| 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 GRAPH_RDLOCK |
| 111 | qed_check_l1_table(QEDCheck *check, QEDTable *table) |
| 112 | { |
| 113 | BDRVQEDState *s = check->s; |
| 114 | unsigned int i, num_invalid_l1 = 0; |
| 115 | int ret, last_error = 0; |
| 116 | |
| 117 | /* Mark L1 table clusters used */ |
| 118 | qed_set_used_clusters(check, s->header.l1_table_offset, |
| 119 | s->header.table_size); |
| 120 | |
| 121 | for (i = 0; i < s->table_nelems; i++) { |
| 122 | unsigned int num_invalid_l2; |
| 123 | uint64_t offset = table->offsets[i]; |
| 124 | |
| 125 | if (qed_offset_is_unalloc_cluster(offset)) { |
| 126 | continue; |
| 127 | } |
| 128 | |
| 129 | /* Detect invalid L2 offset */ |
| 130 | if (!qed_check_table_offset(s, offset)) { |
| 131 | /* Clear invalid offset */ |
| 132 | if (check->fix) { |
| 133 | table->offsets[i] = 0; |
| 134 | check->result->corruptions_fixed++; |
| 135 | } else { |
| 136 | check->result->corruptions++; |
| 137 | } |
| 138 | |
| 139 | num_invalid_l1++; |
| 140 | continue; |
| 141 | } |
| 142 | |
| 143 | if (!qed_set_used_clusters(check, offset, s->header.table_size)) { |
| 144 | continue; /* skip an invalid table */ |
| 145 | } |
| 146 | |
| 147 | ret = qed_read_l2_table_sync(s, &check->request, offset); |
| 148 | if (ret) { |
| 149 | check->result->check_errors++; |
| 150 | last_error = ret; |
| 151 | continue; |
| 152 | } |
| 153 | |
| 154 | num_invalid_l2 = qed_check_l2_table(check, |
| 155 | check->request.l2_table->table); |
| 156 | |
| 157 | /* Write out fixed L2 table */ |
| 158 | if (num_invalid_l2 > 0 && check->fix) { |
| 159 | ret = qed_write_l2_table_sync(s, &check->request, 0, |
| 160 | s->table_nelems, false); |
| 161 | if (ret) { |
| 162 | check->result->check_errors++; |
| 163 | last_error = ret; |
| 164 | continue; |
| 165 | } |
| 166 | } |
| 167 | } |
| 168 | |
| 169 | /* Drop reference to final table */ |
| 170 | qed_unref_l2_cache_entry(check->request.l2_table); |
| 171 | check->request.l2_table = NULL; |
| 172 | |
| 173 | /* Write out fixed L1 table */ |
| 174 | if (num_invalid_l1 > 0 && check->fix) { |
| 175 | ret = qed_write_l1_table_sync(s, 0, s->table_nelems); |
| 176 | if (ret) { |
| 177 | check->result->check_errors++; |
| 178 | last_error = ret; |
| 179 | } |
| 180 | } |
| 181 | |
| 182 | return last_error; |
| 183 | } |
| 184 | |
| 185 | /** |
| 186 | * Check for unreferenced (leaked) clusters |
| 187 | */ |
| 188 | static void qed_check_for_leaks(QEDCheck *check) |
| 189 | { |
| 190 | BDRVQEDState *s = check->s; |
| 191 | uint64_t i; |
| 192 | |
| 193 | for (i = s->header.header_size; i < check->nclusters; i++) { |
| 194 | if (!qed_test_bit(check->used_clusters, i)) { |
| 195 | check->result->leaks++; |
| 196 | } |
| 197 | } |
| 198 | } |
| 199 | |
| 200 | /** |
| 201 | * Mark an image clean once it passes check or has been repaired |
| 202 | */ |
| 203 | static void coroutine_fn GRAPH_RDLOCK |
| 204 | qed_check_mark_clean(BDRVQEDState *s, BdrvCheckResult *result) |
| 205 | { |
| 206 | /* Skip if there were unfixable corruptions or I/O errors */ |
| 207 | if (result->corruptions > 0 || result->check_errors > 0) { |
| 208 | return; |
| 209 | } |
| 210 | |
| 211 | /* Skip if image is already marked clean */ |
| 212 | if (!(s->header.features & QED_F_NEED_CHECK)) { |
| 213 | return; |
| 214 | } |
| 215 | |
| 216 | /* Ensure fixes reach storage before clearing check bit */ |
| 217 | bdrv_co_flush(s->bs); |
| 218 | |
| 219 | s->header.features &= ~QED_F_NEED_CHECK; |
| 220 | qed_write_header_sync(s); |
| 221 | } |
| 222 | |
| 223 | /* Called with table_lock held. */ |
| 224 | int coroutine_fn qed_check(BDRVQEDState *s, BdrvCheckResult *result, bool fix) |
| 225 | { |
| 226 | QEDCheck check = { |
| 227 | .s = s, |
| 228 | .result = result, |
| 229 | .nclusters = qed_bytes_to_clusters(s, s->file_size), |
| 230 | .request = { .l2_table = NULL }, |
| 231 | .fix = fix, |
| 232 | }; |
| 233 | int ret; |
| 234 | |
| 235 | check.used_clusters = g_try_new0(uint32_t, (check.nclusters + 31) / 32); |
| 236 | if (check.nclusters && check.used_clusters == NULL) { |
| 237 | return -ENOMEM; |
| 238 | } |
| 239 | |
| 240 | check.result->bfi.total_clusters = |
| 241 | DIV_ROUND_UP(s->header.image_size, s->header.cluster_size); |
| 242 | ret = qed_check_l1_table(&check, s->l1_table); |
| 243 | if (ret == 0) { |
| 244 | /* Only check for leaks if entire image was scanned successfully */ |
| 245 | qed_check_for_leaks(&check); |
| 246 | |
| 247 | if (fix) { |
| 248 | qed_check_mark_clean(s, result); |
| 249 | } |
| 250 | } |
| 251 | |
| 252 | g_free(check.used_clusters); |
| 253 | return ret; |
| 254 | } |