qcow2-refcount.c 38.2 KB
Newer Older
K
Kevin Wolf 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
/*
 * Block driver for the QCOW version 2 format
 *
 * Copyright (c) 2004-2006 Fabrice Bellard
 *
 * Permission is hereby granted, free of charge, to any person obtaining a copy
 * of this software and associated documentation files (the "Software"), to deal
 * in the Software without restriction, including without limitation the rights
 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 * copies of the Software, and to permit persons to whom the Software is
 * furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
 * THE SOFTWARE.
 */

#include "qemu-common.h"
#include "block_int.h"
#include "block/qcow2.h"

static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size);
30
static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
K
Kevin Wolf 已提交
31 32 33
                            int64_t offset, int64_t length,
                            int addend);

34 35 36

static int cache_refcount_updates = 0;

37
static int write_refcount_block(BlockDriverState *bs)
38
{
39
    BDRVQcowState *s = bs->opaque;
40 41 42 43 44 45
    size_t size = s->cluster_size;

    if (s->refcount_block_cache_offset == 0) {
        return 0;
    }

46
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_UPDATE);
47 48
    if (bdrv_pwrite_sync(bs->file, s->refcount_block_cache_offset,
            s->refcount_block_cache, size) < 0)
49 50 51 52 53 54 55
    {
        return -EIO;
    }

    return 0;
}

K
Kevin Wolf 已提交
56 57 58
/*********************************************************/
/* refcount handling */

K
Kevin Wolf 已提交
59
int qcow2_refcount_init(BlockDriverState *bs)
K
Kevin Wolf 已提交
60 61 62 63 64 65 66 67
{
    BDRVQcowState *s = bs->opaque;
    int ret, refcount_table_size2, i;

    s->refcount_block_cache = qemu_malloc(s->cluster_size);
    refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
    s->refcount_table = qemu_malloc(refcount_table_size2);
    if (s->refcount_table_size > 0) {
68 69
        BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_LOAD);
        ret = bdrv_pread(bs->file, s->refcount_table_offset,
K
Kevin Wolf 已提交
70 71 72 73 74 75 76 77 78 79 80
                         s->refcount_table, refcount_table_size2);
        if (ret != refcount_table_size2)
            goto fail;
        for(i = 0; i < s->refcount_table_size; i++)
            be64_to_cpus(&s->refcount_table[i]);
    }
    return 0;
 fail:
    return -ENOMEM;
}

K
Kevin Wolf 已提交
81
void qcow2_refcount_close(BlockDriverState *bs)
K
Kevin Wolf 已提交
82 83 84 85 86 87 88 89 90 91 92 93
{
    BDRVQcowState *s = bs->opaque;
    qemu_free(s->refcount_block_cache);
    qemu_free(s->refcount_table);
}


static int load_refcount_block(BlockDriverState *bs,
                               int64_t refcount_block_offset)
{
    BDRVQcowState *s = bs->opaque;
    int ret;
94 95

    if (cache_refcount_updates) {
96 97 98 99
        ret = write_refcount_block(bs);
        if (ret < 0) {
            return ret;
        }
100 101
    }

102 103
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_LOAD);
    ret = bdrv_pread(bs->file, refcount_block_offset, s->refcount_block_cache,
K
Kevin Wolf 已提交
104
                     s->cluster_size);
105 106 107 108
    if (ret < 0) {
        return ret;
    }

K
Kevin Wolf 已提交
109 110 111 112
    s->refcount_block_cache_offset = refcount_block_offset;
    return 0;
}

113 114 115 116 117
/*
 * Returns the refcount of the cluster given by its index. Any non-negative
 * return value is the refcount of the cluster, negative values are -errno
 * and indicate an error.
 */
K
Kevin Wolf 已提交
118 119 120 121 122
static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
{
    BDRVQcowState *s = bs->opaque;
    int refcount_table_index, block_index;
    int64_t refcount_block_offset;
123
    int ret;
K
Kevin Wolf 已提交
124 125 126 127 128 129 130 131 132

    refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
    if (refcount_table_index >= s->refcount_table_size)
        return 0;
    refcount_block_offset = s->refcount_table[refcount_table_index];
    if (!refcount_block_offset)
        return 0;
    if (refcount_block_offset != s->refcount_block_cache_offset) {
        /* better than nothing: return allocated if read error */
133 134 135 136
        ret = load_refcount_block(bs, refcount_block_offset);
        if (ret < 0) {
            return ret;
        }
K
Kevin Wolf 已提交
137 138 139 140 141 142
    }
    block_index = cluster_index &
        ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
    return be16_to_cpu(s->refcount_block_cache[block_index]);
}

143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160
/*
 * Rounds the refcount table size up to avoid growing the table for each single
 * refcount block that is allocated.
 */
static unsigned int next_refcount_table_size(BDRVQcowState *s,
    unsigned int min_size)
{
    unsigned int min_clusters = (min_size >> (s->cluster_bits - 3)) + 1;
    unsigned int refcount_table_clusters =
        MAX(1, s->refcount_table_size >> (s->cluster_bits - 3));

    while (min_clusters > refcount_table_clusters) {
        refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
    }

    return refcount_table_clusters << (s->cluster_bits - 3);
}

161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178

/* Checks if two offsets are described by the same refcount block */
static int in_same_refcount_block(BDRVQcowState *s, uint64_t offset_a,
    uint64_t offset_b)
{
    uint64_t block_a = offset_a >> (2 * s->cluster_bits - REFCOUNT_SHIFT);
    uint64_t block_b = offset_b >> (2 * s->cluster_bits - REFCOUNT_SHIFT);

    return (block_a == block_b);
}

/*
 * Loads a refcount block. If it doesn't exist yet, it is allocated first
 * (including growing the refcount table if needed).
 *
 * Returns the offset of the refcount block on success or -errno in error case
 */
static int64_t alloc_refcount_block(BlockDriverState *bs, int64_t cluster_index)
K
Kevin Wolf 已提交
179 180
{
    BDRVQcowState *s = bs->opaque;
181 182 183
    unsigned int refcount_table_index;
    int ret;

184
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
K
Kevin Wolf 已提交
185

186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228
    /* Find the refcount block for the given cluster */
    refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);

    if (refcount_table_index < s->refcount_table_size) {

        uint64_t refcount_block_offset =
            s->refcount_table[refcount_table_index];

        /* If it's already there, we're done */
        if (refcount_block_offset) {
            if (refcount_block_offset != s->refcount_block_cache_offset) {
                ret = load_refcount_block(bs, refcount_block_offset);
                if (ret < 0) {
                    return ret;
                }
            }
            return refcount_block_offset;
        }
    }

    /*
     * If we came here, we need to allocate something. Something is at least
     * a cluster for the new refcount block. It may also include a new refcount
     * table if the old refcount table is too small.
     *
     * Note that allocating clusters here needs some special care:
     *
     * - We can't use the normal qcow2_alloc_clusters(), it would try to
     *   increase the refcount and very likely we would end up with an endless
     *   recursion. Instead we must place the refcount blocks in a way that
     *   they can describe them themselves.
     *
     * - We need to consider that at this point we are inside update_refcounts
     *   and doing the initial refcount increase. This means that some clusters
     *   have already been allocated by the caller, but their refcount isn't
     *   accurate yet. free_cluster_index tells us where this allocation ends
     *   as long as we don't overwrite it by freeing clusters.
     *
     * - alloc_clusters_noref and qcow2_free_clusters may load a different
     *   refcount block into the cache
     */

    if (cache_refcount_updates) {
229
        ret = write_refcount_block(bs);
230 231 232 233 234 235
        if (ret < 0) {
            return ret;
        }
    }

    /* Allocate the refcount block itself and mark it as used */
236 237 238 239
    int64_t new_block = alloc_clusters_noref(bs, s->cluster_size);
    if (new_block < 0) {
        return new_block;
    }
K
Kevin Wolf 已提交
240 241

#ifdef DEBUG_ALLOC2
242 243 244
    fprintf(stderr, "qcow2: Allocate refcount block %d for %" PRIx64
        " at %" PRIx64 "\n",
        refcount_table_index, cluster_index << s->cluster_bits, new_block);
K
Kevin Wolf 已提交
245
#endif
246 247

    if (in_same_refcount_block(s, new_block, cluster_index << s->cluster_bits)) {
248 249 250 251
        /* Zero the new refcount block before updating it */
        memset(s->refcount_block_cache, 0, s->cluster_size);
        s->refcount_block_cache_offset = new_block;

252 253 254 255 256 257 258 259 260 261 262
        /* The block describes itself, need to update the cache */
        int block_index = (new_block >> s->cluster_bits) &
            ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
        s->refcount_block_cache[block_index] = cpu_to_be16(1);
    } else {
        /* Described somewhere else. This can recurse at most twice before we
         * arrive at a block that describes itself. */
        ret = update_refcount(bs, new_block, s->cluster_size, 1);
        if (ret < 0) {
            goto fail_block;
        }
263 264 265 266 267

        /* Initialize the new refcount block only after updating its refcount,
         * update_refcount uses the refcount cache itself */
        memset(s->refcount_block_cache, 0, s->cluster_size);
        s->refcount_block_cache_offset = new_block;
268 269 270
    }

    /* Now the new refcount block needs to be written to disk */
271
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE);
272
    ret = bdrv_pwrite_sync(bs->file, new_block, s->refcount_block_cache,
273 274 275 276 277 278 279 280
        s->cluster_size);
    if (ret < 0) {
        goto fail_block;
    }

    /* If the refcount table is big enough, just hook the block up there */
    if (refcount_table_index < s->refcount_table_size) {
        uint64_t data64 = cpu_to_be64(new_block);
281
        BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_HOOKUP);
282
        ret = bdrv_pwrite_sync(bs->file,
283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303
            s->refcount_table_offset + refcount_table_index * sizeof(uint64_t),
            &data64, sizeof(data64));
        if (ret < 0) {
            goto fail_block;
        }

        s->refcount_table[refcount_table_index] = new_block;
        return new_block;
    }

    /*
     * If we come here, we need to grow the refcount table. Again, a new
     * refcount table needs some space and we can't simply allocate to avoid
     * endless recursion.
     *
     * Therefore let's grab new refcount blocks at the end of the image, which
     * will describe themselves and the new refcount table. This way we can
     * reference them only in the new table and do the switch to the new
     * refcount table at once without producing an inconsistent state in
     * between.
     */
304
    BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_GROW);
K
Kevin Wolf 已提交
305

306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343
    /* Calculate the number of refcount blocks needed so far */
    uint64_t refcount_block_clusters = 1 << (s->cluster_bits - REFCOUNT_SHIFT);
    uint64_t blocks_used = (s->free_cluster_index +
        refcount_block_clusters - 1) / refcount_block_clusters;

    /* And now we need at least one block more for the new metadata */
    uint64_t table_size = next_refcount_table_size(s, blocks_used + 1);
    uint64_t last_table_size;
    uint64_t blocks_clusters;
    do {
        uint64_t table_clusters = size_to_clusters(s, table_size);
        blocks_clusters = 1 +
            ((table_clusters + refcount_block_clusters - 1)
            / refcount_block_clusters);
        uint64_t meta_clusters = table_clusters + blocks_clusters;

        last_table_size = table_size;
        table_size = next_refcount_table_size(s, blocks_used +
            ((meta_clusters + refcount_block_clusters - 1)
            / refcount_block_clusters));

    } while (last_table_size != table_size);

#ifdef DEBUG_ALLOC2
    fprintf(stderr, "qcow2: Grow refcount table %" PRId32 " => %" PRId64 "\n",
        s->refcount_table_size, table_size);
#endif

    /* Create the new refcount table and blocks */
    uint64_t meta_offset = (blocks_used * refcount_block_clusters) *
        s->cluster_size;
    uint64_t table_offset = meta_offset + blocks_clusters * s->cluster_size;
    uint16_t *new_blocks = qemu_mallocz(blocks_clusters * s->cluster_size);
    uint64_t *new_table = qemu_mallocz(table_size * sizeof(uint64_t));

    assert(meta_offset >= (s->free_cluster_index * s->cluster_size));

    /* Fill the new refcount table */
K
Kevin Wolf 已提交
344
    memcpy(new_table, s->refcount_table,
345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360
        s->refcount_table_size * sizeof(uint64_t));
    new_table[refcount_table_index] = new_block;

    int i;
    for (i = 0; i < blocks_clusters; i++) {
        new_table[blocks_used + i] = meta_offset + (i * s->cluster_size);
    }

    /* Fill the refcount blocks */
    uint64_t table_clusters = size_to_clusters(s, table_size * sizeof(uint64_t));
    int block = 0;
    for (i = 0; i < table_clusters + blocks_clusters; i++) {
        new_blocks[block++] = cpu_to_be16(1);
    }

    /* Write refcount blocks to disk */
361
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_BLOCKS);
362
    ret = bdrv_pwrite_sync(bs->file, meta_offset, new_blocks,
363 364 365 366 367 368 369 370 371 372 373
        blocks_clusters * s->cluster_size);
    qemu_free(new_blocks);
    if (ret < 0) {
        goto fail_table;
    }

    /* Write refcount table to disk */
    for(i = 0; i < table_size; i++) {
        cpu_to_be64s(&new_table[i]);
    }

374
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_WRITE_TABLE);
375
    ret = bdrv_pwrite_sync(bs->file, table_offset, new_table,
376 377 378 379 380 381
        table_size * sizeof(uint64_t));
    if (ret < 0) {
        goto fail_table;
    }

    for(i = 0; i < table_size; i++) {
K
Kevin Wolf 已提交
382
        cpu_to_be64s(&new_table[i]);
383
    }
K
Kevin Wolf 已提交
384

385 386
    /* Hook up the new refcount table in the qcow2 header */
    uint8_t data[12];
K
Kevin Wolf 已提交
387
    cpu_to_be64w((uint64_t*)data, table_offset);
388
    cpu_to_be32w((uint32_t*)(data + 8), table_clusters);
389
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC_SWITCH_TABLE);
390
    ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, refcount_table_offset),
391 392 393
        data, sizeof(data));
    if (ret < 0) {
        goto fail_table;
394 395
    }

396 397 398 399
    /* And switch it in memory */
    uint64_t old_table_offset = s->refcount_table_offset;
    uint64_t old_table_size = s->refcount_table_size;

K
Kevin Wolf 已提交
400 401
    qemu_free(s->refcount_table);
    s->refcount_table = new_table;
402
    s->refcount_table_size = table_size;
K
Kevin Wolf 已提交
403 404
    s->refcount_table_offset = table_offset;

405 406
    /* Free old table. Remember, we must not change free_cluster_index */
    uint64_t old_free_cluster_index = s->free_cluster_index;
K
Kevin Wolf 已提交
407
    qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
408
    s->free_cluster_index = old_free_cluster_index;
K
Kevin Wolf 已提交
409

410 411 412
    ret = load_refcount_block(bs, new_block);
    if (ret < 0) {
        goto fail_block;
K
Kevin Wolf 已提交
413 414
    }

415
    return new_block;
K
Kevin Wolf 已提交
416

417 418 419 420 421
fail_table:
    qemu_free(new_table);
fail_block:
    s->refcount_block_cache_offset = 0;
    return ret;
K
Kevin Wolf 已提交
422 423
}

424
#define REFCOUNTS_PER_SECTOR (512 >> REFCOUNT_SHIFT)
425
static int write_refcount_block_entries(BlockDriverState *bs,
426 427
    int64_t refcount_block_offset, int first_index, int last_index)
{
428
    BDRVQcowState *s = bs->opaque;
429
    size_t size;
430
    int ret;
431

432 433 434 435
    if (cache_refcount_updates) {
        return 0;
    }

436 437 438 439
    if (first_index < 0) {
        return 0;
    }

440 441 442 443 444
    first_index &= ~(REFCOUNTS_PER_SECTOR - 1);
    last_index = (last_index + REFCOUNTS_PER_SECTOR)
        & ~(REFCOUNTS_PER_SECTOR - 1);

    size = (last_index - first_index) << REFCOUNT_SHIFT;
445

446
    BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_UPDATE_PART);
447
    ret = bdrv_pwrite_sync(bs->file,
448
        refcount_block_offset + (first_index << REFCOUNT_SHIFT),
449 450 451
        &s->refcount_block_cache[first_index], size);
    if (ret < 0) {
        return ret;
452 453 454 455 456
    }

    return 0;
}

K
Kevin Wolf 已提交
457
/* XXX: cache several refcount block clusters ? */
458 459
static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
    int64_t offset, int64_t length, int addend)
K
Kevin Wolf 已提交
460 461 462 463 464 465
{
    BDRVQcowState *s = bs->opaque;
    int64_t start, last, cluster_offset;
    int64_t refcount_block_offset = 0;
    int64_t table_index = -1, old_table_index;
    int first_index = -1, last_index = -1;
466
    int ret;
K
Kevin Wolf 已提交
467 468

#ifdef DEBUG_ALLOC2
469
    printf("update_refcount: offset=%" PRId64 " size=%" PRId64 " addend=%d\n",
K
Kevin Wolf 已提交
470 471
           offset, length, addend);
#endif
K
Kevin Wolf 已提交
472
    if (length < 0) {
K
Kevin Wolf 已提交
473
        return -EINVAL;
K
Kevin Wolf 已提交
474 475 476 477
    } else if (length == 0) {
        return 0;
    }

K
Kevin Wolf 已提交
478 479 480 481 482 483 484
    start = offset & ~(s->cluster_size - 1);
    last = (offset + length - 1) & ~(s->cluster_size - 1);
    for(cluster_offset = start; cluster_offset <= last;
        cluster_offset += s->cluster_size)
    {
        int block_index, refcount;
        int64_t cluster_index = cluster_offset >> s->cluster_bits;
485
        int64_t new_block;
K
Kevin Wolf 已提交
486 487 488 489 490

        /* Only write refcount block to disk when we are done with it */
        old_table_index = table_index;
        table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
        if ((old_table_index >= 0) && (table_index != old_table_index)) {
491

492 493 494 495
            ret = write_refcount_block_entries(bs, refcount_block_offset,
                first_index, last_index);
            if (ret < 0) {
                return ret;
K
Kevin Wolf 已提交
496 497 498 499 500 501 502
            }

            first_index = -1;
            last_index = -1;
        }

        /* Load the refcount block and allocate it if needed */
503 504 505 506
        new_block = alloc_refcount_block(bs, cluster_index);
        if (new_block < 0) {
            ret = new_block;
            goto fail;
K
Kevin Wolf 已提交
507
        }
508
        refcount_block_offset = new_block;
K
Kevin Wolf 已提交
509 510 511 512 513 514 515 516 517 518 519 520 521

        /* we can update the count and save it */
        block_index = cluster_index &
            ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
        if (first_index == -1 || block_index < first_index) {
            first_index = block_index;
        }
        if (block_index > last_index) {
            last_index = block_index;
        }

        refcount = be16_to_cpu(s->refcount_block_cache[block_index]);
        refcount += addend;
522 523 524 525
        if (refcount < 0 || refcount > 0xffff) {
            ret = -EINVAL;
            goto fail;
        }
K
Kevin Wolf 已提交
526 527 528 529 530 531
        if (refcount == 0 && cluster_index < s->free_cluster_index) {
            s->free_cluster_index = cluster_index;
        }
        s->refcount_block_cache[block_index] = cpu_to_be16(refcount);
    }

532 533 534
    ret = 0;
fail:

K
Kevin Wolf 已提交
535 536
    /* Write last changed block to disk */
    if (refcount_block_offset != 0) {
537 538 539 540 541
        int wret;
        wret = write_refcount_block_entries(bs, refcount_block_offset,
            first_index, last_index);
        if (wret < 0) {
            return ret < 0 ? ret : wret;
K
Kevin Wolf 已提交
542 543 544
        }
    }

545 546 547 548 549 550 551 552 553 554
    /*
     * Try do undo any updates if an error is returned (This may succeed in
     * some cases like ENOSPC for allocating a new refcount block)
     */
    if (ret < 0) {
        int dummy;
        dummy = update_refcount(bs, offset, cluster_offset - offset, -addend);
    }

    return ret;
K
Kevin Wolf 已提交
555 556
}

557 558 559 560 561 562 563
/*
 * Increases or decreases the refcount of a given cluster by one.
 * addend must be 1 or -1.
 *
 * If the return value is non-negative, it is the new refcount of the cluster.
 * If it is negative, it is -errno and indicates an error.
 */
K
Kevin Wolf 已提交
564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589
static int update_cluster_refcount(BlockDriverState *bs,
                                   int64_t cluster_index,
                                   int addend)
{
    BDRVQcowState *s = bs->opaque;
    int ret;

    ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend);
    if (ret < 0) {
        return ret;
    }

    return get_refcount(bs, cluster_index);
}



/*********************************************************/
/* cluster allocation functions */



/* return < 0 if error */
static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
{
    BDRVQcowState *s = bs->opaque;
590
    int i, nb_clusters, refcount;
K
Kevin Wolf 已提交
591 592 593 594

    nb_clusters = size_to_clusters(s, size);
retry:
    for(i = 0; i < nb_clusters; i++) {
595
        int64_t next_cluster_index = s->free_cluster_index++;
596 597 598 599 600
        refcount = get_refcount(bs, next_cluster_index);

        if (refcount < 0) {
            return refcount;
        } else if (refcount != 0) {
K
Kevin Wolf 已提交
601
            goto retry;
602
        }
K
Kevin Wolf 已提交
603 604
    }
#ifdef DEBUG_ALLOC2
605
    printf("alloc_clusters: size=%" PRId64 " -> %" PRId64 "\n",
K
Kevin Wolf 已提交
606 607 608 609 610 611
            size,
            (s->free_cluster_index - nb_clusters) << s->cluster_bits);
#endif
    return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
}

K
Kevin Wolf 已提交
612
int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size)
K
Kevin Wolf 已提交
613 614
{
    int64_t offset;
615
    int ret;
K
Kevin Wolf 已提交
616

617
    BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
K
Kevin Wolf 已提交
618
    offset = alloc_clusters_noref(bs, size);
619 620 621 622
    if (offset < 0) {
        return offset;
    }

623 624 625 626
    ret = update_refcount(bs, offset, size, 1);
    if (ret < 0) {
        return ret;
    }
K
Kevin Wolf 已提交
627 628 629 630 631
    return offset;
}

/* only used to allocate compressed sectors. We try to allocate
   contiguous sectors. size must be <= cluster_size */
K
Kevin Wolf 已提交
632
int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
K
Kevin Wolf 已提交
633 634 635 636 637
{
    BDRVQcowState *s = bs->opaque;
    int64_t offset, cluster_offset;
    int free_in_cluster;

638
    BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_BYTES);
K
Kevin Wolf 已提交
639 640
    assert(size > 0 && size <= s->cluster_size);
    if (s->free_byte_offset == 0) {
K
Kevin Wolf 已提交
641
        s->free_byte_offset = qcow2_alloc_clusters(bs, s->cluster_size);
642 643 644
        if (s->free_byte_offset < 0) {
            return s->free_byte_offset;
        }
K
Kevin Wolf 已提交
645 646 647 648 649 650 651 652 653 654 655 656 657 658
    }
 redo:
    free_in_cluster = s->cluster_size -
        (s->free_byte_offset & (s->cluster_size - 1));
    if (size <= free_in_cluster) {
        /* enough space in current cluster */
        offset = s->free_byte_offset;
        s->free_byte_offset += size;
        free_in_cluster -= size;
        if (free_in_cluster == 0)
            s->free_byte_offset = 0;
        if ((offset & (s->cluster_size - 1)) != 0)
            update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
    } else {
K
Kevin Wolf 已提交
659
        offset = qcow2_alloc_clusters(bs, s->cluster_size);
660 661 662
        if (offset < 0) {
            return offset;
        }
K
Kevin Wolf 已提交
663 664 665 666 667 668 669 670 671 672 673 674 675 676
        cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
        if ((cluster_offset + s->cluster_size) == offset) {
            /* we are lucky: contiguous data */
            offset = s->free_byte_offset;
            update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
            s->free_byte_offset += size;
        } else {
            s->free_byte_offset = offset;
            goto redo;
        }
    }
    return offset;
}

K
Kevin Wolf 已提交
677
void qcow2_free_clusters(BlockDriverState *bs,
K
Kevin Wolf 已提交
678 679
                          int64_t offset, int64_t size)
{
680 681
    int ret;

682
    BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
683 684 685
    ret = update_refcount(bs, offset, size, -1);
    if (ret < 0) {
        fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
686
        /* TODO Remember the clusters to free them later and avoid leaking */
687
    }
K
Kevin Wolf 已提交
688 689
}

690 691 692 693 694 695 696
/*
 * free_any_clusters
 *
 * free clusters according to its type: compressed or not
 *
 */

K
Kevin Wolf 已提交
697
void qcow2_free_any_clusters(BlockDriverState *bs,
698 699 700 701 702 703 704 705 706 707
    uint64_t cluster_offset, int nb_clusters)
{
    BDRVQcowState *s = bs->opaque;

    /* free the cluster */

    if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
        int nb_csectors;
        nb_csectors = ((cluster_offset >> s->csize_shift) &
                       s->csize_mask) + 1;
K
Kevin Wolf 已提交
708 709 710
        qcow2_free_clusters(bs,
            (cluster_offset & s->cluster_offset_mask) & ~511,
            nb_csectors * 512);
711 712 713
        return;
    }

K
Kevin Wolf 已提交
714
    qcow2_free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits);
715 716 717 718

    return;
}

K
Kevin Wolf 已提交
719 720 721 722 723 724 725


/*********************************************************/
/* snapshots and image creation */



K
Kevin Wolf 已提交
726 727
void qcow2_create_refcount_update(QCowCreateState *s, int64_t offset,
    int64_t size)
K
Kevin Wolf 已提交
728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744
{
    int refcount;
    int64_t start, last, cluster_offset;
    uint16_t *p;

    start = offset & ~(s->cluster_size - 1);
    last = (offset + size - 1)  & ~(s->cluster_size - 1);
    for(cluster_offset = start; cluster_offset <= last;
        cluster_offset += s->cluster_size) {
        p = &s->refcount_block[cluster_offset >> s->cluster_bits];
        refcount = be16_to_cpu(*p);
        refcount++;
        *p = cpu_to_be16(refcount);
    }
}

/* update the refcounts of snapshots and the copied flag */
K
Kevin Wolf 已提交
745 746
int qcow2_update_snapshot_refcount(BlockDriverState *bs,
    int64_t l1_table_offset, int l1_size, int addend)
K
Kevin Wolf 已提交
747 748 749 750 751 752
{
    BDRVQcowState *s = bs->opaque;
    uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
    int64_t old_offset, old_l2_offset;
    int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;

K
Kevin Wolf 已提交
753
    qcow2_l2_cache_reset(bs);
754
    cache_refcount_updates = 1;
K
Kevin Wolf 已提交
755 756 757 758 759

    l2_table = NULL;
    l1_table = NULL;
    l1_size2 = l1_size * sizeof(uint64_t);
    if (l1_table_offset != s->l1_table_offset) {
760 761 762 763 764
        if (l1_size2 != 0) {
            l1_table = qemu_mallocz(align_offset(l1_size2, 512));
        } else {
            l1_table = NULL;
        }
K
Kevin Wolf 已提交
765
        l1_allocated = 1;
766
        if (bdrv_pread(bs->file, l1_table_offset,
K
Kevin Wolf 已提交
767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785
                       l1_table, l1_size2) != l1_size2)
            goto fail;
        for(i = 0;i < l1_size; i++)
            be64_to_cpus(&l1_table[i]);
    } else {
        assert(l1_size == s->l1_size);
        l1_table = s->l1_table;
        l1_allocated = 0;
    }

    l2_size = s->l2_size * sizeof(uint64_t);
    l2_table = qemu_malloc(l2_size);
    l1_modified = 0;
    for(i = 0; i < l1_size; i++) {
        l2_offset = l1_table[i];
        if (l2_offset) {
            old_l2_offset = l2_offset;
            l2_offset &= ~QCOW_OFLAG_COPIED;
            l2_modified = 0;
786
            if (bdrv_pread(bs->file, l2_offset, l2_table, l2_size) != l2_size)
K
Kevin Wolf 已提交
787 788 789 790 791 792 793 794 795
                goto fail;
            for(j = 0; j < s->l2_size; j++) {
                offset = be64_to_cpu(l2_table[j]);
                if (offset != 0) {
                    old_offset = offset;
                    offset &= ~QCOW_OFLAG_COPIED;
                    if (offset & QCOW_OFLAG_COMPRESSED) {
                        nb_csectors = ((offset >> s->csize_shift) &
                                       s->csize_mask) + 1;
796 797 798 799 800 801 802 803 804
                        if (addend != 0) {
                            int ret;
                            ret = update_refcount(bs,
                                (offset & s->cluster_offset_mask) & ~511,
                                nb_csectors * 512, addend);
                            if (ret < 0) {
                                goto fail;
                            }
                        }
K
Kevin Wolf 已提交
805 806 807 808 809 810 811 812
                        /* compressed clusters are never modified */
                        refcount = 2;
                    } else {
                        if (addend != 0) {
                            refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
                        } else {
                            refcount = get_refcount(bs, offset >> s->cluster_bits);
                        }
813 814 815 816

                        if (refcount < 0) {
                            goto fail;
                        }
K
Kevin Wolf 已提交
817 818 819 820 821 822 823 824 825 826 827 828
                    }

                    if (refcount == 1) {
                        offset |= QCOW_OFLAG_COPIED;
                    }
                    if (offset != old_offset) {
                        l2_table[j] = cpu_to_be64(offset);
                        l2_modified = 1;
                    }
                }
            }
            if (l2_modified) {
829 830
                if (bdrv_pwrite_sync(bs->file,
                                l2_offset, l2_table, l2_size) < 0)
K
Kevin Wolf 已提交
831 832 833 834 835 836 837 838
                    goto fail;
            }

            if (addend != 0) {
                refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
            } else {
                refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
            }
839 840 841
            if (refcount < 0) {
                goto fail;
            } else if (refcount == 1) {
K
Kevin Wolf 已提交
842 843 844 845 846 847 848 849 850 851 852
                l2_offset |= QCOW_OFLAG_COPIED;
            }
            if (l2_offset != old_l2_offset) {
                l1_table[i] = l2_offset;
                l1_modified = 1;
            }
        }
    }
    if (l1_modified) {
        for(i = 0; i < l1_size; i++)
            cpu_to_be64s(&l1_table[i]);
853 854
        if (bdrv_pwrite_sync(bs->file, l1_table_offset, l1_table,
                        l1_size2) < 0)
K
Kevin Wolf 已提交
855 856 857 858 859 860 861
            goto fail;
        for(i = 0; i < l1_size; i++)
            be64_to_cpus(&l1_table[i]);
    }
    if (l1_allocated)
        qemu_free(l1_table);
    qemu_free(l2_table);
862
    cache_refcount_updates = 0;
863
    write_refcount_block(bs);
K
Kevin Wolf 已提交
864 865 866 867 868
    return 0;
 fail:
    if (l1_allocated)
        qemu_free(l1_table);
    qemu_free(l2_table);
869
    cache_refcount_updates = 0;
870
    write_refcount_block(bs);
K
Kevin Wolf 已提交
871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943
    return -EIO;
}




/*********************************************************/
/* refcount checking functions */



/*
 * Increases the refcount for a range of clusters in a given refcount table.
 * This is used to construct a temporary refcount table out of L1 and L2 tables
 * which can be compared the the refcount table saved in the image.
 *
 * Returns the number of errors in the image that were found
 */
static int inc_refcounts(BlockDriverState *bs,
                          uint16_t *refcount_table,
                          int refcount_table_size,
                          int64_t offset, int64_t size)
{
    BDRVQcowState *s = bs->opaque;
    int64_t start, last, cluster_offset;
    int k;
    int errors = 0;

    if (size <= 0)
        return 0;

    start = offset & ~(s->cluster_size - 1);
    last = (offset + size - 1) & ~(s->cluster_size - 1);
    for(cluster_offset = start; cluster_offset <= last;
        cluster_offset += s->cluster_size) {
        k = cluster_offset >> s->cluster_bits;
        if (k < 0 || k >= refcount_table_size) {
            fprintf(stderr, "ERROR: invalid cluster offset=0x%" PRIx64 "\n",
                cluster_offset);
            errors++;
        } else {
            if (++refcount_table[k] == 0) {
                fprintf(stderr, "ERROR: overflow cluster offset=0x%" PRIx64
                    "\n", cluster_offset);
                errors++;
            }
        }
    }

    return errors;
}

/*
 * Increases the refcount in the given refcount table for the all clusters
 * referenced in the L2 table. While doing so, performs some checks on L2
 * entries.
 *
 * Returns the number of errors found by the checks or -errno if an internal
 * error occurred.
 */
static int check_refcounts_l2(BlockDriverState *bs,
    uint16_t *refcount_table, int refcount_table_size, int64_t l2_offset,
    int check_copied)
{
    BDRVQcowState *s = bs->opaque;
    uint64_t *l2_table, offset;
    int i, l2_size, nb_csectors, refcount;
    int errors = 0;

    /* Read L2 table from disk */
    l2_size = s->l2_size * sizeof(uint64_t);
    l2_table = qemu_malloc(l2_size);

944
    if (bdrv_pread(bs->file, l2_offset, l2_table, l2_size) != l2_size)
K
Kevin Wolf 已提交
945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973
        goto fail;

    /* Do the actual checks */
    for(i = 0; i < s->l2_size; i++) {
        offset = be64_to_cpu(l2_table[i]);
        if (offset != 0) {
            if (offset & QCOW_OFLAG_COMPRESSED) {
                /* Compressed clusters don't have QCOW_OFLAG_COPIED */
                if (offset & QCOW_OFLAG_COPIED) {
                    fprintf(stderr, "ERROR: cluster %" PRId64 ": "
                        "copied flag must never be set for compressed "
                        "clusters\n", offset >> s->cluster_bits);
                    offset &= ~QCOW_OFLAG_COPIED;
                    errors++;
                }

                /* Mark cluster as used */
                nb_csectors = ((offset >> s->csize_shift) &
                               s->csize_mask) + 1;
                offset &= s->cluster_offset_mask;
                errors += inc_refcounts(bs, refcount_table,
                              refcount_table_size,
                              offset & ~511, nb_csectors * 512);
            } else {
                /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
                if (check_copied) {
                    uint64_t entry = offset;
                    offset &= ~QCOW_OFLAG_COPIED;
                    refcount = get_refcount(bs, offset >> s->cluster_bits);
974 975 976 977
                    if (refcount < 0) {
                        fprintf(stderr, "Can't get refcount for offset %"
                            PRIx64 ": %s\n", entry, strerror(-refcount));
                    }
K
Kevin Wolf 已提交
978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035
                    if ((refcount == 1) != ((entry & QCOW_OFLAG_COPIED) != 0)) {
                        fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
                            PRIx64 " refcount=%d\n", entry, refcount);
                        errors++;
                    }
                }

                /* Mark cluster as used */
                offset &= ~QCOW_OFLAG_COPIED;
                errors += inc_refcounts(bs, refcount_table,
                              refcount_table_size,
                              offset, s->cluster_size);

                /* Correct offsets are cluster aligned */
                if (offset & (s->cluster_size - 1)) {
                    fprintf(stderr, "ERROR offset=%" PRIx64 ": Cluster is not "
                        "properly aligned; L2 entry corrupted.\n", offset);
                    errors++;
                }
            }
        }
    }

    qemu_free(l2_table);
    return errors;

fail:
    fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
    qemu_free(l2_table);
    return -EIO;
}

/*
 * Increases the refcount for the L1 table, its L2 tables and all referenced
 * clusters in the given refcount table. While doing so, performs some checks
 * on L1 and L2 entries.
 *
 * Returns the number of errors found by the checks or -errno if an internal
 * error occurred.
 */
static int check_refcounts_l1(BlockDriverState *bs,
                              uint16_t *refcount_table,
                              int refcount_table_size,
                              int64_t l1_table_offset, int l1_size,
                              int check_copied)
{
    BDRVQcowState *s = bs->opaque;
    uint64_t *l1_table, l2_offset, l1_size2;
    int i, refcount, ret;
    int errors = 0;

    l1_size2 = l1_size * sizeof(uint64_t);

    /* Mark L1 table as used */
    errors += inc_refcounts(bs, refcount_table, refcount_table_size,
                  l1_table_offset, l1_size2);

    /* Read L1 table entries from disk */
1036 1037 1038 1039
    if (l1_size2 == 0) {
        l1_table = NULL;
    } else {
        l1_table = qemu_malloc(l1_size2);
1040
        if (bdrv_pread(bs->file, l1_table_offset,
1041 1042 1043 1044 1045
                       l1_table, l1_size2) != l1_size2)
            goto fail;
        for(i = 0;i < l1_size; i++)
            be64_to_cpus(&l1_table[i]);
    }
K
Kevin Wolf 已提交
1046 1047 1048 1049 1050 1051 1052 1053 1054

    /* Do the actual checks */
    for(i = 0; i < l1_size; i++) {
        l2_offset = l1_table[i];
        if (l2_offset) {
            /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
            if (check_copied) {
                refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED)
                    >> s->cluster_bits);
1055 1056 1057 1058
                if (refcount < 0) {
                    fprintf(stderr, "Can't get refcount for l2_offset %"
                        PRIx64 ": %s\n", l2_offset, strerror(-refcount));
                }
K
Kevin Wolf 已提交
1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103
                if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
                    fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
                        " refcount=%d\n", l2_offset, refcount);
                    errors++;
                }
            }

            /* Mark L2 table as used */
            l2_offset &= ~QCOW_OFLAG_COPIED;
            errors += inc_refcounts(bs, refcount_table,
                          refcount_table_size,
                          l2_offset,
                          s->cluster_size);

            /* L2 tables are cluster aligned */
            if (l2_offset & (s->cluster_size - 1)) {
                fprintf(stderr, "ERROR l2_offset=%" PRIx64 ": Table is not "
                    "cluster aligned; L1 entry corrupted\n", l2_offset);
                errors++;
            }

            /* Process and check L2 entries */
            ret = check_refcounts_l2(bs, refcount_table, refcount_table_size,
                l2_offset, check_copied);
            if (ret < 0) {
                goto fail;
            }
            errors += ret;
        }
    }
    qemu_free(l1_table);
    return errors;

fail:
    fprintf(stderr, "ERROR: I/O error in check_refcounts_l1\n");
    qemu_free(l1_table);
    return -EIO;
}

/*
 * Checks an image for refcount consistency.
 *
 * Returns 0 if no errors are found, the number of errors in case the image is
 * detected as corrupted, and -errno when an internal error occured.
 */
K
Kevin Wolf 已提交
1104
int qcow2_check_refcounts(BlockDriverState *bs)
K
Kevin Wolf 已提交
1105 1106 1107 1108 1109 1110 1111 1112
{
    BDRVQcowState *s = bs->opaque;
    int64_t size;
    int nb_clusters, refcount1, refcount2, i;
    QCowSnapshot *sn;
    uint16_t *refcount_table;
    int ret, errors = 0;

1113
    size = bdrv_getlength(bs->file);
K
Kevin Wolf 已提交
1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144
    nb_clusters = size_to_clusters(s, size);
    refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));

    /* header */
    errors += inc_refcounts(bs, refcount_table, nb_clusters,
                  0, s->cluster_size);

    /* current L1 table */
    ret = check_refcounts_l1(bs, refcount_table, nb_clusters,
                       s->l1_table_offset, s->l1_size, 1);
    if (ret < 0) {
        return ret;
    }
    errors += ret;

    /* snapshots */
    for(i = 0; i < s->nb_snapshots; i++) {
        sn = s->snapshots + i;
        check_refcounts_l1(bs, refcount_table, nb_clusters,
                           sn->l1_table_offset, sn->l1_size, 0);
    }
    errors += inc_refcounts(bs, refcount_table, nb_clusters,
                  s->snapshots_offset, s->snapshots_size);

    /* refcount data */
    errors += inc_refcounts(bs, refcount_table, nb_clusters,
                  s->refcount_table_offset,
                  s->refcount_table_size * sizeof(uint64_t));
    for(i = 0; i < s->refcount_table_size; i++) {
        int64_t offset;
        offset = s->refcount_table[i];
1145 1146 1147 1148 1149 1150 1151 1152

        /* Refcount blocks are cluster aligned */
        if (offset & (s->cluster_size - 1)) {
            fprintf(stderr, "ERROR refcount block %d is not "
                "cluster aligned; refcount table entry corrupted\n", i);
            errors++;
        }

K
Kevin Wolf 已提交
1153 1154 1155
        if (offset != 0) {
            errors += inc_refcounts(bs, refcount_table, nb_clusters,
                          offset, s->cluster_size);
1156 1157 1158 1159
            if (refcount_table[offset / s->cluster_size] != 1) {
                fprintf(stderr, "ERROR refcount block %d refcount=%d\n",
                    i, refcount_table[offset / s->cluster_size]);
            }
K
Kevin Wolf 已提交
1160 1161 1162 1163 1164 1165
        }
    }

    /* compare ref counts */
    for(i = 0; i < nb_clusters; i++) {
        refcount1 = get_refcount(bs, i);
1166 1167 1168 1169 1170
        if (refcount1 < 0) {
            fprintf(stderr, "Can't get refcount for cluster %d: %s\n",
                i, strerror(-refcount1));
        }

K
Kevin Wolf 已提交
1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183
        refcount2 = refcount_table[i];
        if (refcount1 != refcount2) {
            fprintf(stderr, "ERROR cluster %d refcount=%d reference=%d\n",
                   i, refcount1, refcount2);
            errors++;
        }
    }

    qemu_free(refcount_table);

    return errors;
}