block/qed-check.c

   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 "qemu/osdep.h"
  15 #include "block/block-io.h"
  16 #include "qed.h"
  17
  18 typedef struct {
  19     BDRVQEDState *s;
  20     BdrvCheckResult *result;
  21     bool fix;                           /* whether to fix invalid offsets */
  22
  23     uint64_t nclusters;
  24     uint32_t *used_clusters;            /* referenced cluster bitmap */
  25
  26     QEDRequest request;
  27 } QEDCheck;
  28
  29 static bool qed_test_bit(uint32_t *bitmap, uint64_t n) {
  30     return !!(bitmap[n / 32] & (1 << (n % 32)));
  31 }
  32
  33 static void qed_set_bit(uint32_t *bitmap, uint64_t n) {
  34     bitmap[n / 32] |= 1 << (n % 32);
  35 }
  36
  37 /**
  38  * Set bitmap bits for clusters
  39  *
  40  * @check:          Check structure
  41  * @offset:         Starting offset in bytes
  42  * @n:              Number of clusters
  43  */
  44 static bool qed_set_used_clusters(QEDCheck *check, uint64_t offset,
  45                                   unsigned int n)
  46 {
  47     uint64_t cluster = qed_bytes_to_clusters(check->s, offset);
  48     unsigned int corruptions = 0;
  49
  50     while (n-- != 0) {
  51         /* Clusters should only be referenced once */
  52         if (qed_test_bit(check->used_clusters, cluster)) {
  53             corruptions++;
  54         }
  55
  56         qed_set_bit(check->used_clusters, cluster);
  57         cluster++;
  58     }
  59
  60     check->result->corruptions += corruptions;
  61     return corruptions == 0;
  62 }
  63
  64 /**
  65  * Check an L2 table
  66  *
  67  * @ret:            Number of invalid cluster offsets
  68  */
  69 static unsigned int qed_check_l2_table(QEDCheck *check, QEDTable *table)
  70 {
  71     BDRVQEDState *s = check->s;
  72     unsigned int i, num_invalid = 0;
  73     uint64_t last_offset = 0;
  74
  75     for (i = 0; i < s->table_nelems; i++) {
  76         uint64_t offset = table->offsets[i];
  77
  78         if (qed_offset_is_unalloc_cluster(offset) ||
  79             qed_offset_is_zero_cluster(offset)) {
  80             continue;
  81         }
  82         check->result->bfi.allocated_clusters++;
  83         if (last_offset && (last_offset + s->header.cluster_size != offset)) {
  84             check->result->bfi.fragmented_clusters++;
  85         }
  86         last_offset = offset;
  87
  88         /* Detect invalid cluster offset */
  89         if (!qed_check_cluster_offset(s, offset)) {
  90             if (check->fix) {
  91                 table->offsets[i] = 0;
  92                 check->result->corruptions_fixed++;
  93             } else {
  94                 check->result->corruptions++;
  95             }
  96
  97             num_invalid++;
  98             continue;
  99         }
 100
 101         qed_set_used_clusters(check, offset, 1);
 102     }
 103
 104     return num_invalid;
 105 }
 106
 107 /**
 108  * Descend tables and check each cluster is referenced once only
 109  */
 110 static int coroutine_fn qed_check_l1_table(QEDCheck *check, QEDTable *table)
 111 {
 112     BDRVQEDState *s = check->s;
 113     unsigned int i, num_invalid_l1 = 0;
 114     int ret, last_error = 0;
 115
 116     /* Mark L1 table clusters used */
 117     qed_set_used_clusters(check, s->header.l1_table_offset,
 118                           s->header.table_size);
 119
 120     for (i = 0; i < s->table_nelems; i++) {
 121         unsigned int num_invalid_l2;
 122         uint64_t offset = table->offsets[i];
 123
 124         if (qed_offset_is_unalloc_cluster(offset)) {
 125             continue;
 126         }
 127
 128         /* Detect invalid L2 offset */
 129         if (!qed_check_table_offset(s, offset)) {
 130             /* Clear invalid offset */
 131             if (check->fix) {
 132                 table->offsets[i] = 0;
 133                 check->result->corruptions_fixed++;
 134             } else {
 135                 check->result->corruptions++;
 136             }
 137
 138             num_invalid_l1++;
 139             continue;
 140         }
 141
 142         if (!qed_set_used_clusters(check, offset, s->header.table_size)) {
 143             continue; /* skip an invalid table */
 144         }
 145
 146         ret = qed_read_l2_table_sync(s, &check->request, offset);
 147         if (ret) {
 148             check->result->check_errors++;
 149             last_error = ret;
 150             continue;
 151         }
 152
 153         num_invalid_l2 = qed_check_l2_table(check,
 154                                             check->request.l2_table->table);
 155
 156         /* Write out fixed L2 table */
 157         if (num_invalid_l2 > 0 && check->fix) {
 158             ret = qed_write_l2_table_sync(s, &check->request, 0,
 159                                           s->table_nelems, false);
 160             if (ret) {
 161                 check->result->check_errors++;
 162                 last_error = ret;
 163                 continue;
 164             }
 165         }
 166     }
 167
 168     /* Drop reference to final table */
 169     qed_unref_l2_cache_entry(check->request.l2_table);
 170     check->request.l2_table = NULL;
 171
 172     /* Write out fixed L1 table */
 173     if (num_invalid_l1 > 0 && check->fix) {
 174         ret = qed_write_l1_table_sync(s, 0, s->table_nelems);
 175         if (ret) {
 176             check->result->check_errors++;
 177             last_error = ret;
 178         }
 179     }
 180
 181     return last_error;
 182 }
 183
 184 /**
 185  * Check for unreferenced (leaked) clusters
 186  */
 187 static void qed_check_for_leaks(QEDCheck *check)
 188 {
 189     BDRVQEDState *s = check->s;
 190     uint64_t i;
 191
 192     for (i = s->header.header_size; i < check->nclusters; i++) {
 193         if (!qed_test_bit(check->used_clusters, i)) {
 194             check->result->leaks++;
 195         }
 196     }
 197 }
 198
 199 /**
 200  * Mark an image clean once it passes check or has been repaired
 201  */
 202 static void qed_check_mark_clean(BDRVQEDState *s, BdrvCheckResult *result)
 203 {
 204     /* Skip if there were unfixable corruptions or I/O errors */
 205     if (result->corruptions > 0 || result->check_errors > 0) {
 206         return;
 207     }
 208
 209     /* Skip if image is already marked clean */
 210     if (!(s->header.features & QED_F_NEED_CHECK)) {
 211         return;
 212     }
 213
 214     /* Ensure fixes reach storage before clearing check bit */
 215     bdrv_flush(s->bs);
 216
 217     s->header.features &= ~QED_F_NEED_CHECK;
 218     qed_write_header_sync(s);
 219 }
 220
 221 /* Called with table_lock held.  */
 222 int coroutine_fn qed_check(BDRVQEDState *s, BdrvCheckResult *result, bool fix)
 223 {
 224     QEDCheck check = {
 225         .s = s,
 226         .result = result,
 227         .nclusters = qed_bytes_to_clusters(s, s->file_size),
 228         .request = { .l2_table = NULL },
 229         .fix = fix,
 230     };
 231     int ret;
 232
 233     check.used_clusters = g_try_new0(uint32_t, (check.nclusters + 31) / 32);
 234     if (check.nclusters && check.used_clusters == NULL) {
 235         return -ENOMEM;
 236     }
 237
 238     check.result->bfi.total_clusters =
 239         DIV_ROUND_UP(s->header.image_size, s->header.cluster_size);
 240     ret = qed_check_l1_table(&check, s->l1_table);
 241     if (ret == 0) {
 242         /* Only check for leaks if entire image was scanned successfully */
 243         qed_check_for_leaks(&check);
 244
 245         if (fix) {
 246             qed_check_mark_clean(s, result);
 247         }
 248     }
 249
 250     g_free(check.used_clusters);
 251     return ret;
 252 }