tests/test-bdrv-drain.c

   1 /*
   2  * Block node draining tests
   3  *
   4  * Copyright (c) 2017 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.h"
  27 #include "block/blockjob_int.h"
  28 #include "sysemu/block-backend.h"
  29 #include "qapi/error.h"
  30 #include "iothread.h"
  31
  32 static QemuEvent done_event;
  33
  34 typedef struct BDRVTestState {
  35     int drain_count;
  36     AioContext *bh_indirection_ctx;
  37     bool sleep_in_drain_begin;
  38 } BDRVTestState;
  39
  40 static void coroutine_fn bdrv_test_co_drain_begin(BlockDriverState *bs)
  41 {
  42     BDRVTestState *s = bs->opaque;
  43     s->drain_count++;
  44     if (s->sleep_in_drain_begin) {
  45         qemu_co_sleep_ns(QEMU_CLOCK_REALTIME, 100000);
  46     }
  47 }
  48
  49 static void coroutine_fn bdrv_test_co_drain_end(BlockDriverState *bs)
  50 {
  51     BDRVTestState *s = bs->opaque;
  52     s->drain_count--;
  53 }
  54
  55 static void bdrv_test_close(BlockDriverState *bs)
  56 {
  57     BDRVTestState *s = bs->opaque;
  58     g_assert_cmpint(s->drain_count, >, 0);
  59 }
  60
  61 static void co_reenter_bh(void *opaque)
  62 {
  63     aio_co_wake(opaque);
  64 }
  65
  66 static int coroutine_fn bdrv_test_co_preadv(BlockDriverState *bs,
  67                                             uint64_t offset, uint64_t bytes,
  68                                             QEMUIOVector *qiov, int flags)
  69 {
  70     BDRVTestState *s = bs->opaque;
  71
  72     /* We want this request to stay until the polling loop in drain waits for
  73      * it to complete. We need to sleep a while as bdrv_drain_invoke() comes
  74      * first and polls its result, too, but it shouldn't accidentally complete
  75      * this request yet. */
  76     qemu_co_sleep_ns(QEMU_CLOCK_REALTIME, 100000);
  77
  78     if (s->bh_indirection_ctx) {
  79         aio_bh_schedule_oneshot(s->bh_indirection_ctx, co_reenter_bh,
  80                                 qemu_coroutine_self());
  81         qemu_coroutine_yield();
  82     }
  83
  84     return 0;
  85 }
  86
  87 static void bdrv_test_child_perm(BlockDriverState *bs, BdrvChild *c,
  88                                  const BdrvChildRole *role,
  89                                  BlockReopenQueue *reopen_queue,
  90                                  uint64_t perm, uint64_t shared,
  91                                  uint64_t *nperm, uint64_t *nshared)
  92 {
  93     /* bdrv_format_default_perms() accepts only these two, so disguise
  94      * detach_by_driver_cb_role as one of them. */
  95     if (role != &child_file && role != &child_backing) {
  96         role = &child_file;
  97     }
  98
  99     bdrv_format_default_perms(bs, c, role, reopen_queue, perm, shared,
 100                               nperm, nshared);
 101 }
 102
 103 static BlockDriver bdrv_test = {
 104     .format_name            = "test",
 105     .instance_size          = sizeof(BDRVTestState),
 106
 107     .bdrv_close             = bdrv_test_close,
 108     .bdrv_co_preadv         = bdrv_test_co_preadv,
 109
 110     .bdrv_co_drain_begin    = bdrv_test_co_drain_begin,
 111     .bdrv_co_drain_end      = bdrv_test_co_drain_end,
 112
 113     .bdrv_child_perm        = bdrv_test_child_perm,
 114 };
 115
 116 static void aio_ret_cb(void *opaque, int ret)
 117 {
 118     int *aio_ret = opaque;
 119     *aio_ret = ret;
 120 }
 121
 122 typedef struct CallInCoroutineData {
 123     void (*entry)(void);
 124     bool done;
 125 } CallInCoroutineData;
 126
 127 static coroutine_fn void call_in_coroutine_entry(void *opaque)
 128 {
 129     CallInCoroutineData *data = opaque;
 130
 131     data->entry();
 132     data->done = true;
 133 }
 134
 135 static void call_in_coroutine(void (*entry)(void))
 136 {
 137     Coroutine *co;
 138     CallInCoroutineData data = {
 139         .entry  = entry,
 140         .done   = false,
 141     };
 142
 143     co = qemu_coroutine_create(call_in_coroutine_entry, &data);
 144     qemu_coroutine_enter(co);
 145     while (!data.done) {
 146         aio_poll(qemu_get_aio_context(), true);
 147     }
 148 }
 149
 150 enum drain_type {
 151     BDRV_DRAIN_ALL,
 152     BDRV_DRAIN,
 153     BDRV_SUBTREE_DRAIN,
 154     DRAIN_TYPE_MAX,
 155 };
 156
 157 static void do_drain_begin(enum drain_type drain_type, BlockDriverState *bs)
 158 {
 159     switch (drain_type) {
 160     case BDRV_DRAIN_ALL:        bdrv_drain_all_begin(); break;
 161     case BDRV_DRAIN:            bdrv_drained_begin(bs); break;
 162     case BDRV_SUBTREE_DRAIN:    bdrv_subtree_drained_begin(bs); break;
 163     default:                    g_assert_not_reached();
 164     }
 165 }
 166
 167 static void do_drain_end(enum drain_type drain_type, BlockDriverState *bs)
 168 {
 169     switch (drain_type) {
 170     case BDRV_DRAIN_ALL:        bdrv_drain_all_end(); break;
 171     case BDRV_DRAIN:            bdrv_drained_end(bs); break;
 172     case BDRV_SUBTREE_DRAIN:    bdrv_subtree_drained_end(bs); break;
 173     default:                    g_assert_not_reached();
 174     }
 175 }
 176
 177 static void test_drv_cb_common(enum drain_type drain_type, bool recursive)
 178 {
 179     BlockBackend *blk;
 180     BlockDriverState *bs, *backing;
 181     BDRVTestState *s, *backing_s;
 182     BlockAIOCB *acb;
 183     int aio_ret;
 184
 185     QEMUIOVector qiov;
 186     struct iovec iov = {
 187         .iov_base = NULL,
 188         .iov_len = 0,
 189     };
 190     qemu_iovec_init_external(&qiov, &iov, 1);
 191
 192     blk = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
 193     bs = bdrv_new_open_driver(&bdrv_test, "test-node", BDRV_O_RDWR,
 194                               &error_abort);
 195     s = bs->opaque;
 196     blk_insert_bs(blk, bs, &error_abort);
 197
 198     backing = bdrv_new_open_driver(&bdrv_test, "backing", 0, &error_abort);
 199     backing_s = backing->opaque;
 200     bdrv_set_backing_hd(bs, backing, &error_abort);
 201
 202     /* Simple bdrv_drain_all_begin/end pair, check that CBs are called */
 203     g_assert_cmpint(s->drain_count, ==, 0);
 204     g_assert_cmpint(backing_s->drain_count, ==, 0);
 205
 206     do_drain_begin(drain_type, bs);
 207
 208     g_assert_cmpint(s->drain_count, ==, 1);
 209     g_assert_cmpint(backing_s->drain_count, ==, !!recursive);
 210
 211     do_drain_end(drain_type, bs);
 212
 213     g_assert_cmpint(s->drain_count, ==, 0);
 214     g_assert_cmpint(backing_s->drain_count, ==, 0);
 215
 216     /* Now do the same while a request is pending */
 217     aio_ret = -EINPROGRESS;
 218     acb = blk_aio_preadv(blk, 0, &qiov, 0, aio_ret_cb, &aio_ret);
 219     g_assert(acb != NULL);
 220     g_assert_cmpint(aio_ret, ==, -EINPROGRESS);
 221
 222     g_assert_cmpint(s->drain_count, ==, 0);
 223     g_assert_cmpint(backing_s->drain_count, ==, 0);
 224
 225     do_drain_begin(drain_type, bs);
 226
 227     g_assert_cmpint(aio_ret, ==, 0);
 228     g_assert_cmpint(s->drain_count, ==, 1);
 229     g_assert_cmpint(backing_s->drain_count, ==, !!recursive);
 230
 231     do_drain_end(drain_type, bs);
 232
 233     g_assert_cmpint(s->drain_count, ==, 0);
 234     g_assert_cmpint(backing_s->drain_count, ==, 0);
 235
 236     bdrv_unref(backing);
 237     bdrv_unref(bs);
 238     blk_unref(blk);
 239 }
 240
 241 static void test_drv_cb_drain_all(void)
 242 {
 243     test_drv_cb_common(BDRV_DRAIN_ALL, true);
 244 }
 245
 246 static void test_drv_cb_drain(void)
 247 {
 248     test_drv_cb_common(BDRV_DRAIN, false);
 249 }
 250
 251 static void test_drv_cb_drain_subtree(void)
 252 {
 253     test_drv_cb_common(BDRV_SUBTREE_DRAIN, true);
 254 }
 255
 256 static void test_drv_cb_co_drain_all(void)
 257 {
 258     call_in_coroutine(test_drv_cb_drain_all);
 259 }
 260
 261 static void test_drv_cb_co_drain(void)
 262 {
 263     call_in_coroutine(test_drv_cb_drain);
 264 }
 265
 266 static void test_drv_cb_co_drain_subtree(void)
 267 {
 268     call_in_coroutine(test_drv_cb_drain_subtree);
 269 }
 270
 271 static void test_quiesce_common(enum drain_type drain_type, bool recursive)
 272 {
 273     BlockBackend *blk;
 274     BlockDriverState *bs, *backing;
 275
 276     blk = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
 277     bs = bdrv_new_open_driver(&bdrv_test, "test-node", BDRV_O_RDWR,
 278                               &error_abort);
 279     blk_insert_bs(blk, bs, &error_abort);
 280
 281     backing = bdrv_new_open_driver(&bdrv_test, "backing", 0, &error_abort);
 282     bdrv_set_backing_hd(bs, backing, &error_abort);
 283
 284     g_assert_cmpint(bs->quiesce_counter, ==, 0);
 285     g_assert_cmpint(backing->quiesce_counter, ==, 0);
 286
 287     do_drain_begin(drain_type, bs);
 288
 289     g_assert_cmpint(bs->quiesce_counter, ==, 1);
 290     g_assert_cmpint(backing->quiesce_counter, ==, !!recursive);
 291
 292     do_drain_end(drain_type, bs);
 293
 294     g_assert_cmpint(bs->quiesce_counter, ==, 0);
 295     g_assert_cmpint(backing->quiesce_counter, ==, 0);
 296
 297     bdrv_unref(backing);
 298     bdrv_unref(bs);
 299     blk_unref(blk);
 300 }
 301
 302 static void test_quiesce_drain_all(void)
 303 {
 304     test_quiesce_common(BDRV_DRAIN_ALL, true);
 305 }
 306
 307 static void test_quiesce_drain(void)
 308 {
 309     test_quiesce_common(BDRV_DRAIN, false);
 310 }
 311
 312 static void test_quiesce_drain_subtree(void)
 313 {
 314     test_quiesce_common(BDRV_SUBTREE_DRAIN, true);
 315 }
 316
 317 static void test_quiesce_co_drain_all(void)
 318 {
 319     call_in_coroutine(test_quiesce_drain_all);
 320 }
 321
 322 static void test_quiesce_co_drain(void)
 323 {
 324     call_in_coroutine(test_quiesce_drain);
 325 }
 326
 327 static void test_quiesce_co_drain_subtree(void)
 328 {
 329     call_in_coroutine(test_quiesce_drain_subtree);
 330 }
 331
 332 static void test_nested(void)
 333 {
 334     BlockBackend *blk;
 335     BlockDriverState *bs, *backing;
 336     BDRVTestState *s, *backing_s;
 337     enum drain_type outer, inner;
 338
 339     blk = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
 340     bs = bdrv_new_open_driver(&bdrv_test, "test-node", BDRV_O_RDWR,
 341                               &error_abort);
 342     s = bs->opaque;
 343     blk_insert_bs(blk, bs, &error_abort);
 344
 345     backing = bdrv_new_open_driver(&bdrv_test, "backing", 0, &error_abort);
 346     backing_s = backing->opaque;
 347     bdrv_set_backing_hd(bs, backing, &error_abort);
 348
 349     for (outer = 0; outer < DRAIN_TYPE_MAX; outer++) {
 350         for (inner = 0; inner < DRAIN_TYPE_MAX; inner++) {
 351             int backing_quiesce = (outer != BDRV_DRAIN) +
 352                                   (inner != BDRV_DRAIN);
 353
 354             g_assert_cmpint(bs->quiesce_counter, ==, 0);
 355             g_assert_cmpint(backing->quiesce_counter, ==, 0);
 356             g_assert_cmpint(s->drain_count, ==, 0);
 357             g_assert_cmpint(backing_s->drain_count, ==, 0);
 358
 359             do_drain_begin(outer, bs);
 360             do_drain_begin(inner, bs);
 361
 362             g_assert_cmpint(bs->quiesce_counter, ==, 2);
 363             g_assert_cmpint(backing->quiesce_counter, ==, backing_quiesce);
 364             g_assert_cmpint(s->drain_count, ==, 2);
 365             g_assert_cmpint(backing_s->drain_count, ==, backing_quiesce);
 366
 367             do_drain_end(inner, bs);
 368             do_drain_end(outer, bs);
 369
 370             g_assert_cmpint(bs->quiesce_counter, ==, 0);
 371             g_assert_cmpint(backing->quiesce_counter, ==, 0);
 372             g_assert_cmpint(s->drain_count, ==, 0);
 373             g_assert_cmpint(backing_s->drain_count, ==, 0);
 374         }
 375     }
 376
 377     bdrv_unref(backing);
 378     bdrv_unref(bs);
 379     blk_unref(blk);
 380 }
 381
 382 static void test_multiparent(void)
 383 {
 384     BlockBackend *blk_a, *blk_b;
 385     BlockDriverState *bs_a, *bs_b, *backing;
 386     BDRVTestState *a_s, *b_s, *backing_s;
 387
 388     blk_a = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
 389     bs_a = bdrv_new_open_driver(&bdrv_test, "test-node-a", BDRV_O_RDWR,
 390                                 &error_abort);
 391     a_s = bs_a->opaque;
 392     blk_insert_bs(blk_a, bs_a, &error_abort);
 393
 394     blk_b = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
 395     bs_b = bdrv_new_open_driver(&bdrv_test, "test-node-b", BDRV_O_RDWR,
 396                                 &error_abort);
 397     b_s = bs_b->opaque;
 398     blk_insert_bs(blk_b, bs_b, &error_abort);
 399
 400     backing = bdrv_new_open_driver(&bdrv_test, "backing", 0, &error_abort);
 401     backing_s = backing->opaque;
 402     bdrv_set_backing_hd(bs_a, backing, &error_abort);
 403     bdrv_set_backing_hd(bs_b, backing, &error_abort);
 404
 405     g_assert_cmpint(bs_a->quiesce_counter, ==, 0);
 406     g_assert_cmpint(bs_b->quiesce_counter, ==, 0);
 407     g_assert_cmpint(backing->quiesce_counter, ==, 0);
 408     g_assert_cmpint(a_s->drain_count, ==, 0);
 409     g_assert_cmpint(b_s->drain_count, ==, 0);
 410     g_assert_cmpint(backing_s->drain_count, ==, 0);
 411
 412     do_drain_begin(BDRV_SUBTREE_DRAIN, bs_a);
 413
 414     g_assert_cmpint(bs_a->quiesce_counter, ==, 1);
 415     g_assert_cmpint(bs_b->quiesce_counter, ==, 1);
 416     g_assert_cmpint(backing->quiesce_counter, ==, 1);
 417     g_assert_cmpint(a_s->drain_count, ==, 1);
 418     g_assert_cmpint(b_s->drain_count, ==, 1);
 419     g_assert_cmpint(backing_s->drain_count, ==, 1);
 420
 421     do_drain_begin(BDRV_SUBTREE_DRAIN, bs_b);
 422
 423     g_assert_cmpint(bs_a->quiesce_counter, ==, 2);
 424     g_assert_cmpint(bs_b->quiesce_counter, ==, 2);
 425     g_assert_cmpint(backing->quiesce_counter, ==, 2);
 426     g_assert_cmpint(a_s->drain_count, ==, 2);
 427     g_assert_cmpint(b_s->drain_count, ==, 2);
 428     g_assert_cmpint(backing_s->drain_count, ==, 2);
 429
 430     do_drain_end(BDRV_SUBTREE_DRAIN, bs_b);
 431
 432     g_assert_cmpint(bs_a->quiesce_counter, ==, 1);
 433     g_assert_cmpint(bs_b->quiesce_counter, ==, 1);
 434     g_assert_cmpint(backing->quiesce_counter, ==, 1);
 435     g_assert_cmpint(a_s->drain_count, ==, 1);
 436     g_assert_cmpint(b_s->drain_count, ==, 1);
 437     g_assert_cmpint(backing_s->drain_count, ==, 1);
 438
 439     do_drain_end(BDRV_SUBTREE_DRAIN, bs_a);
 440
 441     g_assert_cmpint(bs_a->quiesce_counter, ==, 0);
 442     g_assert_cmpint(bs_b->quiesce_counter, ==, 0);
 443     g_assert_cmpint(backing->quiesce_counter, ==, 0);
 444     g_assert_cmpint(a_s->drain_count, ==, 0);
 445     g_assert_cmpint(b_s->drain_count, ==, 0);
 446     g_assert_cmpint(backing_s->drain_count, ==, 0);
 447
 448     bdrv_unref(backing);
 449     bdrv_unref(bs_a);
 450     bdrv_unref(bs_b);
 451     blk_unref(blk_a);
 452     blk_unref(blk_b);
 453 }
 454
 455 static void test_graph_change_drain_subtree(void)
 456 {
 457     BlockBackend *blk_a, *blk_b;
 458     BlockDriverState *bs_a, *bs_b, *backing;
 459     BDRVTestState *a_s, *b_s, *backing_s;
 460
 461     blk_a = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
 462     bs_a = bdrv_new_open_driver(&bdrv_test, "test-node-a", BDRV_O_RDWR,
 463                                 &error_abort);
 464     a_s = bs_a->opaque;
 465     blk_insert_bs(blk_a, bs_a, &error_abort);
 466
 467     blk_b = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
 468     bs_b = bdrv_new_open_driver(&bdrv_test, "test-node-b", BDRV_O_RDWR,
 469                                 &error_abort);
 470     b_s = bs_b->opaque;
 471     blk_insert_bs(blk_b, bs_b, &error_abort);
 472
 473     backing = bdrv_new_open_driver(&bdrv_test, "backing", 0, &error_abort);
 474     backing_s = backing->opaque;
 475     bdrv_set_backing_hd(bs_a, backing, &error_abort);
 476
 477     g_assert_cmpint(bs_a->quiesce_counter, ==, 0);
 478     g_assert_cmpint(bs_b->quiesce_counter, ==, 0);
 479     g_assert_cmpint(backing->quiesce_counter, ==, 0);
 480     g_assert_cmpint(a_s->drain_count, ==, 0);
 481     g_assert_cmpint(b_s->drain_count, ==, 0);
 482     g_assert_cmpint(backing_s->drain_count, ==, 0);
 483
 484     do_drain_begin(BDRV_SUBTREE_DRAIN, bs_a);
 485     do_drain_begin(BDRV_SUBTREE_DRAIN, bs_a);
 486     do_drain_begin(BDRV_SUBTREE_DRAIN, bs_a);
 487     do_drain_begin(BDRV_SUBTREE_DRAIN, bs_b);
 488     do_drain_begin(BDRV_SUBTREE_DRAIN, bs_b);
 489
 490     bdrv_set_backing_hd(bs_b, backing, &error_abort);
 491     g_assert_cmpint(bs_a->quiesce_counter, ==, 5);
 492     g_assert_cmpint(bs_b->quiesce_counter, ==, 5);
 493     g_assert_cmpint(backing->quiesce_counter, ==, 5);
 494     g_assert_cmpint(a_s->drain_count, ==, 5);
 495     g_assert_cmpint(b_s->drain_count, ==, 5);
 496     g_assert_cmpint(backing_s->drain_count, ==, 5);
 497
 498     bdrv_set_backing_hd(bs_b, NULL, &error_abort);
 499     g_assert_cmpint(bs_a->quiesce_counter, ==, 3);
 500     g_assert_cmpint(bs_b->quiesce_counter, ==, 2);
 501     g_assert_cmpint(backing->quiesce_counter, ==, 3);
 502     g_assert_cmpint(a_s->drain_count, ==, 3);
 503     g_assert_cmpint(b_s->drain_count, ==, 2);
 504     g_assert_cmpint(backing_s->drain_count, ==, 3);
 505
 506     bdrv_set_backing_hd(bs_b, backing, &error_abort);
 507     g_assert_cmpint(bs_a->quiesce_counter, ==, 5);
 508     g_assert_cmpint(bs_b->quiesce_counter, ==, 5);
 509     g_assert_cmpint(backing->quiesce_counter, ==, 5);
 510     g_assert_cmpint(a_s->drain_count, ==, 5);
 511     g_assert_cmpint(b_s->drain_count, ==, 5);
 512     g_assert_cmpint(backing_s->drain_count, ==, 5);
 513
 514     do_drain_end(BDRV_SUBTREE_DRAIN, bs_b);
 515     do_drain_end(BDRV_SUBTREE_DRAIN, bs_b);
 516     do_drain_end(BDRV_SUBTREE_DRAIN, bs_a);
 517     do_drain_end(BDRV_SUBTREE_DRAIN, bs_a);
 518     do_drain_end(BDRV_SUBTREE_DRAIN, bs_a);
 519
 520     g_assert_cmpint(bs_a->quiesce_counter, ==, 0);
 521     g_assert_cmpint(bs_b->quiesce_counter, ==, 0);
 522     g_assert_cmpint(backing->quiesce_counter, ==, 0);
 523     g_assert_cmpint(a_s->drain_count, ==, 0);
 524     g_assert_cmpint(b_s->drain_count, ==, 0);
 525     g_assert_cmpint(backing_s->drain_count, ==, 0);
 526
 527     bdrv_unref(backing);
 528     bdrv_unref(bs_a);
 529     bdrv_unref(bs_b);
 530     blk_unref(blk_a);
 531     blk_unref(blk_b);
 532 }
 533
 534 static void test_graph_change_drain_all(void)
 535 {
 536     BlockBackend *blk_a, *blk_b;
 537     BlockDriverState *bs_a, *bs_b;
 538     BDRVTestState *a_s, *b_s;
 539
 540     /* Create node A with a BlockBackend */
 541     blk_a = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
 542     bs_a = bdrv_new_open_driver(&bdrv_test, "test-node-a", BDRV_O_RDWR,
 543                                 &error_abort);
 544     a_s = bs_a->opaque;
 545     blk_insert_bs(blk_a, bs_a, &error_abort);
 546
 547     g_assert_cmpint(bs_a->quiesce_counter, ==, 0);
 548     g_assert_cmpint(a_s->drain_count, ==, 0);
 549
 550     /* Call bdrv_drain_all_begin() */
 551     bdrv_drain_all_begin();
 552
 553     g_assert_cmpint(bs_a->quiesce_counter, ==, 1);
 554     g_assert_cmpint(a_s->drain_count, ==, 1);
 555
 556     /* Create node B with a BlockBackend */
 557     blk_b = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
 558     bs_b = bdrv_new_open_driver(&bdrv_test, "test-node-b", BDRV_O_RDWR,
 559                                 &error_abort);
 560     b_s = bs_b->opaque;
 561     blk_insert_bs(blk_b, bs_b, &error_abort);
 562
 563     g_assert_cmpint(bs_a->quiesce_counter, ==, 1);
 564     g_assert_cmpint(bs_b->quiesce_counter, ==, 1);
 565     g_assert_cmpint(a_s->drain_count, ==, 1);
 566     g_assert_cmpint(b_s->drain_count, ==, 1);
 567
 568     /* Unref and finally delete node A */
 569     blk_unref(blk_a);
 570
 571     g_assert_cmpint(bs_a->quiesce_counter, ==, 1);
 572     g_assert_cmpint(bs_b->quiesce_counter, ==, 1);
 573     g_assert_cmpint(a_s->drain_count, ==, 1);
 574     g_assert_cmpint(b_s->drain_count, ==, 1);
 575
 576     bdrv_unref(bs_a);
 577
 578     g_assert_cmpint(bs_b->quiesce_counter, ==, 1);
 579     g_assert_cmpint(b_s->drain_count, ==, 1);
 580
 581     /* End the drained section */
 582     bdrv_drain_all_end();
 583
 584     g_assert_cmpint(bs_b->quiesce_counter, ==, 0);
 585     g_assert_cmpint(b_s->drain_count, ==, 0);
 586
 587     bdrv_unref(bs_b);
 588     blk_unref(blk_b);
 589 }
 590
 591 struct test_iothread_data {
 592     BlockDriverState *bs;
 593     enum drain_type drain_type;
 594     int *aio_ret;
 595 };
 596
 597 static void test_iothread_drain_entry(void *opaque)
 598 {
 599     struct test_iothread_data *data = opaque;
 600
 601     aio_context_acquire(bdrv_get_aio_context(data->bs));
 602     do_drain_begin(data->drain_type, data->bs);
 603     g_assert_cmpint(*data->aio_ret, ==, 0);
 604     do_drain_end(data->drain_type, data->bs);
 605     aio_context_release(bdrv_get_aio_context(data->bs));
 606
 607     qemu_event_set(&done_event);
 608 }
 609
 610 static void test_iothread_aio_cb(void *opaque, int ret)
 611 {
 612     int *aio_ret = opaque;
 613     *aio_ret = ret;
 614     qemu_event_set(&done_event);
 615 }
 616
 617 /*
 618  * Starts an AIO request on a BDS that runs in the AioContext of iothread 1.
 619  * The request involves a BH on iothread 2 before it can complete.
 620  *
 621  * @drain_thread = 0 means that do_drain_begin/end are called from the main
 622  * thread, @drain_thread = 1 means that they are called from iothread 1. Drain
 623  * for this BDS cannot be called from iothread 2 because only the main thread
 624  * may do cross-AioContext polling.
 625  */
 626 static void test_iothread_common(enum drain_type drain_type, int drain_thread)
 627 {
 628     BlockBackend *blk;
 629     BlockDriverState *bs;
 630     BDRVTestState *s;
 631     BlockAIOCB *acb;
 632     int aio_ret;
 633     struct test_iothread_data data;
 634
 635     IOThread *a = iothread_new();
 636     IOThread *b = iothread_new();
 637     AioContext *ctx_a = iothread_get_aio_context(a);
 638     AioContext *ctx_b = iothread_get_aio_context(b);
 639
 640     QEMUIOVector qiov;
 641     struct iovec iov = {
 642         .iov_base = NULL,
 643         .iov_len = 0,
 644     };
 645     qemu_iovec_init_external(&qiov, &iov, 1);
 646
 647     /* bdrv_drain_all() may only be called from the main loop thread */
 648     if (drain_type == BDRV_DRAIN_ALL && drain_thread != 0) {
 649         goto out;
 650     }
 651
 652     blk = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
 653     bs = bdrv_new_open_driver(&bdrv_test, "test-node", BDRV_O_RDWR,
 654                               &error_abort);
 655     s = bs->opaque;
 656     blk_insert_bs(blk, bs, &error_abort);
 657
 658     blk_set_aio_context(blk, ctx_a);
 659     aio_context_acquire(ctx_a);
 660
 661     s->bh_indirection_ctx = ctx_b;
 662
 663     aio_ret = -EINPROGRESS;
 664     if (drain_thread == 0) {
 665         acb = blk_aio_preadv(blk, 0, &qiov, 0, test_iothread_aio_cb, &aio_ret);
 666     } else {
 667         acb = blk_aio_preadv(blk, 0, &qiov, 0, aio_ret_cb, &aio_ret);
 668     }
 669     g_assert(acb != NULL);
 670     g_assert_cmpint(aio_ret, ==, -EINPROGRESS);
 671
 672     aio_context_release(ctx_a);
 673
 674     data = (struct test_iothread_data) {
 675         .bs         = bs,
 676         .drain_type = drain_type,
 677         .aio_ret    = &aio_ret,
 678     };
 679
 680     switch (drain_thread) {
 681     case 0:
 682         if (drain_type != BDRV_DRAIN_ALL) {
 683             aio_context_acquire(ctx_a);
 684         }
 685
 686         /* The request is running on the IOThread a. Draining its block device
 687          * will make sure that it has completed as far as the BDS is concerned,
 688          * but the drain in this thread can continue immediately after
 689          * bdrv_dec_in_flight() and aio_ret might be assigned only slightly
 690          * later. */
 691         qemu_event_reset(&done_event);
 692         do_drain_begin(drain_type, bs);
 693         g_assert_cmpint(bs->in_flight, ==, 0);
 694
 695         if (drain_type != BDRV_DRAIN_ALL) {
 696             aio_context_release(ctx_a);
 697         }
 698         qemu_event_wait(&done_event);
 699         if (drain_type != BDRV_DRAIN_ALL) {
 700             aio_context_acquire(ctx_a);
 701         }
 702
 703         g_assert_cmpint(aio_ret, ==, 0);
 704         do_drain_end(drain_type, bs);
 705
 706         if (drain_type != BDRV_DRAIN_ALL) {
 707             aio_context_release(ctx_a);
 708         }
 709         break;
 710     case 1:
 711         qemu_event_reset(&done_event);
 712         aio_bh_schedule_oneshot(ctx_a, test_iothread_drain_entry, &data);
 713         qemu_event_wait(&done_event);
 714         break;
 715     default:
 716         g_assert_not_reached();
 717     }
 718
 719     aio_context_acquire(ctx_a);
 720     blk_set_aio_context(blk, qemu_get_aio_context());
 721     aio_context_release(ctx_a);
 722
 723     bdrv_unref(bs);
 724     blk_unref(blk);
 725
 726 out:
 727     iothread_join(a);
 728     iothread_join(b);
 729 }
 730
 731 static void test_iothread_drain_all(void)
 732 {
 733     test_iothread_common(BDRV_DRAIN_ALL, 0);
 734     test_iothread_common(BDRV_DRAIN_ALL, 1);
 735 }
 736
 737 static void test_iothread_drain(void)
 738 {
 739     test_iothread_common(BDRV_DRAIN, 0);
 740     test_iothread_common(BDRV_DRAIN, 1);
 741 }
 742
 743 static void test_iothread_drain_subtree(void)
 744 {
 745     test_iothread_common(BDRV_SUBTREE_DRAIN, 0);
 746     test_iothread_common(BDRV_SUBTREE_DRAIN, 1);
 747 }
 748
 749
 750 typedef struct TestBlockJob {
 751     BlockJob common;
 752     bool should_complete;
 753 } TestBlockJob;
 754
 755 static void test_job_completed(Job *job, void *opaque)
 756 {
 757     job_completed(job, 0, NULL);
 758 }
 759
 760 static void coroutine_fn test_job_start(void *opaque)
 761 {
 762     TestBlockJob *s = opaque;
 763
 764     job_transition_to_ready(&s->common.job);
 765     while (!s->should_complete) {
 766         /* Avoid block_job_sleep_ns() because it marks the job as !busy. We
 767          * want to emulate some actual activity (probably some I/O) here so
 768          * that drain has to wait for this acitivity to stop. */
 769         qemu_co_sleep_ns(QEMU_CLOCK_REALTIME, 100000);
 770         job_pause_point(&s->common.job);
 771     }
 772
 773     job_defer_to_main_loop(&s->common.job, test_job_completed, NULL);
 774 }
 775
 776 static void test_job_complete(Job *job, Error **errp)
 777 {
 778     TestBlockJob *s = container_of(job, TestBlockJob, common.job);
 779     s->should_complete = true;
 780 }
 781
 782 BlockJobDriver test_job_driver = {
 783     .job_driver = {
 784         .instance_size  = sizeof(TestBlockJob),
 785         .free           = block_job_free,
 786         .user_resume    = block_job_user_resume,
 787         .drain          = block_job_drain,
 788         .start          = test_job_start,
 789         .complete       = test_job_complete,
 790     },
 791 };
 792
 793 static void test_blockjob_common(enum drain_type drain_type)
 794 {
 795     BlockBackend *blk_src, *blk_target;
 796     BlockDriverState *src, *target;
 797     BlockJob *job;
 798     int ret;
 799
 800     src = bdrv_new_open_driver(&bdrv_test, "source", BDRV_O_RDWR,
 801                                &error_abort);
 802     blk_src = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
 803     blk_insert_bs(blk_src, src, &error_abort);
 804
 805     target = bdrv_new_open_driver(&bdrv_test, "target", BDRV_O_RDWR,
 806                                   &error_abort);
 807     blk_target = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
 808     blk_insert_bs(blk_target, target, &error_abort);
 809
 810     job = block_job_create("job0", &test_job_driver, NULL, src, 0, BLK_PERM_ALL,
 811                            0, 0, NULL, NULL, &error_abort);
 812     block_job_add_bdrv(job, "target", target, 0, BLK_PERM_ALL, &error_abort);
 813     job_start(&job->job);
 814
 815     g_assert_cmpint(job->job.pause_count, ==, 0);
 816     g_assert_false(job->job.paused);
 817     g_assert_true(job->job.busy); /* We're in job_sleep_ns() */
 818
 819     do_drain_begin(drain_type, src);
 820
 821     if (drain_type == BDRV_DRAIN_ALL) {
 822         /* bdrv_drain_all() drains both src and target */
 823         g_assert_cmpint(job->job.pause_count, ==, 2);
 824     } else {
 825         g_assert_cmpint(job->job.pause_count, ==, 1);
 826     }
 827     g_assert_true(job->job.paused);
 828     g_assert_false(job->job.busy); /* The job is paused */
 829
 830     do_drain_end(drain_type, src);
 831
 832     g_assert_cmpint(job->job.pause_count, ==, 0);
 833     g_assert_false(job->job.paused);
 834     g_assert_true(job->job.busy); /* We're in qemu_co_sleep_ns() */
 835
 836     do_drain_begin(drain_type, target);
 837
 838     if (drain_type == BDRV_DRAIN_ALL) {
 839         /* bdrv_drain_all() drains both src and target */
 840         g_assert_cmpint(job->job.pause_count, ==, 2);
 841     } else {
 842         g_assert_cmpint(job->job.pause_count, ==, 1);
 843     }
 844     g_assert_true(job->job.paused);
 845     g_assert_false(job->job.busy); /* The job is paused */
 846
 847     do_drain_end(drain_type, target);
 848
 849     g_assert_cmpint(job->job.pause_count, ==, 0);
 850     g_assert_false(job->job.paused);
 851     g_assert_true(job->job.busy); /* We're in job_sleep_ns() */
 852
 853     ret = job_complete_sync(&job->job, &error_abort);
 854     g_assert_cmpint(ret, ==, 0);
 855
 856     blk_unref(blk_src);
 857     blk_unref(blk_target);
 858     bdrv_unref(src);
 859     bdrv_unref(target);
 860 }
 861
 862 static void test_blockjob_drain_all(void)
 863 {
 864     test_blockjob_common(BDRV_DRAIN_ALL);
 865 }
 866
 867 static void test_blockjob_drain(void)
 868 {
 869     test_blockjob_common(BDRV_DRAIN);
 870 }
 871
 872 static void test_blockjob_drain_subtree(void)
 873 {
 874     test_blockjob_common(BDRV_SUBTREE_DRAIN);
 875 }
 876
 877
 878 typedef struct BDRVTestTopState {
 879     BdrvChild *wait_child;
 880 } BDRVTestTopState;
 881
 882 static void bdrv_test_top_close(BlockDriverState *bs)
 883 {
 884     BdrvChild *c, *next_c;
 885     QLIST_FOREACH_SAFE(c, &bs->children, next, next_c) {
 886         bdrv_unref_child(bs, c);
 887     }
 888 }
 889
 890 static int coroutine_fn bdrv_test_top_co_preadv(BlockDriverState *bs,
 891                                                 uint64_t offset, uint64_t bytes,
 892                                                 QEMUIOVector *qiov, int flags)
 893 {
 894     BDRVTestTopState *tts = bs->opaque;
 895     return bdrv_co_preadv(tts->wait_child, offset, bytes, qiov, flags);
 896 }
 897
 898 static BlockDriver bdrv_test_top_driver = {
 899     .format_name            = "test_top_driver",
 900     .instance_size          = sizeof(BDRVTestTopState),
 901
 902     .bdrv_close             = bdrv_test_top_close,
 903     .bdrv_co_preadv         = bdrv_test_top_co_preadv,
 904
 905     .bdrv_child_perm        = bdrv_format_default_perms,
 906 };
 907
 908 typedef struct TestCoDeleteByDrainData {
 909     BlockBackend *blk;
 910     bool detach_instead_of_delete;
 911     bool done;
 912 } TestCoDeleteByDrainData;
 913
 914 static void coroutine_fn test_co_delete_by_drain(void *opaque)
 915 {
 916     TestCoDeleteByDrainData *dbdd = opaque;
 917     BlockBackend *blk = dbdd->blk;
 918     BlockDriverState *bs = blk_bs(blk);
 919     BDRVTestTopState *tts = bs->opaque;
 920     void *buffer = g_malloc(65536);
 921     QEMUIOVector qiov;
 922     struct iovec iov = {
 923         .iov_base = buffer,
 924         .iov_len  = 65536,
 925     };
 926
 927     qemu_iovec_init_external(&qiov, &iov, 1);
 928
 929     /* Pretend some internal write operation from parent to child.
 930      * Important: We have to read from the child, not from the parent!
 931      * Draining works by first propagating it all up the tree to the
 932      * root and then waiting for drainage from root to the leaves
 933      * (protocol nodes).  If we have a request waiting on the root,
 934      * everything will be drained before we go back down the tree, but
 935      * we do not want that.  We want to be in the middle of draining
 936      * when this following requests returns. */
 937     bdrv_co_preadv(tts->wait_child, 0, 65536, &qiov, 0);
 938
 939     g_assert_cmpint(bs->refcnt, ==, 1);
 940
 941     if (!dbdd->detach_instead_of_delete) {
 942         blk_unref(blk);
 943     } else {
 944         BdrvChild *c, *next_c;
 945         QLIST_FOREACH_SAFE(c, &bs->children, next, next_c) {
 946             bdrv_unref_child(bs, c);
 947         }
 948     }
 949
 950     dbdd->done = true;
 951 }
 952
 953 /**
 954  * Test what happens when some BDS has some children, you drain one of
 955  * them and this results in the BDS being deleted.
 956  *
 957  * If @detach_instead_of_delete is set, the BDS is not going to be
 958  * deleted but will only detach all of its children.
 959  */
 960 static void do_test_delete_by_drain(bool detach_instead_of_delete,
 961                                     enum drain_type drain_type)
 962 {
 963     BlockBackend *blk;
 964     BlockDriverState *bs, *child_bs, *null_bs;
 965     BDRVTestTopState *tts;
 966     TestCoDeleteByDrainData dbdd;
 967     Coroutine *co;
 968
 969     bs = bdrv_new_open_driver(&bdrv_test_top_driver, "top", BDRV_O_RDWR,
 970                               &error_abort);
 971     bs->total_sectors = 65536 >> BDRV_SECTOR_BITS;
 972     tts = bs->opaque;
 973
 974     null_bs = bdrv_open("null-co://", NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
 975                         &error_abort);
 976     bdrv_attach_child(bs, null_bs, "null-child", &child_file, &error_abort);
 977
 978     /* This child will be the one to pass to requests through to, and
 979      * it will stall until a drain occurs */
 980     child_bs = bdrv_new_open_driver(&bdrv_test, "child", BDRV_O_RDWR,
 981                                     &error_abort);
 982     child_bs->total_sectors = 65536 >> BDRV_SECTOR_BITS;
 983     /* Takes our reference to child_bs */
 984     tts->wait_child = bdrv_attach_child(bs, child_bs, "wait-child", &child_file,
 985                                         &error_abort);
 986
 987     /* This child is just there to be deleted
 988      * (for detach_instead_of_delete == true) */
 989     null_bs = bdrv_open("null-co://", NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
 990                         &error_abort);
 991     bdrv_attach_child(bs, null_bs, "null-child", &child_file, &error_abort);
 992
 993     blk = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
 994     blk_insert_bs(blk, bs, &error_abort);
 995
 996     /* Referenced by blk now */
 997     bdrv_unref(bs);
 998
 999     g_assert_cmpint(bs->refcnt, ==, 1);
1000     g_assert_cmpint(child_bs->refcnt, ==, 1);
1001     g_assert_cmpint(null_bs->refcnt, ==, 1);
1002
1003
1004     dbdd = (TestCoDeleteByDrainData){
1005         .blk = blk,
1006         .detach_instead_of_delete = detach_instead_of_delete,
1007         .done = false,
1008     };
1009     co = qemu_coroutine_create(test_co_delete_by_drain, &dbdd);
1010     qemu_coroutine_enter(co);
1011
1012     /* Drain the child while the read operation is still pending.
1013      * This should result in the operation finishing and
1014      * test_co_delete_by_drain() resuming.  Thus, @bs will be deleted
1015      * and the coroutine will exit while this drain operation is still
1016      * in progress. */
1017     switch (drain_type) {
1018     case BDRV_DRAIN:
1019         bdrv_ref(child_bs);
1020         bdrv_drain(child_bs);
1021         bdrv_unref(child_bs);
1022         break;
1023     case BDRV_SUBTREE_DRAIN:
1024         /* Would have to ref/unref bs here for !detach_instead_of_delete, but
1025          * then the whole test becomes pointless because the graph changes
1026          * don't occur during the drain any more. */
1027         assert(detach_instead_of_delete);
1028         bdrv_subtree_drained_begin(bs);
1029         bdrv_subtree_drained_end(bs);
1030         break;
1031     case BDRV_DRAIN_ALL:
1032         bdrv_drain_all_begin();
1033         bdrv_drain_all_end();
1034         break;
1035     default:
1036         g_assert_not_reached();
1037     }
1038
1039     while (!dbdd.done) {
1040         aio_poll(qemu_get_aio_context(), true);
1041     }
1042
1043     if (detach_instead_of_delete) {
1044         /* Here, the reference has not passed over to the coroutine,
1045          * so we have to delete the BB ourselves */
1046         blk_unref(blk);
1047     }
1048 }
1049
1050 static void test_delete_by_drain(void)
1051 {
1052     do_test_delete_by_drain(false, BDRV_DRAIN);
1053 }
1054
1055 static void test_detach_by_drain_all(void)
1056 {
1057     do_test_delete_by_drain(true, BDRV_DRAIN_ALL);
1058 }
1059
1060 static void test_detach_by_drain(void)
1061 {
1062     do_test_delete_by_drain(true, BDRV_DRAIN);
1063 }
1064
1065 static void test_detach_by_drain_subtree(void)
1066 {
1067     do_test_delete_by_drain(true, BDRV_SUBTREE_DRAIN);
1068 }
1069
1070
1071 struct detach_by_parent_data {
1072     BlockDriverState *parent_b;
1073     BdrvChild *child_b;
1074     BlockDriverState *c;
1075     BdrvChild *child_c;
1076     bool by_parent_cb;
1077 };
1078 static struct detach_by_parent_data detach_by_parent_data;
1079
1080 static void detach_indirect_bh(void *opaque)
1081 {
1082     struct detach_by_parent_data *data = opaque;
1083
1084     bdrv_unref_child(data->parent_b, data->child_b);
1085
1086     bdrv_ref(data->c);
1087     data->child_c = bdrv_attach_child(data->parent_b, data->c, "PB-C",
1088                                       &child_file, &error_abort);
1089 }
1090
1091 static void detach_by_parent_aio_cb(void *opaque, int ret)
1092 {
1093     struct detach_by_parent_data *data = &detach_by_parent_data;
1094
1095     g_assert_cmpint(ret, ==, 0);
1096     if (data->by_parent_cb) {
1097         detach_indirect_bh(data);
1098     }
1099 }
1100
1101 static void detach_by_driver_cb_drained_begin(BdrvChild *child)
1102 {
1103     aio_bh_schedule_oneshot(qemu_get_current_aio_context(),
1104                             detach_indirect_bh, &detach_by_parent_data);
1105     child_file.drained_begin(child);
1106 }
1107
1108 static BdrvChildRole detach_by_driver_cb_role;
1109
1110 /*
1111  * Initial graph:
1112  *
1113  * PA     PB
1114  *    \ /   \
1115  *     A     B     C
1116  *
1117  * by_parent_cb == true:  Test that parent callbacks don't poll
1118  *
1119  *     PA has a pending write request whose callback changes the child nodes of
1120  *     PB: It removes B and adds C instead. The subtree of PB is drained, which
1121  *     will indirectly drain the write request, too.
1122  *
1123  * by_parent_cb == false: Test that bdrv_drain_invoke() doesn't poll
1124  *
1125  *     PA's BdrvChildRole has a .drained_begin callback that schedules a BH
1126  *     that does the same graph change. If bdrv_drain_invoke() calls it, the
1127  *     state is messed up, but if it is only polled in the single
1128  *     BDRV_POLL_WHILE() at the end of the drain, this should work fine.
1129  */
1130 static void test_detach_indirect(bool by_parent_cb)
1131 {
1132     BlockBackend *blk;
1133     BlockDriverState *parent_a, *parent_b, *a, *b, *c;
1134     BdrvChild *child_a, *child_b;
1135     BlockAIOCB *acb;
1136
1137     QEMUIOVector qiov;
1138     struct iovec iov = {
1139         .iov_base = NULL,
1140         .iov_len = 0,
1141     };
1142     qemu_iovec_init_external(&qiov, &iov, 1);
1143
1144     if (!by_parent_cb) {
1145         detach_by_driver_cb_role = child_file;
1146         detach_by_driver_cb_role.drained_begin =
1147             detach_by_driver_cb_drained_begin;
1148     }
1149
1150     /* Create all involved nodes */
1151     parent_a = bdrv_new_open_driver(&bdrv_test, "parent-a", BDRV_O_RDWR,
1152                                     &error_abort);
1153     parent_b = bdrv_new_open_driver(&bdrv_test, "parent-b", 0,
1154                                     &error_abort);
1155
1156     a = bdrv_new_open_driver(&bdrv_test, "a", BDRV_O_RDWR, &error_abort);
1157     b = bdrv_new_open_driver(&bdrv_test, "b", BDRV_O_RDWR, &error_abort);
1158     c = bdrv_new_open_driver(&bdrv_test, "c", BDRV_O_RDWR, &error_abort);
1159
1160     /* blk is a BB for parent-a */
1161     blk = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
1162     blk_insert_bs(blk, parent_a, &error_abort);
1163     bdrv_unref(parent_a);
1164
1165     /* If we want to get bdrv_drain_invoke() to call aio_poll(), the driver
1166      * callback must not return immediately. */
1167     if (!by_parent_cb) {
1168         BDRVTestState *s = parent_a->opaque;
1169         s->sleep_in_drain_begin = true;
1170     }
1171
1172     /* Set child relationships */
1173     bdrv_ref(b);
1174     bdrv_ref(a);
1175     child_b = bdrv_attach_child(parent_b, b, "PB-B", &child_file, &error_abort);
1176     child_a = bdrv_attach_child(parent_b, a, "PB-A", &child_backing, &error_abort);
1177
1178     bdrv_ref(a);
1179     bdrv_attach_child(parent_a, a, "PA-A",
1180                       by_parent_cb ? &child_file : &detach_by_driver_cb_role,
1181                       &error_abort);
1182
1183     g_assert_cmpint(parent_a->refcnt, ==, 1);
1184     g_assert_cmpint(parent_b->refcnt, ==, 1);
1185     g_assert_cmpint(a->refcnt, ==, 3);
1186     g_assert_cmpint(b->refcnt, ==, 2);
1187     g_assert_cmpint(c->refcnt, ==, 1);
1188
1189     g_assert(QLIST_FIRST(&parent_b->children) == child_a);
1190     g_assert(QLIST_NEXT(child_a, next) == child_b);
1191     g_assert(QLIST_NEXT(child_b, next) == NULL);
1192
1193     /* Start the evil write request */
1194     detach_by_parent_data = (struct detach_by_parent_data) {
1195         .parent_b = parent_b,
1196         .child_b = child_b,
1197         .c = c,
1198         .by_parent_cb = by_parent_cb,
1199     };
1200     acb = blk_aio_preadv(blk, 0, &qiov, 0, detach_by_parent_aio_cb, NULL);
1201     g_assert(acb != NULL);
1202
1203     /* Drain and check the expected result */
1204     bdrv_subtree_drained_begin(parent_b);
1205
1206     g_assert(detach_by_parent_data.child_c != NULL);
1207
1208     g_assert_cmpint(parent_a->refcnt, ==, 1);
1209     g_assert_cmpint(parent_b->refcnt, ==, 1);
1210     g_assert_cmpint(a->refcnt, ==, 3);
1211     g_assert_cmpint(b->refcnt, ==, 1);
1212     g_assert_cmpint(c->refcnt, ==, 2);
1213
1214     g_assert(QLIST_FIRST(&parent_b->children) == detach_by_parent_data.child_c);
1215     g_assert(QLIST_NEXT(detach_by_parent_data.child_c, next) == child_a);
1216     g_assert(QLIST_NEXT(child_a, next) == NULL);
1217
1218     g_assert_cmpint(parent_a->quiesce_counter, ==, 1);
1219     g_assert_cmpint(parent_b->quiesce_counter, ==, 1);
1220     g_assert_cmpint(a->quiesce_counter, ==, 1);
1221     g_assert_cmpint(b->quiesce_counter, ==, 0);
1222     g_assert_cmpint(c->quiesce_counter, ==, 1);
1223
1224     bdrv_subtree_drained_end(parent_b);
1225
1226     bdrv_unref(parent_b);
1227     blk_unref(blk);
1228
1229     /* XXX Once bdrv_close() unref's children instead of just detaching them,
1230      * this won't be necessary any more. */
1231     bdrv_unref(a);
1232     bdrv_unref(a);
1233     bdrv_unref(c);
1234
1235     g_assert_cmpint(a->refcnt, ==, 1);
1236     g_assert_cmpint(b->refcnt, ==, 1);
1237     g_assert_cmpint(c->refcnt, ==, 1);
1238     bdrv_unref(a);
1239     bdrv_unref(b);
1240     bdrv_unref(c);
1241 }
1242
1243 static void test_detach_by_parent_cb(void)
1244 {
1245     test_detach_indirect(true);
1246 }
1247
1248 static void test_detach_by_driver_cb(void)
1249 {
1250     test_detach_indirect(false);
1251 }
1252
1253 static void test_append_to_drained(void)
1254 {
1255     BlockBackend *blk;
1256     BlockDriverState *base, *overlay;
1257     BDRVTestState *base_s, *overlay_s;
1258
1259     blk = blk_new(BLK_PERM_ALL, BLK_PERM_ALL);
1260     base = bdrv_new_open_driver(&bdrv_test, "base", BDRV_O_RDWR, &error_abort);
1261     base_s = base->opaque;
1262     blk_insert_bs(blk, base, &error_abort);
1263
1264     overlay = bdrv_new_open_driver(&bdrv_test, "overlay", BDRV_O_RDWR,
1265                                    &error_abort);
1266     overlay_s = overlay->opaque;
1267
1268     do_drain_begin(BDRV_DRAIN, base);
1269     g_assert_cmpint(base->quiesce_counter, ==, 1);
1270     g_assert_cmpint(base_s->drain_count, ==, 1);
1271     g_assert_cmpint(base->in_flight, ==, 0);
1272
1273     /* Takes ownership of overlay, so we don't have to unref it later */
1274     bdrv_append(overlay, base, &error_abort);
1275     g_assert_cmpint(base->in_flight, ==, 0);
1276     g_assert_cmpint(overlay->in_flight, ==, 0);
1277
1278     g_assert_cmpint(base->quiesce_counter, ==, 1);
1279     g_assert_cmpint(base_s->drain_count, ==, 1);
1280     g_assert_cmpint(overlay->quiesce_counter, ==, 1);
1281     g_assert_cmpint(overlay_s->drain_count, ==, 1);
1282
1283     do_drain_end(BDRV_DRAIN, base);
1284
1285     g_assert_cmpint(base->quiesce_counter, ==, 0);
1286     g_assert_cmpint(base_s->drain_count, ==, 0);
1287     g_assert_cmpint(overlay->quiesce_counter, ==, 0);
1288     g_assert_cmpint(overlay_s->drain_count, ==, 0);
1289
1290     bdrv_unref(base);
1291     blk_unref(blk);
1292 }
1293
1294 int main(int argc, char **argv)
1295 {
1296     int ret;
1297
1298     bdrv_init();
1299     qemu_init_main_loop(&error_abort);
1300
1301     g_test_init(&argc, &argv, NULL);
1302     qemu_event_init(&done_event, false);
1303
1304     g_test_add_func("/bdrv-drain/driver-cb/drain_all", test_drv_cb_drain_all);
1305     g_test_add_func("/bdrv-drain/driver-cb/drain", test_drv_cb_drain);
1306     g_test_add_func("/bdrv-drain/driver-cb/drain_subtree",
1307                     test_drv_cb_drain_subtree);
1308
1309     g_test_add_func("/bdrv-drain/driver-cb/co/drain_all",
1310                     test_drv_cb_co_drain_all);
1311     g_test_add_func("/bdrv-drain/driver-cb/co/drain", test_drv_cb_co_drain);
1312     g_test_add_func("/bdrv-drain/driver-cb/co/drain_subtree",
1313                     test_drv_cb_co_drain_subtree);
1314
1315
1316     g_test_add_func("/bdrv-drain/quiesce/drain_all", test_quiesce_drain_all);
1317     g_test_add_func("/bdrv-drain/quiesce/drain", test_quiesce_drain);
1318     g_test_add_func("/bdrv-drain/quiesce/drain_subtree",
1319                     test_quiesce_drain_subtree);
1320
1321     g_test_add_func("/bdrv-drain/quiesce/co/drain_all",
1322                     test_quiesce_co_drain_all);
1323     g_test_add_func("/bdrv-drain/quiesce/co/drain", test_quiesce_co_drain);
1324     g_test_add_func("/bdrv-drain/quiesce/co/drain_subtree",
1325                     test_quiesce_co_drain_subtree);
1326
1327     g_test_add_func("/bdrv-drain/nested", test_nested);
1328     g_test_add_func("/bdrv-drain/multiparent", test_multiparent);
1329
1330     g_test_add_func("/bdrv-drain/graph-change/drain_subtree",
1331                     test_graph_change_drain_subtree);
1332     g_test_add_func("/bdrv-drain/graph-change/drain_all",
1333                     test_graph_change_drain_all);
1334
1335     g_test_add_func("/bdrv-drain/iothread/drain_all", test_iothread_drain_all);
1336     g_test_add_func("/bdrv-drain/iothread/drain", test_iothread_drain);
1337     g_test_add_func("/bdrv-drain/iothread/drain_subtree",
1338                     test_iothread_drain_subtree);
1339
1340     g_test_add_func("/bdrv-drain/blockjob/drain_all", test_blockjob_drain_all);
1341     g_test_add_func("/bdrv-drain/blockjob/drain", test_blockjob_drain);
1342     g_test_add_func("/bdrv-drain/blockjob/drain_subtree",
1343                     test_blockjob_drain_subtree);
1344
1345     g_test_add_func("/bdrv-drain/deletion/drain", test_delete_by_drain);
1346     g_test_add_func("/bdrv-drain/detach/drain_all", test_detach_by_drain_all);
1347     g_test_add_func("/bdrv-drain/detach/drain", test_detach_by_drain);
1348     g_test_add_func("/bdrv-drain/detach/drain_subtree", test_detach_by_drain_subtree);
1349     g_test_add_func("/bdrv-drain/detach/parent_cb", test_detach_by_parent_cb);
1350     g_test_add_func("/bdrv-drain/detach/driver_cb", test_detach_by_driver_cb);
1351
1352     g_test_add_func("/bdrv-drain/attach/drain", test_append_to_drained);
1353
1354     ret = g_test_run();
1355     qemu_event_destroy(&done_event);
1356     return ret;
1357 }