1 /* 2 * copy-before-write filter driver 3 * 4 * The driver performs Copy-Before-Write (CBW) operation: it is injected above 5 * some node, and before each write it copies _old_ data to the target node. 6 * 7 * Copyright (c) 2018-2021 Virtuozzo International GmbH. 8 * 9 * Author: 10 * Sementsov-Ogievskiy Vladimir <vsementsov@virtuozzo.com> 11 * 12 * This program is free software; you can redistribute it and/or modify 13 * it under the terms of the GNU General Public License as published by 14 * the Free Software Foundation; either version 2 of the License, or 15 * (at your option) any later version. 16 * 17 * This program is distributed in the hope that it will be useful, 18 * but WITHOUT ANY WARRANTY; without even the implied warranty of 19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 20 * GNU General Public License for more details. 21 * 22 * You should have received a copy of the GNU General Public License 23 * along with this program. If not, see <http://www.gnu.org/licenses/>. 24 */ 25 26 #include "qemu/osdep.h" 27 #include "qapi/qmp/qjson.h" 28 29 #include "sysemu/block-backend.h" 30 #include "qemu/cutils.h" 31 #include "qapi/error.h" 32 #include "block/block_int.h" 33 #include "block/qdict.h" 34 #include "block/block-copy.h" 35 #include "block/dirty-bitmap.h" 36 37 #include "block/copy-before-write.h" 38 #include "block/reqlist.h" 39 40 #include "qapi/qapi-visit-block-core.h" 41 42 typedef struct BDRVCopyBeforeWriteState { 43 BlockCopyState *bcs; 44 BdrvChild *target; 45 OnCbwError on_cbw_error; 46 uint32_t cbw_timeout_ns; 47 48 /* 49 * @lock: protects access to @access_bitmap, @done_bitmap and 50 * @frozen_read_reqs 51 */ 52 CoMutex lock; 53 54 /* 55 * @access_bitmap: represents areas allowed for reading by fleecing user. 56 * Reading from non-dirty areas leads to -EACCES. 57 */ 58 BdrvDirtyBitmap *access_bitmap; 59 60 /* 61 * @done_bitmap: represents areas that was successfully copied to @target by 62 * copy-before-write operations. 63 */ 64 BdrvDirtyBitmap *done_bitmap; 65 66 /* 67 * @frozen_read_reqs: current read requests for fleecing user in bs->file 68 * node. These areas must not be rewritten by guest. 69 */ 70 BlockReqList frozen_read_reqs; 71 72 /* 73 * @snapshot_error is normally zero. But on first copy-before-write failure 74 * when @on_cbw_error == ON_CBW_ERROR_BREAK_SNAPSHOT, @snapshot_error takes 75 * value of this error (<0). After that all in-flight and further 76 * snapshot-API requests will fail with that error. 77 */ 78 int snapshot_error; 79 } BDRVCopyBeforeWriteState; 80 81 static coroutine_fn int cbw_co_preadv( 82 BlockDriverState *bs, int64_t offset, int64_t bytes, 83 QEMUIOVector *qiov, BdrvRequestFlags flags) 84 { 85 return bdrv_co_preadv(bs->file, offset, bytes, qiov, flags); 86 } 87 88 static void block_copy_cb(void *opaque) 89 { 90 BlockDriverState *bs = opaque; 91 92 bdrv_dec_in_flight(bs); 93 } 94 95 /* 96 * Do copy-before-write operation. 97 * 98 * On failure guest request must be failed too. 99 * 100 * On success, we also wait for all in-flight fleecing read requests in source 101 * node, and it's guaranteed that after cbw_do_copy_before_write() successful 102 * return there are no such requests and they will never appear. 103 */ 104 static coroutine_fn int cbw_do_copy_before_write(BlockDriverState *bs, 105 uint64_t offset, uint64_t bytes, BdrvRequestFlags flags) 106 { 107 BDRVCopyBeforeWriteState *s = bs->opaque; 108 int ret; 109 uint64_t off, end; 110 int64_t cluster_size = block_copy_cluster_size(s->bcs); 111 112 if (flags & BDRV_REQ_WRITE_UNCHANGED) { 113 return 0; 114 } 115 116 if (s->snapshot_error) { 117 return 0; 118 } 119 120 off = QEMU_ALIGN_DOWN(offset, cluster_size); 121 end = QEMU_ALIGN_UP(offset + bytes, cluster_size); 122 123 /* 124 * Increase in_flight, so that in case of timed-out block-copy, the 125 * remaining background block_copy() request (which can't be immediately 126 * cancelled by timeout) is presented in bs->in_flight. This way we are 127 * sure that on bs close() we'll previously wait for all timed-out but yet 128 * running block_copy calls. 129 */ 130 bdrv_inc_in_flight(bs); 131 ret = block_copy(s->bcs, off, end - off, true, s->cbw_timeout_ns, 132 block_copy_cb, bs); 133 if (ret < 0 && s->on_cbw_error == ON_CBW_ERROR_BREAK_GUEST_WRITE) { 134 return ret; 135 } 136 137 WITH_QEMU_LOCK_GUARD(&s->lock) { 138 if (ret < 0) { 139 assert(s->on_cbw_error == ON_CBW_ERROR_BREAK_SNAPSHOT); 140 if (!s->snapshot_error) { 141 s->snapshot_error = ret; 142 } 143 } else { 144 bdrv_set_dirty_bitmap(s->done_bitmap, off, end - off); 145 } 146 reqlist_wait_all(&s->frozen_read_reqs, off, end - off, &s->lock); 147 } 148 149 return 0; 150 } 151 152 static int coroutine_fn cbw_co_pdiscard(BlockDriverState *bs, 153 int64_t offset, int64_t bytes) 154 { 155 int ret = cbw_do_copy_before_write(bs, offset, bytes, 0); 156 if (ret < 0) { 157 return ret; 158 } 159 160 return bdrv_co_pdiscard(bs->file, offset, bytes); 161 } 162 163 static int coroutine_fn cbw_co_pwrite_zeroes(BlockDriverState *bs, 164 int64_t offset, int64_t bytes, BdrvRequestFlags flags) 165 { 166 int ret = cbw_do_copy_before_write(bs, offset, bytes, flags); 167 if (ret < 0) { 168 return ret; 169 } 170 171 return bdrv_co_pwrite_zeroes(bs->file, offset, bytes, flags); 172 } 173 174 static coroutine_fn int cbw_co_pwritev(BlockDriverState *bs, 175 int64_t offset, 176 int64_t bytes, 177 QEMUIOVector *qiov, 178 BdrvRequestFlags flags) 179 { 180 int ret = cbw_do_copy_before_write(bs, offset, bytes, flags); 181 if (ret < 0) { 182 return ret; 183 } 184 185 return bdrv_co_pwritev(bs->file, offset, bytes, qiov, flags); 186 } 187 188 static int coroutine_fn cbw_co_flush(BlockDriverState *bs) 189 { 190 if (!bs->file) { 191 return 0; 192 } 193 194 return bdrv_co_flush(bs->file->bs); 195 } 196 197 /* 198 * If @offset not accessible - return NULL. 199 * 200 * Otherwise, set @pnum to some bytes that accessible from @file (@file is set 201 * to bs->file or to s->target). Return newly allocated BlockReq object that 202 * should be than passed to cbw_snapshot_read_unlock(). 203 * 204 * It's guaranteed that guest writes will not interact in the region until 205 * cbw_snapshot_read_unlock() called. 206 */ 207 static coroutine_fn BlockReq * 208 cbw_snapshot_read_lock(BlockDriverState *bs, int64_t offset, int64_t bytes, 209 int64_t *pnum, BdrvChild **file) 210 { 211 BDRVCopyBeforeWriteState *s = bs->opaque; 212 BlockReq *req = g_new(BlockReq, 1); 213 bool done; 214 215 QEMU_LOCK_GUARD(&s->lock); 216 217 if (s->snapshot_error) { 218 g_free(req); 219 return NULL; 220 } 221 222 if (bdrv_dirty_bitmap_next_zero(s->access_bitmap, offset, bytes) != -1) { 223 g_free(req); 224 return NULL; 225 } 226 227 done = bdrv_dirty_bitmap_status(s->done_bitmap, offset, bytes, pnum); 228 if (done) { 229 /* 230 * Special invalid BlockReq, that is handled in 231 * cbw_snapshot_read_unlock(). We don't need to lock something to read 232 * from s->target. 233 */ 234 *req = (BlockReq) {.offset = -1, .bytes = -1}; 235 *file = s->target; 236 } else { 237 reqlist_init_req(&s->frozen_read_reqs, req, offset, bytes); 238 *file = bs->file; 239 } 240 241 return req; 242 } 243 244 static coroutine_fn void 245 cbw_snapshot_read_unlock(BlockDriverState *bs, BlockReq *req) 246 { 247 BDRVCopyBeforeWriteState *s = bs->opaque; 248 249 if (req->offset == -1 && req->bytes == -1) { 250 g_free(req); 251 return; 252 } 253 254 QEMU_LOCK_GUARD(&s->lock); 255 256 reqlist_remove_req(req); 257 g_free(req); 258 } 259 260 static coroutine_fn int 261 cbw_co_preadv_snapshot(BlockDriverState *bs, int64_t offset, int64_t bytes, 262 QEMUIOVector *qiov, size_t qiov_offset) 263 { 264 BlockReq *req; 265 BdrvChild *file; 266 int ret; 267 268 /* TODO: upgrade to async loop using AioTask */ 269 while (bytes) { 270 int64_t cur_bytes; 271 272 req = cbw_snapshot_read_lock(bs, offset, bytes, &cur_bytes, &file); 273 if (!req) { 274 return -EACCES; 275 } 276 277 ret = bdrv_co_preadv_part(file, offset, cur_bytes, 278 qiov, qiov_offset, 0); 279 cbw_snapshot_read_unlock(bs, req); 280 if (ret < 0) { 281 return ret; 282 } 283 284 bytes -= cur_bytes; 285 offset += cur_bytes; 286 qiov_offset += cur_bytes; 287 } 288 289 return 0; 290 } 291 292 static int coroutine_fn 293 cbw_co_snapshot_block_status(BlockDriverState *bs, 294 bool want_zero, int64_t offset, int64_t bytes, 295 int64_t *pnum, int64_t *map, 296 BlockDriverState **file) 297 { 298 BDRVCopyBeforeWriteState *s = bs->opaque; 299 BlockReq *req; 300 int ret; 301 int64_t cur_bytes; 302 BdrvChild *child; 303 304 req = cbw_snapshot_read_lock(bs, offset, bytes, &cur_bytes, &child); 305 if (!req) { 306 return -EACCES; 307 } 308 309 ret = bdrv_block_status(child->bs, offset, cur_bytes, pnum, map, file); 310 if (child == s->target) { 311 /* 312 * We refer to s->target only for areas that we've written to it. 313 * And we can not report unallocated blocks in s->target: this will 314 * break generic block-status-above logic, that will go to 315 * copy-before-write filtered child in this case. 316 */ 317 assert(ret & BDRV_BLOCK_ALLOCATED); 318 } 319 320 cbw_snapshot_read_unlock(bs, req); 321 322 return ret; 323 } 324 325 static int coroutine_fn cbw_co_pdiscard_snapshot(BlockDriverState *bs, 326 int64_t offset, int64_t bytes) 327 { 328 BDRVCopyBeforeWriteState *s = bs->opaque; 329 330 WITH_QEMU_LOCK_GUARD(&s->lock) { 331 bdrv_reset_dirty_bitmap(s->access_bitmap, offset, bytes); 332 } 333 334 block_copy_reset(s->bcs, offset, bytes); 335 336 return bdrv_co_pdiscard(s->target, offset, bytes); 337 } 338 339 static void cbw_refresh_filename(BlockDriverState *bs) 340 { 341 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), 342 bs->file->bs->filename); 343 } 344 345 static void cbw_child_perm(BlockDriverState *bs, BdrvChild *c, 346 BdrvChildRole role, 347 BlockReopenQueue *reopen_queue, 348 uint64_t perm, uint64_t shared, 349 uint64_t *nperm, uint64_t *nshared) 350 { 351 if (!(role & BDRV_CHILD_FILTERED)) { 352 /* 353 * Target child 354 * 355 * Share write to target (child_file), to not interfere 356 * with guest writes to its disk which may be in target backing chain. 357 * Can't resize during a backup block job because we check the size 358 * only upfront. 359 */ 360 *nshared = BLK_PERM_ALL & ~BLK_PERM_RESIZE; 361 *nperm = BLK_PERM_WRITE; 362 } else { 363 /* Source child */ 364 bdrv_default_perms(bs, c, role, reopen_queue, 365 perm, shared, nperm, nshared); 366 367 if (!QLIST_EMPTY(&bs->parents)) { 368 if (perm & BLK_PERM_WRITE) { 369 *nperm = *nperm | BLK_PERM_CONSISTENT_READ; 370 } 371 *nshared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE); 372 } 373 } 374 } 375 376 static BlockdevOptions *cbw_parse_options(QDict *options, Error **errp) 377 { 378 BlockdevOptions *opts = NULL; 379 Visitor *v = NULL; 380 381 qdict_put_str(options, "driver", "copy-before-write"); 382 383 v = qobject_input_visitor_new_flat_confused(options, errp); 384 if (!v) { 385 goto out; 386 } 387 388 visit_type_BlockdevOptions(v, NULL, &opts, errp); 389 if (!opts) { 390 goto out; 391 } 392 393 /* 394 * Delete options which we are going to parse through BlockdevOptions 395 * object for original options. 396 */ 397 qdict_extract_subqdict(options, NULL, "bitmap"); 398 qdict_del(options, "on-cbw-error"); 399 qdict_del(options, "cbw-timeout"); 400 401 out: 402 visit_free(v); 403 qdict_del(options, "driver"); 404 405 return opts; 406 } 407 408 static int cbw_open(BlockDriverState *bs, QDict *options, int flags, 409 Error **errp) 410 { 411 BDRVCopyBeforeWriteState *s = bs->opaque; 412 BdrvDirtyBitmap *bitmap = NULL; 413 int64_t cluster_size; 414 g_autoptr(BlockdevOptions) full_opts = NULL; 415 BlockdevOptionsCbw *opts; 416 int ret; 417 418 full_opts = cbw_parse_options(options, errp); 419 if (!full_opts) { 420 return -EINVAL; 421 } 422 assert(full_opts->driver == BLOCKDEV_DRIVER_COPY_BEFORE_WRITE); 423 opts = &full_opts->u.copy_before_write; 424 425 ret = bdrv_open_file_child(NULL, options, "file", bs, errp); 426 if (ret < 0) { 427 return ret; 428 } 429 430 s->target = bdrv_open_child(NULL, options, "target", bs, &child_of_bds, 431 BDRV_CHILD_DATA, false, errp); 432 if (!s->target) { 433 return -EINVAL; 434 } 435 436 if (opts->bitmap) { 437 bitmap = block_dirty_bitmap_lookup(opts->bitmap->node, 438 opts->bitmap->name, NULL, errp); 439 if (!bitmap) { 440 return -EINVAL; 441 } 442 } 443 s->on_cbw_error = opts->has_on_cbw_error ? opts->on_cbw_error : 444 ON_CBW_ERROR_BREAK_GUEST_WRITE; 445 s->cbw_timeout_ns = opts->has_cbw_timeout ? 446 opts->cbw_timeout * NANOSECONDS_PER_SECOND : 0; 447 448 bs->total_sectors = bs->file->bs->total_sectors; 449 bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED | 450 (BDRV_REQ_FUA & bs->file->bs->supported_write_flags); 451 bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED | 452 ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK) & 453 bs->file->bs->supported_zero_flags); 454 455 s->bcs = block_copy_state_new(bs->file, s->target, bitmap, errp); 456 if (!s->bcs) { 457 error_prepend(errp, "Cannot create block-copy-state: "); 458 return -EINVAL; 459 } 460 461 cluster_size = block_copy_cluster_size(s->bcs); 462 463 s->done_bitmap = bdrv_create_dirty_bitmap(bs, cluster_size, NULL, errp); 464 if (!s->done_bitmap) { 465 return -EINVAL; 466 } 467 bdrv_disable_dirty_bitmap(s->done_bitmap); 468 469 /* s->access_bitmap starts equal to bcs bitmap */ 470 s->access_bitmap = bdrv_create_dirty_bitmap(bs, cluster_size, NULL, errp); 471 if (!s->access_bitmap) { 472 return -EINVAL; 473 } 474 bdrv_disable_dirty_bitmap(s->access_bitmap); 475 bdrv_dirty_bitmap_merge_internal(s->access_bitmap, 476 block_copy_dirty_bitmap(s->bcs), NULL, 477 true); 478 479 qemu_co_mutex_init(&s->lock); 480 QLIST_INIT(&s->frozen_read_reqs); 481 482 return 0; 483 } 484 485 static void cbw_close(BlockDriverState *bs) 486 { 487 BDRVCopyBeforeWriteState *s = bs->opaque; 488 489 bdrv_release_dirty_bitmap(s->access_bitmap); 490 bdrv_release_dirty_bitmap(s->done_bitmap); 491 492 block_copy_state_free(s->bcs); 493 s->bcs = NULL; 494 } 495 496 BlockDriver bdrv_cbw_filter = { 497 .format_name = "copy-before-write", 498 .instance_size = sizeof(BDRVCopyBeforeWriteState), 499 500 .bdrv_open = cbw_open, 501 .bdrv_close = cbw_close, 502 503 .bdrv_co_preadv = cbw_co_preadv, 504 .bdrv_co_pwritev = cbw_co_pwritev, 505 .bdrv_co_pwrite_zeroes = cbw_co_pwrite_zeroes, 506 .bdrv_co_pdiscard = cbw_co_pdiscard, 507 .bdrv_co_flush = cbw_co_flush, 508 509 .bdrv_co_preadv_snapshot = cbw_co_preadv_snapshot, 510 .bdrv_co_pdiscard_snapshot = cbw_co_pdiscard_snapshot, 511 .bdrv_co_snapshot_block_status = cbw_co_snapshot_block_status, 512 513 .bdrv_refresh_filename = cbw_refresh_filename, 514 515 .bdrv_child_perm = cbw_child_perm, 516 517 .is_filter = true, 518 }; 519 520 BlockDriverState *bdrv_cbw_append(BlockDriverState *source, 521 BlockDriverState *target, 522 const char *filter_node_name, 523 BlockCopyState **bcs, 524 Error **errp) 525 { 526 BDRVCopyBeforeWriteState *state; 527 BlockDriverState *top; 528 QDict *opts; 529 530 assert(source->total_sectors == target->total_sectors); 531 GLOBAL_STATE_CODE(); 532 533 opts = qdict_new(); 534 qdict_put_str(opts, "driver", "copy-before-write"); 535 if (filter_node_name) { 536 qdict_put_str(opts, "node-name", filter_node_name); 537 } 538 qdict_put_str(opts, "file", bdrv_get_node_name(source)); 539 qdict_put_str(opts, "target", bdrv_get_node_name(target)); 540 541 top = bdrv_insert_node(source, opts, BDRV_O_RDWR, errp); 542 if (!top) { 543 return NULL; 544 } 545 546 state = top->opaque; 547 *bcs = state->bcs; 548 549 return top; 550 } 551 552 void bdrv_cbw_drop(BlockDriverState *bs) 553 { 554 GLOBAL_STATE_CODE(); 555 bdrv_drop_filter(bs, &error_abort); 556 bdrv_unref(bs); 557 } 558 559 static void cbw_init(void) 560 { 561 bdrv_register(&bdrv_cbw_filter); 562 } 563 564 block_init(cbw_init); 565