qemu/block/qed-check.c
<<
>>
Prefs
   1/*
   2 * QEMU Enhanced Disk Format Consistency Check
   3 *
   4 * Copyright IBM, Corp. 2010
   5 *
   6 * Authors:
   7 *  Stefan Hajnoczi   <stefanha@linux.vnet.ibm.com>
   8 *
   9 * This work is licensed under the terms of the GNU LGPL, version 2 or later.
  10 * See the COPYING.LIB file in the top-level directory.
  11 *
  12 */
  13
  14#include "qed.h"
  15
  16typedef struct {
  17    BDRVQEDState *s;
  18    BdrvCheckResult *result;
  19    bool fix;                           /* whether to fix invalid offsets */
  20
  21    uint64_t nclusters;
  22    uint32_t *used_clusters;            /* referenced cluster bitmap */
  23
  24    QEDRequest request;
  25} QEDCheck;
  26
  27static bool qed_test_bit(uint32_t *bitmap, uint64_t n) {
  28    return !!(bitmap[n / 32] & (1 << (n % 32)));
  29}
  30
  31static void qed_set_bit(uint32_t *bitmap, uint64_t n) {
  32    bitmap[n / 32] |= 1 << (n % 32);
  33}
  34
  35/**
  36 * Set bitmap bits for clusters
  37 *
  38 * @check:          Check structure
  39 * @offset:         Starting offset in bytes
  40 * @n:              Number of clusters
  41 */
  42static bool qed_set_used_clusters(QEDCheck *check, uint64_t offset,
  43                                  unsigned int n)
  44{
  45    uint64_t cluster = qed_bytes_to_clusters(check->s, offset);
  46    unsigned int corruptions = 0;
  47
  48    while (n-- != 0) {
  49        /* Clusters should only be referenced once */
  50        if (qed_test_bit(check->used_clusters, cluster)) {
  51            corruptions++;
  52        }
  53
  54        qed_set_bit(check->used_clusters, cluster);
  55        cluster++;
  56    }
  57
  58    check->result->corruptions += corruptions;
  59    return corruptions == 0;
  60}
  61
  62/**
  63 * Check an L2 table
  64 *
  65 * @ret:            Number of invalid cluster offsets
  66 */
  67static unsigned int qed_check_l2_table(QEDCheck *check, QEDTable *table)
  68{
  69    BDRVQEDState *s = check->s;
  70    unsigned int i, num_invalid = 0;
  71    uint64_t last_offset = 0;
  72
  73    for (i = 0; i < s->table_nelems; i++) {
  74        uint64_t offset = table->offsets[i];
  75
  76        if (qed_offset_is_unalloc_cluster(offset) ||
  77            qed_offset_is_zero_cluster(offset)) {
  78            continue;
  79        }
  80        check->result->bfi.allocated_clusters++;
  81        if (last_offset && (last_offset + s->header.cluster_size != offset)) {
  82            check->result->bfi.fragmented_clusters++;
  83        }
  84        last_offset = offset;
  85
  86        /* Detect invalid cluster offset */
  87        if (!qed_check_cluster_offset(s, offset)) {
  88            if (check->fix) {
  89                table->offsets[i] = 0;
  90                check->result->corruptions_fixed++;
  91            } else {
  92                check->result->corruptions++;
  93            }
  94
  95            num_invalid++;
  96            continue;
  97        }
  98
  99        qed_set_used_clusters(check, offset, 1);
 100    }
 101
 102    return num_invalid;
 103}
 104
 105/**
 106 * Descend tables and check each cluster is referenced once only
 107 */
 108static int qed_check_l1_table(QEDCheck *check, QEDTable *table)
 109{
 110    BDRVQEDState *s = check->s;
 111    unsigned int i, num_invalid_l1 = 0;
 112    int ret, last_error = 0;
 113
 114    /* Mark L1 table clusters used */
 115    qed_set_used_clusters(check, s->header.l1_table_offset,
 116                          s->header.table_size);
 117
 118    for (i = 0; i < s->table_nelems; i++) {
 119        unsigned int num_invalid_l2;
 120        uint64_t offset = table->offsets[i];
 121
 122        if (qed_offset_is_unalloc_cluster(offset)) {
 123            continue;
 124        }
 125
 126        /* Detect invalid L2 offset */
 127        if (!qed_check_table_offset(s, offset)) {
 128            /* Clear invalid offset */
 129            if (check->fix) {
 130                table->offsets[i] = 0;
 131                check->result->corruptions_fixed++;
 132            } else {
 133                check->result->corruptions++;
 134            }
 135
 136            num_invalid_l1++;
 137            continue;
 138        }
 139
 140        if (!qed_set_used_clusters(check, offset, s->header.table_size)) {
 141            continue; /* skip an invalid table */
 142        }
 143
 144        ret = qed_read_l2_table_sync(s, &check->request, offset);
 145        if (ret) {
 146            check->result->check_errors++;
 147            last_error = ret;
 148            continue;
 149        }
 150
 151        num_invalid_l2 = qed_check_l2_table(check,
 152                                            check->request.l2_table->table);
 153
 154        /* Write out fixed L2 table */
 155        if (num_invalid_l2 > 0 && check->fix) {
 156            ret = qed_write_l2_table_sync(s, &check->request, 0,
 157                                          s->table_nelems, false);
 158            if (ret) {
 159                check->result->check_errors++;
 160                last_error = ret;
 161                continue;
 162            }
 163        }
 164    }
 165
 166    /* Drop reference to final table */
 167    qed_unref_l2_cache_entry(check->request.l2_table);
 168    check->request.l2_table = NULL;
 169
 170    /* Write out fixed L1 table */
 171    if (num_invalid_l1 > 0 && check->fix) {
 172        ret = qed_write_l1_table_sync(s, 0, s->table_nelems);
 173        if (ret) {
 174            check->result->check_errors++;
 175            last_error = ret;
 176        }
 177    }
 178
 179    return last_error;
 180}
 181
 182/**
 183 * Check for unreferenced (leaked) clusters
 184 */
 185static void qed_check_for_leaks(QEDCheck *check)
 186{
 187    BDRVQEDState *s = check->s;
 188    uint64_t i;
 189
 190    for (i = s->header.header_size; i < check->nclusters; i++) {
 191        if (!qed_test_bit(check->used_clusters, i)) {
 192            check->result->leaks++;
 193        }
 194    }
 195}
 196
 197/**
 198 * Mark an image clean once it passes check or has been repaired
 199 */
 200static void qed_check_mark_clean(BDRVQEDState *s, BdrvCheckResult *result)
 201{
 202    /* Skip if there were unfixable corruptions or I/O errors */
 203    if (result->corruptions > 0 || result->check_errors > 0) {
 204        return;
 205    }
 206
 207    /* Skip if image is already marked clean */
 208    if (!(s->header.features & QED_F_NEED_CHECK)) {
 209        return;
 210    }
 211
 212    /* Ensure fixes reach storage before clearing check bit */
 213    bdrv_flush(s->bs);
 214
 215    s->header.features &= ~QED_F_NEED_CHECK;
 216    qed_write_header_sync(s);
 217}
 218
 219int qed_check(BDRVQEDState *s, BdrvCheckResult *result, bool fix)
 220{
 221    QEDCheck check = {
 222        .s = s,
 223        .result = result,
 224        .nclusters = qed_bytes_to_clusters(s, s->file_size),
 225        .request = { .l2_table = NULL },
 226        .fix = fix,
 227    };
 228    int ret;
 229
 230    check.used_clusters = g_malloc0(((check.nclusters + 31) / 32) *
 231                                       sizeof(check.used_clusters[0]));
 232
 233    check.result->bfi.total_clusters =
 234        (s->header.image_size + s->header.cluster_size - 1) /
 235            s->header.cluster_size;
 236    ret = qed_check_l1_table(&check, s->l1_table);
 237    if (ret == 0) {
 238        /* Only check for leaks if entire image was scanned successfully */
 239        qed_check_for_leaks(&check);
 240
 241        if (fix) {
 242            qed_check_mark_clean(s, result);
 243        }
 244    }
 245
 246    g_free(check.used_clusters);
 247    return ret;
 248}
 249