 * calls that hit this node. We have to check for our collapse cases 404 * and we have a few optimizations strewn in as well. 405 */ 406 407static daddr_t 408blst_meta_alloc( 409 blmeta_t *scan, 410 daddr_t blk, 411 daddr_t count, 412 daddr_t radix, 413 int skip 414) { 415 int i; 416 int next_skip = ((u_int)skip / BLIST_META_RADIX); 417 418 if (scan->u.bmu_avail == 0) { 419 /* 420 * ALL-ALLOCATED special case 421 */ 422 scan->bm_bighint = count; 423 return(SWAPBLK_NONE); 424 } 425 426 if (scan->u.bmu_avail == radix) { 427 radix /= BLIST_META_RADIX; 428 429 /* 430 * ALL-FREE special case, initialize uninitialize 431 * sublevel. 432 */ 433 for (i = 1; i <= skip; i += next_skip) { 434 if (scan[i].bm_bighint == (daddr_t)-1) 435 break; 436 if (next_skip == 1) { 437 scan[i].u.bmu_bitmap = (u_daddr_t)-1; 438 scan[i].bm_bighint = BLIST_BMAP_RADIX; 439 } else { 440 scan[i].bm_bighint = radix; 441 scan[i].u.bmu_avail = radix; 442 } 443 } 444 } else { 445 radix /= BLIST_META_RADIX; 446 } 447 448 for (i = 1; i <= skip; i += next_skip) { 449 if (count <= scan[i].bm_bighint) { 450 /* 451 * count fits in object 452 */ 453 daddr_t r; 454 if (next_skip == 1) { 455 r = blst_leaf_alloc(&scan[i], blk, count); 456 } else { 457 r = blst_meta_alloc(&scan[i], blk, count, radix, next_skip - 1); 458 } 459 if (r != SWAPBLK_NONE) { 460 scan->u.bmu_avail -= count; 461 if (scan->bm_bighint > scan->u.bmu_avail) 462 scan->bm_bighint = scan->u.bmu_avail; 463 return(r); 464 } 465 } else if (scan[i].bm_bighint == (daddr_t)-1) { 466 /* 467 * Terminator 468 */ 469 break; 470 } else if (count > radix) { 471 /* 472 * count does not fit in object even if it were 473 * complete free. 474 */ 475 panic("blist_meta_alloc: allocation too large"); 476 } 477 blk += radix; 478 } 479 480 /* 481 * We couldn't allocate count in this subtree, update bighint. 482 */ 483 if (scan->bm_bighint >= count) 484 scan->bm_bighint = count - 1; 485 return(SWAPBLK_NONE); 486} 487 488/* 489 * BLST_LEAF_FREE() - free allocated block from leaf bitmap 490 * 491 */ 492 493static void 494blst_leaf_free( 495 blmeta_t *scan, 496 daddr_t blk, 497 int count 498) { 499 /* 500 * free some data in this bitmap 501 * 502 * e.g. 503 * 0000111111111110000 504 * \_________/\__/ 505 * v n 506 */ 507 int n = blk & (BLIST_BMAP_RADIX - 1); 508 u_daddr_t mask; 509 510 mask = ((u_daddr_t)-1 << n) & 511 ((u_daddr_t)-1 >> (BLIST_BMAP_RADIX - count - n)); 512 513 if (scan->u.bmu_bitmap & mask) 514 panic("blst_radix_free: freeing free block"); 515 scan->u.bmu_bitmap |= mask; 516 517 /* 518 * We could probably do a better job here. We are required to make 519 * bighint at least as large as the biggest contiguous block of 520 * data. If we just shoehorn it, a little extra overhead will 521 * be incured on the next allocation (but only that one typically). 522 */ 523 scan->bm_bighint = BLIST_BMAP_RADIX; 524} 525 526/* 527 * BLST_META_FREE() - free allocated blocks from radix tree meta info 528 * 529 * This support routine frees a range of blocks from the bitmap. 530 * The range must be entirely enclosed by this radix node. If a 531 * meta node, we break the range down recursively to free blocks 532 * in subnodes (which means that this code can free an arbitrary 533 * range whereas the allocation code cannot allocate an arbitrary 534 * range). 535 */ 536 537static void 538blst_meta_free( 539 blmeta_t *scan, 540 daddr_t freeBlk, 541 daddr_t count, 542 daddr_t radix, 543 int skip, 544 daddr_t blk 545) { 546 int i; 547 int next_skip = ((u_int)skip / BLIST_META_RADIX); 548 549#if 0 550 printf("free (%llx,%lld) FROM (%llx,%lld)\n", 551 (long long)freeBlk, (long long)count, 552 (long long)blk, (long long)radix 553 ); 554#endif 555 556 if (scan->u.bmu_avail == 0) { 557 /* 558 * ALL-ALLOCATED special case, with possible 559 * shortcut to ALL-FREE special case. 560 */ 561 scan->u.bmu_avail = count; 562 scan->bm_bighint = count; 563 564 if (count != radix) { 565 for (i = 1; i <= skip; i += next_skip) { 566 if (scan[i].bm_bighint == (daddr_t)-1) 567 break; 568 scan[i].bm_bighint = 0; 569 if (next_skip == 1) { 570 scan[i].u.bmu_bitmap = 0; 571 } else { 572 scan[i].u.bmu_avail = 0; 573 } 574 } 575 /* fall through */ 576 } 577 } else { 578 scan->u.bmu_avail += count; 579 /* scan->bm_bighint = radix; */ 580 } 581 582 /* 583 * ALL-FREE special case. 584 */ 585 586 if (scan->u.bmu_avail == radix) 587 return; 588 if (scan->u.bmu_avail > radix) 589 panic("blst_meta_free: freeing already free blocks (%lld) %lld/%lld", 590 (long long)count, (long long)scan->u.bmu_avail, 591 (long long)radix); 592 593 /* 594 * Break the free down into its components 595 */ 596 597 radix /= BLIST_META_RADIX; 598 599 i = (freeBlk - blk) / radix; 600 blk += i * radix; 601 i = i * next_skip + 1; 602 603 while (i <= skip && blk < freeBlk + count) { 604 daddr_t v; 605 606 v = blk + radix - freeBlk; 607 if (v > count) 608 v = count; 609 610 if (scan->bm_bighint == (daddr_t)-1) 611 panic("blst_meta_free: freeing unexpected range"); 612 613 if (next_skip == 1) { 614 blst_leaf_free(&scan[i], freeBlk, v); 615 } else { 616 blst_meta_free(&scan[i], freeBlk, v, radix, next_skip - 1, blk); 617 } 618 if (scan->bm_bighint < scan[i].bm_bighint) 619 scan->bm_bighint = scan[i].bm_bighint; 620 count -= v; 621 freeBlk += v; 622 blk += radix; 623 i += next_skip; 624 } 625} 626 627/* 628 * BLIST_RADIX_COPY() - copy one radix tree to another 629 * 630 * Locates free space in the source tree and frees it in the destination 631 * tree. The space may not already be free in the destination. 632 */ 633 634static void blst_copy( 635 blmeta_t *scan, 636 daddr_t blk, 637 daddr_t radix, 638 daddr_t skip, 639 blist_t dest, 640 daddr_t count 641) { 642 int next_skip; 643 int i; 644 645 /* 646 * Leaf node 647 */ 648 649 if (radix == BLIST_BMAP_RADIX) { 650 u_daddr_t v = scan->u.bmu_bitmap; 651 652 if (v == (u_daddr_t)-1) { 653 blist_free(dest, blk, count); 654 } else if (v != 0) { 655 int i; 656 657 for (i = 0; i < BLIST_BMAP_RADIX && i < count; ++i) { 658 if (v & ((u_daddr_t)1 << i)) 659 blist_free(dest, blk + i, 1); 660 } 661 } 662 return; 663 } 664 665 /* 666 * Meta node 667 */ 668 669 if (scan->u.bmu_avail == 0) { 670 /* 671 * Source all allocated, leave dest allocated 672 */ 673 return; 674 } 675 if (scan->u.bmu_avail == radix) { 676 /* 677 * Source all free, free entire dest 678 */ 679 if (count < radix) 680 blist_free(dest, blk, count); 681 else 682 blist_free(dest, blk, radix); 683 return; 684 } 685 686 687 radix /= BLIST_META_RADIX; 688 next_skip = ((u_int)skip / BLIST_META_RADIX); 689 690 for (i = 1; count && i <= skip; i += next_skip) { 691 if (scan[i].bm_bighint == (daddr_t)-1) 692 break; 693 694 if (count >= radix) { 695 blst_copy( 696 &scan[i], 697 blk, 698 radix, 699 next_skip - 1, 700 dest, 701 radix 702 ); 703 count -= radix; 704 } else { 705 if (count) { 706 blst_copy( 707 &scan[i], 708 blk, 709 radix, 710 next_skip - 1, 711 dest, 712 count 713 ); 714 } 715 count = 0; 716 } 717 blk += radix; 718 } 719} 720 721/* 722 * BLST_LEAF_FILL() - allocate specific blocks in leaf bitmap 723 * 724 * This routine allocates all blocks in the specified range 725 * regardless of any existing allocations in that range. Returns 726 * the number of blocks allocated by the call. 727 */ 728 729static int 730blst_leaf_fill(blmeta_t *scan, daddr_t blk, int count) 731{ 732 int n = blk & (BLIST_BMAP_RADIX - 1); 733 int nblks; 734 u_daddr_t mask, bitmap; 735 736 mask = ((u_daddr_t)-1 << n) & 737 ((u_daddr_t)-1 >> (BLIST_BMAP_RADIX - count - n)); 738 739 /* Count the number of blocks we're about to allocate */ 740 bitmap = scan->u.bmu_bitmap & mask; 741 for (nblks = 0; bitmap != 0; nblks++) 742 bitmap &= bitmap - 1; 743 744 scan->u.bmu_bitmap &= ~mask; 745 return nblks; 746} 747 748/* 749 * BLIST_META_FILL() - allocate specific blocks at a meta node 750 * 751 * This routine allocates the specified range of blocks, 752 * regardless of any existing allocations in the range. The 753 * range must be within the extent of this node. Returns the 754 * number of blocks allocated by the call. 755 */ 756static int 757blst_meta_fill( 758 blmeta_t *scan, 759 daddr_t allocBlk, 760 daddr_t count, 761 daddr_t radix, 762 int skip, 763 daddr_t blk 764) { 765 int i; 766 int next_skip = ((u_int)skip / BLIST_META_RADIX); 767 int nblks = 0; 768 769 if (count > radix) 770 panic("blist_meta_fill: allocation too large"); 771 if (count == radix || scan->u.bmu_avail == 0) { 772 /* 773 * ALL-ALLOCATED special case 774 */ 775 nblks = scan->u.bmu_avail; 776 scan->u.bmu_avail = 0; 777 scan->bm_bighint = count; 778 return nblks; 779 } 780 781 if (scan->u.bmu_avail == radix) { 782 radix /= BLIST_META_RADIX; 783 784 /* 785 * ALL-FREE special case, initialize sublevel 786 */ 787 for (i = 1; i <= skip; i += next_skip) { 788 if (scan[i].bm_bighint == (daddr_t)-1) 789 break; 790 if (next_skip == 1) { 791 scan[i].u.bmu_bitmap = (u_daddr_t)-1; 792 scan[i].bm_bighint = BLIST_BMAP_RADIX; 793 } else { 794 scan[i].bm_bighint = radix; 795 scan[i].u.bmu_avail = radix; 796 } 797 } 798 } else { 799 radix /= BLIST_META_RADIX; 800 } 801 802 i = (allocBlk - blk) / radix; 803 blk += i * radix; 804 i = i * next_skip + 1; 805 806 while (i <= skip && blk < allocBlk + count) { 807 daddr_t v; 808 809 v = blk + radix - allocBlk; 810 if (v > count) 811 v = count; 812 813 if (scan->bm_bighint == (daddr_t)-1) 814 panic("blst_meta_fill: filling unexpected range"); 815 816 if (next_skip == 1) { 817 nblks += blst_leaf_fill(&scan[i], allocBlk, v); 818 } else { 819 nblks += blst_meta_fill(&scan[i], allocBlk, v, 820 radix, next_skip - 1, blk); 821 } 822 count -= v; 823 allocBlk += v; 824 blk += radix; 825 i += next_skip; 826 } 827 scan->u.bmu_avail -= nblks; 828 return nblks; 829} 830 831/* 832 * BLST_RADIX_INIT() - initialize radix tree 833 * 834 * Initialize our meta structures and bitmaps and calculate the exact 835 * amount of space required to manage 'count' blocks - this space may 836 * be considerably less than the calculated radix due to the large 837 * RADIX values we use. 838 */ 839 840static daddr_t 841blst_radix_init(blmeta_t *scan, daddr_t radix, int skip, daddr_t count) 842{ 843 int i; 844 int next_skip; 845 daddr_t memindex = 0; 846 847 /* 848 * Leaf node 849 */ 850 851 if (radix == BLIST_BMAP_RADIX) { 852 if (scan) { 853 scan->bm_bighint = 0; 854 scan->u.bmu_bitmap = 0; 855 } 856 return(memindex); 857 } 858 859 /* 860 * Meta node. If allocating the entire object we can special 861 * case it. However, we need to figure out how much memory 862 * is required to manage 'count' blocks, so we continue on anyway. 863 */ 864 865 if (scan) { 866 scan->bm_bighint = 0; 867 scan->u.bmu_avail = 0; 868 } 869 870 radix /= BLIST_META_RADIX; 871 next_skip = ((u_int)skip / BLIST_META_RADIX); 872 873 for (i = 1; i <= skip; i += next_skip) { 874 if (count >= radix) { 875 /* 876 * Allocate the entire object 877 */ 878 memindex = i + blst_radix_init( 879 ((scan) ? &scan[i] : NULL), 880 radix, 881 next_skip - 1, 882 radix 883 ); 884 count -= radix; 885 } else if (count > 0) { 886 /* 887 * Allocate a partial object 888 */ 889 memindex = i + blst_radix_init( 890 ((scan) ? &scan[i] : NULL), 891 radix, 892 next_skip - 1, 893 count 894 ); 895 count = 0; 896 } else { 897 /* 898 * Add terminator and break out 899 */ 900 if (scan) 901 scan[i].bm_bighint = (daddr_t)-1; 902 break; 903 } 904 } 905 if (memindex < i) 906 memindex = i; 907 return(memindex); 908} 909 910#ifdef BLIST_DEBUG 911 912static void 913blst_radix_print(blmeta_t *scan, daddr_t blk, daddr_t radix, int skip, int tab) 914{ 915 int i; 916 int next_skip; 917 int lastState = 0; 918 919 if (radix == BLIST_BMAP_RADIX) { 920 printf( 921 "%*.*s(%08llx,%lld): bitmap %016llx big=%lld\n", 922 tab, tab, "", 923 (long long)blk, (long long)radix, 924 (long long)scan->u.bmu_bitmap, 925 (long long)scan->bm_bighint 926 ); 927 return; 928 } 929 930 if (scan->u.bmu_avail == 0) { 931 printf( 932 "%*.*s(%08llx,%lld) ALL ALLOCATED\n", 933 tab, tab, "", 934 (long long)blk, 935 (long long)radix 936 ); 937 return; 938 } 939 if (scan->u.bmu_avail == radix) { 940 printf( 941 "%*.*s(%08llx,%lld) ALL FREE\n", 942 tab, tab, "", 943 (long long)blk, 944 (long long)radix 945 ); 946 return; 947 } 948 949 printf( 950 "%*.*s(%08llx,%lld): subtree (%lld/%lld) big=%lld {\n", 951 tab, tab, "", 952 (long long)blk, (long long)radix, 953 (long long)scan->u.bmu_avail, 954 (long long)radix, 955 (long long)scan->bm_bighint 956 ); 957 958 radix /= BLIST_META_RADIX; 959 next_skip = ((u_int)skip / BLIST_META_RADIX); 960 tab += 4; 961 962 for (i = 1; i <= skip; i += next_skip) { 963 if (scan[i].bm_bighint == (daddr_t)-1) { 964 printf( 965 "%*.*s(%08llx,%lld): Terminator\n", 966 tab, tab, "", 967 (long long)blk, (long long)radix 968 ); 969 lastState = 0; 970 break; 971 } 972 blst_radix_print( 973 &scan[i], 974 blk, 975 radix, 976 next_skip - 1, 977 tab 978 ); 979 blk += radix; 980 } 981 tab -= 4; 982 983 printf( 984 "%*.*s}\n", 985 tab, tab, "" 986 ); 987} 988 989#endif 990 991#ifdef BLIST_DEBUG 992 993int 994main(int ac, char **av) 995{ 996 int size = 1024; 997 int i; 998 blist_t bl; 999 1000 for (i = 1; i < ac; ++i) { 1001 const char *ptr = av[i]; 1002 if (*ptr != '-') { 1003 size = strtol(ptr, NULL, 0); 1004 continue; 1005 } 1006 ptr += 2; 1007 fprintf(stderr, "Bad option: %s\n", ptr - 2); 1008 exit(1); 1009 } 1010 bl = blist_create(size, M_WAITOK); 1011 blist_free(bl, 0, size); 1012 1013 for (;;) { 1014 char buf[1024]; 1015 long long da = 0; 1016 long long count = 0; 1017 1018 printf("%lld/%lld/%lld> ", (long long)bl->bl_free, 1019 (long long)size, (long long)bl->bl_radix); 1020 fflush(stdout); 1021 if (fgets(buf, sizeof(buf), stdin) == NULL) 1022 break; 1023 switch(buf[0]) { 1024 case 'r': 1025 if (sscanf(buf + 1, "%lld", &count) == 1) { 1026 blist_resize(&bl, count, 1, M_WAITOK); 1027 } else { 1028 printf("?\n"); 1029 } 1030 case 'p': 1031 blist_print(bl); 1032 break; 1033 case 'a': 1034 if (sscanf(buf + 1, "%lld", &count) == 1) { 1035 daddr_t blk = blist_alloc(bl, count); 1036 printf(" R=%08llx\n", (long long)blk); 1037 } else { 1038 printf("?\n"); 1039 } 1040 break; 1041 case 'f': 1042 if (sscanf(buf + 1, "%llx %lld", &da, &count) == 2) { 1043 blist_free(bl, da, count); 1044 } else { 1045 printf("?\n"); 1046 } 1047 break; 1048 case 'l': 1049 if (sscanf(buf + 1, "%llx %lld", &da, &count) == 2) { 1050 printf(" n=%d\n", 1051 blist_fill(bl, da, count)); 1052 } else { 1053 printf("?\n"); 1054 } 1055 break; 1056 case '?': 1057 case 'h': 1058 puts( 1059 "p -print\n" 1060 "a %d -allocate\n" 1061 "f %x %d -free\n" 1062 "l %x %d -fill\n" 1063 "r %d -resize\n" 1064 "h/? -help" 1065 ); 1066 break; 1067 default: 1068 printf("?\n"); 1069 break; 1070 } 1071 } 1072 return(0); 1073} 1074 1075void 1076panic(const char *ctl, ...) 1077{ 1078 va_list va; 1079 1080 va_start(va, ctl); 1081 vfprintf(stderr, ctl, va); 1082 fprintf(stderr, "\n"); 1083 va_end(va); 1084 exit(1); 1085} 1086 1087#endif 1088 1089