1 /* 2 * L2/refcount table cache for the QCOW2 format 3 * 4 * Copyright (c) 2010 Kevin Wolf <kwolf@redhat.com> 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to deal 8 * in the Software without restriction, including without limitation the rights 9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 * copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22 * THE SOFTWARE. 23 */ 24 25 #include "qemu/osdep.h" 26 #include "block/block-io.h" 27 #include "qemu/memalign.h" 28 #include "qcow2.h" 29 #include "trace.h" 30 31 typedef struct Qcow2CachedTable { 32 int64_t offset; 33 uint64_t lru_counter; 34 int ref; 35 bool dirty; 36 } Qcow2CachedTable; 37 38 struct Qcow2Cache { 39 Qcow2CachedTable *entries; 40 struct Qcow2Cache *depends; 41 int size; 42 int table_size; 43 bool depends_on_flush; 44 void *table_array; 45 uint64_t lru_counter; 46 uint64_t cache_clean_lru_counter; 47 }; 48 49 static inline void *qcow2_cache_get_table_addr(Qcow2Cache *c, int table) 50 { 51 return (uint8_t *) c->table_array + (size_t) table * c->table_size; 52 } 53 54 static inline int qcow2_cache_get_table_idx(Qcow2Cache *c, void *table) 55 { 56 ptrdiff_t table_offset = (uint8_t *) table - (uint8_t *) c->table_array; 57 int idx = table_offset / c->table_size; 58 assert(idx >= 0 && idx < c->size && table_offset % c->table_size == 0); 59 return idx; 60 } 61 62 static inline const char *qcow2_cache_get_name(BDRVQcow2State *s, Qcow2Cache *c) 63 { 64 if (c == s->refcount_block_cache) { 65 return "refcount block"; 66 } else if (c == s->l2_table_cache) { 67 return "L2 table"; 68 } else { 69 /* Do not abort, because this is not critical */ 70 return "unknown"; 71 } 72 } 73 74 static void qcow2_cache_table_release(Qcow2Cache *c, int i, int num_tables) 75 { 76 /* Using MADV_DONTNEED to discard memory is a Linux-specific feature */ 77 #ifdef CONFIG_LINUX 78 void *t = qcow2_cache_get_table_addr(c, i); 79 int align = qemu_real_host_page_size(); 80 size_t mem_size = (size_t) c->table_size * num_tables; 81 size_t offset = QEMU_ALIGN_UP((uintptr_t) t, align) - (uintptr_t) t; 82 size_t length = QEMU_ALIGN_DOWN(mem_size - offset, align); 83 if (mem_size > offset && length > 0) { 84 madvise((uint8_t *) t + offset, length, MADV_DONTNEED); 85 } 86 #endif 87 } 88 89 static inline bool can_clean_entry(Qcow2Cache *c, int i) 90 { 91 Qcow2CachedTable *t = &c->entries[i]; 92 return t->ref == 0 && !t->dirty && t->offset != 0 && 93 t->lru_counter <= c->cache_clean_lru_counter; 94 } 95 96 void qcow2_cache_clean_unused(Qcow2Cache *c) 97 { 98 int i = 0; 99 while (i < c->size) { 100 int to_clean = 0; 101 102 /* Skip the entries that we don't need to clean */ 103 while (i < c->size && !can_clean_entry(c, i)) { 104 i++; 105 } 106 107 /* And count how many we can clean in a row */ 108 while (i < c->size && can_clean_entry(c, i)) { 109 c->entries[i].offset = 0; 110 c->entries[i].lru_counter = 0; 111 i++; 112 to_clean++; 113 } 114 115 if (to_clean > 0) { 116 qcow2_cache_table_release(c, i - to_clean, to_clean); 117 } 118 } 119 120 c->cache_clean_lru_counter = c->lru_counter; 121 } 122 123 Qcow2Cache *qcow2_cache_create(BlockDriverState *bs, int num_tables, 124 unsigned table_size) 125 { 126 BDRVQcow2State *s = bs->opaque; 127 Qcow2Cache *c; 128 129 assert(num_tables > 0); 130 assert(is_power_of_2(table_size)); 131 assert(table_size >= (1 << MIN_CLUSTER_BITS)); 132 assert(table_size <= s->cluster_size); 133 134 c = g_new0(Qcow2Cache, 1); 135 c->size = num_tables; 136 c->table_size = table_size; 137 c->entries = g_try_new0(Qcow2CachedTable, num_tables); 138 c->table_array = qemu_try_blockalign(bs->file->bs, 139 (size_t) num_tables * c->table_size); 140 141 if (!c->entries || !c->table_array) { 142 qemu_vfree(c->table_array); 143 g_free(c->entries); 144 g_free(c); 145 c = NULL; 146 } 147 148 return c; 149 } 150 151 int qcow2_cache_destroy(Qcow2Cache *c) 152 { 153 int i; 154 155 for (i = 0; i < c->size; i++) { 156 assert(c->entries[i].ref == 0); 157 } 158 159 qemu_vfree(c->table_array); 160 g_free(c->entries); 161 g_free(c); 162 163 return 0; 164 } 165 166 static int qcow2_cache_flush_dependency(BlockDriverState *bs, Qcow2Cache *c) 167 { 168 int ret; 169 170 ret = qcow2_cache_flush(bs, c->depends); 171 if (ret < 0) { 172 return ret; 173 } 174 175 c->depends = NULL; 176 c->depends_on_flush = false; 177 178 return 0; 179 } 180 181 static int qcow2_cache_entry_flush(BlockDriverState *bs, Qcow2Cache *c, int i) 182 { 183 BDRVQcow2State *s = bs->opaque; 184 int ret = 0; 185 186 if (!c->entries[i].dirty || !c->entries[i].offset) { 187 return 0; 188 } 189 190 trace_qcow2_cache_entry_flush(qemu_coroutine_self(), 191 c == s->l2_table_cache, i); 192 193 if (c->depends) { 194 ret = qcow2_cache_flush_dependency(bs, c); 195 } else if (c->depends_on_flush) { 196 ret = bdrv_flush(bs->file->bs); 197 if (ret >= 0) { 198 c->depends_on_flush = false; 199 } 200 } 201 202 if (ret < 0) { 203 return ret; 204 } 205 206 if (c == s->refcount_block_cache) { 207 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_REFCOUNT_BLOCK, 208 c->entries[i].offset, c->table_size, false); 209 } else if (c == s->l2_table_cache) { 210 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L2, 211 c->entries[i].offset, c->table_size, false); 212 } else { 213 ret = qcow2_pre_write_overlap_check(bs, 0, 214 c->entries[i].offset, c->table_size, false); 215 } 216 217 if (ret < 0) { 218 return ret; 219 } 220 221 if (c == s->refcount_block_cache) { 222 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_UPDATE_PART); 223 } else if (c == s->l2_table_cache) { 224 BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE); 225 } 226 227 ret = bdrv_pwrite(bs->file, c->entries[i].offset, c->table_size, 228 qcow2_cache_get_table_addr(c, i), 0); 229 if (ret < 0) { 230 return ret; 231 } 232 233 c->entries[i].dirty = false; 234 235 return 0; 236 } 237 238 int qcow2_cache_write(BlockDriverState *bs, Qcow2Cache *c) 239 { 240 BDRVQcow2State *s = bs->opaque; 241 int result = 0; 242 int ret; 243 int i; 244 245 trace_qcow2_cache_flush(qemu_coroutine_self(), c == s->l2_table_cache); 246 247 for (i = 0; i < c->size; i++) { 248 ret = qcow2_cache_entry_flush(bs, c, i); 249 if (ret < 0 && result != -ENOSPC) { 250 result = ret; 251 } 252 } 253 254 return result; 255 } 256 257 int qcow2_cache_flush(BlockDriverState *bs, Qcow2Cache *c) 258 { 259 int result = qcow2_cache_write(bs, c); 260 261 if (result == 0) { 262 int ret = bdrv_flush(bs->file->bs); 263 if (ret < 0) { 264 result = ret; 265 } 266 } 267 268 return result; 269 } 270 271 int qcow2_cache_set_dependency(BlockDriverState *bs, Qcow2Cache *c, 272 Qcow2Cache *dependency) 273 { 274 int ret; 275 276 if (dependency->depends) { 277 ret = qcow2_cache_flush_dependency(bs, dependency); 278 if (ret < 0) { 279 return ret; 280 } 281 } 282 283 if (c->depends && (c->depends != dependency)) { 284 ret = qcow2_cache_flush_dependency(bs, c); 285 if (ret < 0) { 286 return ret; 287 } 288 } 289 290 c->depends = dependency; 291 return 0; 292 } 293 294 void qcow2_cache_depends_on_flush(Qcow2Cache *c) 295 { 296 c->depends_on_flush = true; 297 } 298 299 int qcow2_cache_empty(BlockDriverState *bs, Qcow2Cache *c) 300 { 301 int ret, i; 302 303 ret = qcow2_cache_flush(bs, c); 304 if (ret < 0) { 305 return ret; 306 } 307 308 for (i = 0; i < c->size; i++) { 309 assert(c->entries[i].ref == 0); 310 c->entries[i].offset = 0; 311 c->entries[i].lru_counter = 0; 312 } 313 314 qcow2_cache_table_release(c, 0, c->size); 315 316 c->lru_counter = 0; 317 318 return 0; 319 } 320 321 static int qcow2_cache_do_get(BlockDriverState *bs, Qcow2Cache *c, 322 uint64_t offset, void **table, bool read_from_disk) 323 { 324 BDRVQcow2State *s = bs->opaque; 325 int i; 326 int ret; 327 int lookup_index; 328 uint64_t min_lru_counter = UINT64_MAX; 329 int min_lru_index = -1; 330 331 assert(offset != 0); 332 333 trace_qcow2_cache_get(qemu_coroutine_self(), c == s->l2_table_cache, 334 offset, read_from_disk); 335 336 if (!QEMU_IS_ALIGNED(offset, c->table_size)) { 337 qcow2_signal_corruption(bs, true, -1, -1, "Cannot get entry from %s " 338 "cache: Offset %#" PRIx64 " is unaligned", 339 qcow2_cache_get_name(s, c), offset); 340 return -EIO; 341 } 342 343 /* Check if the table is already cached */ 344 i = lookup_index = (offset / c->table_size * 4) % c->size; 345 do { 346 const Qcow2CachedTable *t = &c->entries[i]; 347 if (t->offset == offset) { 348 goto found; 349 } 350 if (t->ref == 0 && t->lru_counter < min_lru_counter) { 351 min_lru_counter = t->lru_counter; 352 min_lru_index = i; 353 } 354 if (++i == c->size) { 355 i = 0; 356 } 357 } while (i != lookup_index); 358 359 if (min_lru_index == -1) { 360 /* This can't happen in current synchronous code, but leave the check 361 * here as a reminder for whoever starts using AIO with the cache */ 362 abort(); 363 } 364 365 /* Cache miss: write a table back and replace it */ 366 i = min_lru_index; 367 trace_qcow2_cache_get_replace_entry(qemu_coroutine_self(), 368 c == s->l2_table_cache, i); 369 370 ret = qcow2_cache_entry_flush(bs, c, i); 371 if (ret < 0) { 372 return ret; 373 } 374 375 trace_qcow2_cache_get_read(qemu_coroutine_self(), 376 c == s->l2_table_cache, i); 377 c->entries[i].offset = 0; 378 if (read_from_disk) { 379 if (c == s->l2_table_cache) { 380 BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD); 381 } 382 383 ret = bdrv_pread(bs->file, offset, c->table_size, 384 qcow2_cache_get_table_addr(c, i), 0); 385 if (ret < 0) { 386 return ret; 387 } 388 } 389 390 c->entries[i].offset = offset; 391 392 /* And return the right table */ 393 found: 394 c->entries[i].ref++; 395 *table = qcow2_cache_get_table_addr(c, i); 396 397 trace_qcow2_cache_get_done(qemu_coroutine_self(), 398 c == s->l2_table_cache, i); 399 400 return 0; 401 } 402 403 int qcow2_cache_get(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset, 404 void **table) 405 { 406 return qcow2_cache_do_get(bs, c, offset, table, true); 407 } 408 409 int qcow2_cache_get_empty(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset, 410 void **table) 411 { 412 return qcow2_cache_do_get(bs, c, offset, table, false); 413 } 414 415 void qcow2_cache_put(Qcow2Cache *c, void **table) 416 { 417 int i = qcow2_cache_get_table_idx(c, *table); 418 419 c->entries[i].ref--; 420 *table = NULL; 421 422 if (c->entries[i].ref == 0) { 423 c->entries[i].lru_counter = ++c->lru_counter; 424 } 425 426 assert(c->entries[i].ref >= 0); 427 } 428 429 void qcow2_cache_entry_mark_dirty(Qcow2Cache *c, void *table) 430 { 431 int i = qcow2_cache_get_table_idx(c, table); 432 assert(c->entries[i].offset != 0); 433 c->entries[i].dirty = true; 434 } 435 436 void *qcow2_cache_is_table_offset(Qcow2Cache *c, uint64_t offset) 437 { 438 int i; 439 440 for (i = 0; i < c->size; i++) { 441 if (c->entries[i].offset == offset) { 442 return qcow2_cache_get_table_addr(c, i); 443 } 444 } 445 return NULL; 446 } 447 448 void qcow2_cache_discard(Qcow2Cache *c, void *table) 449 { 450 int i = qcow2_cache_get_table_idx(c, table); 451 452 assert(c->entries[i].ref == 0); 453 454 c->entries[i].offset = 0; 455 c->entries[i].lru_counter = 0; 456 c->entries[i].dirty = false; 457 458 qcow2_cache_table_release(c, i, 1); 459 } 460