tests/migration-test.c

   1 /*
   2  * QTest testcase for migration
   3  *
   4  * Copyright (c) 2016-2018 Red Hat, Inc. and/or its affiliates
   5  *   based on the vhost-user-test.c that is:
   6  *      Copyright (c) 2014 Virtual Open Systems Sarl.
   7  *
   8  * This work is licensed under the terms of the GNU GPL, version 2 or later.
   9  * See the COPYING file in the top-level directory.
  10  *
  11  */
  12
  13 #include "qemu/osdep.h"
  14
  15 #include "libqtest.h"
  16 #include "qapi/qmp/qdict.h"
  17 #include "qemu/option.h"
  18 #include "qemu/range.h"
  19 #include "qemu/sockets.h"
  20 #include "chardev/char.h"
  21 #include "sysemu/sysemu.h"
  22
  23 const unsigned start_address = 1024 * 1024;
  24 const unsigned end_address = 100 * 1024 * 1024;
  25 bool got_stop;
  26 static bool uffd_feature_thread_id;
  27
  28 #if defined(__linux__)
  29 #include <sys/syscall.h>
  30 #include <sys/vfs.h>
  31 #endif
  32
  33 #if defined(__linux__) && defined(__NR_userfaultfd) && defined(CONFIG_EVENTFD)
  34 #include <sys/eventfd.h>
  35 #include <sys/ioctl.h>
  36 #include <linux/userfaultfd.h>
  37
  38 static bool ufd_version_check(void)
  39 {
  40     struct uffdio_api api_struct;
  41     uint64_t ioctl_mask;
  42
  43     int ufd = syscall(__NR_userfaultfd, O_CLOEXEC);
  44
  45     if (ufd == -1) {
  46         g_test_message("Skipping test: userfaultfd not available");
  47         return false;
  48     }
  49
  50     api_struct.api = UFFD_API;
  51     api_struct.features = 0;
  52     if (ioctl(ufd, UFFDIO_API, &api_struct)) {
  53         g_test_message("Skipping test: UFFDIO_API failed");
  54         return false;
  55     }
  56     uffd_feature_thread_id = api_struct.features & UFFD_FEATURE_THREAD_ID;
  57
  58     ioctl_mask = (__u64)1 << _UFFDIO_REGISTER |
  59                  (__u64)1 << _UFFDIO_UNREGISTER;
  60     if ((api_struct.ioctls & ioctl_mask) != ioctl_mask) {
  61         g_test_message("Skipping test: Missing userfault feature");
  62         return false;
  63     }
  64
  65     return true;
  66 }
  67
  68 #else
  69 static bool ufd_version_check(void)
  70 {
  71     g_test_message("Skipping test: Userfault not available (builtdtime)");
  72     return false;
  73 }
  74
  75 #endif
  76
  77 static const char *tmpfs;
  78
  79 /* A simple PC boot sector that modifies memory (1-100MB) quickly
  80  * outputting a 'B' every so often if it's still running.
  81  */
  82 #include "tests/migration/x86-a-b-bootblock.h"
  83
  84 static void init_bootfile_x86(const char *bootpath)
  85 {
  86     FILE *bootfile = fopen(bootpath, "wb");
  87
  88     g_assert_cmpint(fwrite(x86_bootsect, 512, 1, bootfile), ==, 1);
  89     fclose(bootfile);
  90 }
  91
  92 /*
  93  * Wait for some output in the serial output file,
  94  * we get an 'A' followed by an endless string of 'B's
  95  * but on the destination we won't have the A.
  96  */
  97 static void wait_for_serial(const char *side)
  98 {
  99     char *serialpath = g_strdup_printf("%s/%s", tmpfs, side);
 100     FILE *serialfile = fopen(serialpath, "r");
 101     const char *arch = qtest_get_arch();
 102     int started = (strcmp(side, "src_serial") == 0 &&
 103                    strcmp(arch, "ppc64") == 0) ? 0 : 1;
 104
 105     g_free(serialpath);
 106     do {
 107         int readvalue = fgetc(serialfile);
 108
 109         if (!started) {
 110             /* SLOF prints its banner before starting test,
 111              * to ignore it, mark the start of the test with '_',
 112              * ignore all characters until this marker
 113              */
 114             switch (readvalue) {
 115             case '_':
 116                 started = 1;
 117                 break;
 118             case EOF:
 119                 fseek(serialfile, 0, SEEK_SET);
 120                 usleep(1000);
 121                 break;
 122             }
 123             continue;
 124         }
 125         switch (readvalue) {
 126         case 'A':
 127             /* Fine */
 128             break;
 129
 130         case 'B':
 131             /* It's alive! */
 132             fclose(serialfile);
 133             return;
 134
 135         case EOF:
 136             started = (strcmp(side, "src_serial") == 0 &&
 137                        strcmp(arch, "ppc64") == 0) ? 0 : 1;
 138             fseek(serialfile, 0, SEEK_SET);
 139             usleep(1000);
 140             break;
 141
 142         default:
 143             fprintf(stderr, "Unexpected %d on %s serial\n", readvalue, side);
 144             g_assert_not_reached();
 145         }
 146     } while (true);
 147 }
 148
 149 /*
 150  * Events can get in the way of responses we are actually waiting for.
 151  */
 152 static QDict *wait_command(QTestState *who, const char *command)
 153 {
 154     const char *event_string;
 155     QDict *response;
 156
 157     response = qtest_qmp(who, command);
 158
 159     while (qdict_haskey(response, "event")) {
 160         /* OK, it was an event */
 161         event_string = qdict_get_str(response, "event");
 162         if (!strcmp(event_string, "STOP")) {
 163             got_stop = true;
 164         }
 165         qobject_unref(response);
 166         response = qtest_qmp_receive(who);
 167     }
 168     return response;
 169 }
 170
 171
 172 /*
 173  * It's tricky to use qemu's migration event capability with qtest,
 174  * events suddenly appearing confuse the qmp()/hmp() responses.
 175  */
 176
 177 static uint64_t get_migration_pass(QTestState *who)
 178 {
 179     QDict *rsp, *rsp_return, *rsp_ram;
 180     uint64_t result;
 181
 182     rsp = wait_command(who, "{ 'execute': 'query-migrate' }");
 183     rsp_return = qdict_get_qdict(rsp, "return");
 184     if (!qdict_haskey(rsp_return, "ram")) {
 185         /* Still in setup */
 186         result = 0;
 187     } else {
 188         rsp_ram = qdict_get_qdict(rsp_return, "ram");
 189         result = qdict_get_try_int(rsp_ram, "dirty-sync-count", 0);
 190     }
 191     qobject_unref(rsp);
 192     return result;
 193 }
 194
 195 static void read_blocktime(QTestState *who)
 196 {
 197     QDict *rsp, *rsp_return;
 198
 199     rsp = wait_command(who, "{ 'execute': 'query-migrate' }");
 200     rsp_return = qdict_get_qdict(rsp, "return");
 201     g_assert(qdict_haskey(rsp_return, "postcopy-blocktime"));
 202     qobject_unref(rsp);
 203 }
 204
 205 static void wait_for_migration_complete(QTestState *who)
 206 {
 207     while (true) {
 208         QDict *rsp, *rsp_return;
 209         bool completed;
 210         const char *status;
 211
 212         rsp = wait_command(who, "{ 'execute': 'query-migrate' }");
 213         rsp_return = qdict_get_qdict(rsp, "return");
 214         status = qdict_get_str(rsp_return, "status");
 215         completed = strcmp(status, "completed") == 0;
 216         g_assert_cmpstr(status, !=,  "failed");
 217         qobject_unref(rsp);
 218         if (completed) {
 219             return;
 220         }
 221         usleep(1000);
 222     }
 223 }
 224
 225 static void wait_for_migration_pass(QTestState *who)
 226 {
 227     uint64_t initial_pass = get_migration_pass(who);
 228     uint64_t pass;
 229
 230     /* Wait for the 1st sync */
 231     while (!got_stop && !initial_pass) {
 232         usleep(1000);
 233         initial_pass = get_migration_pass(who);
 234     }
 235
 236     do {
 237         usleep(1000);
 238         pass = get_migration_pass(who);
 239     } while (pass == initial_pass && !got_stop);
 240 }
 241
 242 static void check_guests_ram(QTestState *who)
 243 {
 244     /* Our ASM test will have been incrementing one byte from each page from
 245      * 1MB to <100MB in order.
 246      * This gives us a constraint that any page's byte should be equal or less
 247      * than the previous pages byte (mod 256); and they should all be equal
 248      * except for one transition at the point where we meet the incrementer.
 249      * (We're running this with the guest stopped).
 250      */
 251     unsigned address;
 252     uint8_t first_byte;
 253     uint8_t last_byte;
 254     bool hit_edge = false;
 255     bool bad = false;
 256
 257     qtest_memread(who, start_address, &first_byte, 1);
 258     last_byte = first_byte;
 259
 260     for (address = start_address + 4096; address < end_address; address += 4096)
 261     {
 262         uint8_t b;
 263         qtest_memread(who, address, &b, 1);
 264         if (b != last_byte) {
 265             if (((b + 1) % 256) == last_byte && !hit_edge) {
 266                 /* This is OK, the guest stopped at the point of
 267                  * incrementing the previous page but didn't get
 268                  * to us yet.
 269                  */
 270                 hit_edge = true;
 271             } else {
 272                 fprintf(stderr, "Memory content inconsistency at %x"
 273                                 " first_byte = %x last_byte = %x current = %x"
 274                                 " hit_edge = %x\n",
 275                                 address, first_byte, last_byte, b, hit_edge);
 276                 bad = true;
 277             }
 278         }
 279         last_byte = b;
 280     }
 281     g_assert_false(bad);
 282 }
 283
 284 static void cleanup(const char *filename)
 285 {
 286     char *path = g_strdup_printf("%s/%s", tmpfs, filename);
 287
 288     unlink(path);
 289     g_free(path);
 290 }
 291
 292 static void migrate_check_parameter(QTestState *who, const char *parameter,
 293                                     const char *value)
 294 {
 295     QDict *rsp, *rsp_return;
 296     char *result;
 297
 298     rsp = wait_command(who, "{ 'execute': 'query-migrate-parameters' }");
 299     rsp_return = qdict_get_qdict(rsp, "return");
 300     result = g_strdup_printf("%" PRId64,
 301                              qdict_get_try_int(rsp_return,  parameter, -1));
 302     g_assert_cmpstr(result, ==, value);
 303     g_free(result);
 304     qobject_unref(rsp);
 305 }
 306
 307 static void migrate_set_parameter(QTestState *who, const char *parameter,
 308                                   const char *value)
 309 {
 310     QDict *rsp;
 311     gchar *cmd;
 312
 313     cmd = g_strdup_printf("{ 'execute': 'migrate-set-parameters',"
 314                           "'arguments': { '%s': %s } }",
 315                           parameter, value);
 316     rsp = qtest_qmp(who, cmd);
 317     g_free(cmd);
 318     g_assert(qdict_haskey(rsp, "return"));
 319     qobject_unref(rsp);
 320     migrate_check_parameter(who, parameter, value);
 321 }
 322
 323 static void migrate_set_capability(QTestState *who, const char *capability,
 324                                    const char *value)
 325 {
 326     QDict *rsp;
 327     gchar *cmd;
 328
 329     cmd = g_strdup_printf("{ 'execute': 'migrate-set-capabilities',"
 330                           "'arguments': { "
 331                           "'capabilities': [ { "
 332                           "'capability': '%s', 'state': %s } ] } }",
 333                           capability, value);
 334     rsp = qtest_qmp(who, cmd);
 335     g_free(cmd);
 336     g_assert(qdict_haskey(rsp, "return"));
 337     qobject_unref(rsp);
 338 }
 339
 340 static void migrate(QTestState *who, const char *uri)
 341 {
 342     QDict *rsp;
 343     gchar *cmd;
 344
 345     cmd = g_strdup_printf("{ 'execute': 'migrate',"
 346                           "'arguments': { 'uri': '%s' } }",
 347                           uri);
 348     rsp = qtest_qmp(who, cmd);
 349     g_free(cmd);
 350     g_assert(qdict_haskey(rsp, "return"));
 351     qobject_unref(rsp);
 352 }
 353
 354 static void migrate_start_postcopy(QTestState *who)
 355 {
 356     QDict *rsp;
 357
 358     rsp = wait_command(who, "{ 'execute': 'migrate-start-postcopy' }");
 359     g_assert(qdict_haskey(rsp, "return"));
 360     qobject_unref(rsp);
 361 }
 362
 363 static void test_migrate_start(QTestState **from, QTestState **to,
 364                                const char *uri, bool hide_stderr)
 365 {
 366     gchar *cmd_src, *cmd_dst;
 367     char *bootpath = g_strdup_printf("%s/bootsect", tmpfs);
 368     const char *arch = qtest_get_arch();
 369     const char *accel = "kvm:tcg";
 370
 371     got_stop = false;
 372
 373     if (strcmp(arch, "i386") == 0 || strcmp(arch, "x86_64") == 0) {
 374         init_bootfile_x86(bootpath);
 375         cmd_src = g_strdup_printf("-machine accel=%s -m 150M"
 376                                   " -name source,debug-threads=on"
 377                                   " -serial file:%s/src_serial"
 378                                   " -drive file=%s,format=raw",
 379                                   accel, tmpfs, bootpath);
 380         cmd_dst = g_strdup_printf("-machine accel=%s -m 150M"
 381                                   " -name target,debug-threads=on"
 382                                   " -serial file:%s/dest_serial"
 383                                   " -drive file=%s,format=raw"
 384                                   " -incoming %s",
 385                                   accel, tmpfs, bootpath, uri);
 386     } else if (strcmp(arch, "ppc64") == 0) {
 387
 388         /* On ppc64, the test only works with kvm-hv, but not with kvm-pr */
 389         if (access("/sys/module/kvm_hv", F_OK)) {
 390             accel = "tcg";
 391         }
 392         cmd_src = g_strdup_printf("-machine accel=%s -m 256M"
 393                                   " -name source,debug-threads=on"
 394                                   " -serial file:%s/src_serial"
 395                                   " -prom-env '"
 396                                   "boot-command=hex .\" _\" begin %x %x "
 397                                   "do i c@ 1 + i c! 1000 +loop .\" B\" 0 "
 398                                   "until'",  accel, tmpfs, end_address,
 399                                   start_address);
 400         cmd_dst = g_strdup_printf("-machine accel=%s -m 256M"
 401                                   " -name target,debug-threads=on"
 402                                   " -serial file:%s/dest_serial"
 403                                   " -incoming %s",
 404                                   accel, tmpfs, uri);
 405     } else {
 406         g_assert_not_reached();
 407     }
 408
 409     g_free(bootpath);
 410
 411     if (hide_stderr) {
 412         gchar *tmp;
 413         tmp = g_strdup_printf("%s 2>/dev/null", cmd_src);
 414         g_free(cmd_src);
 415         cmd_src = tmp;
 416
 417         tmp = g_strdup_printf("%s 2>/dev/null", cmd_dst);
 418         g_free(cmd_dst);
 419         cmd_dst = tmp;
 420     }
 421
 422     *from = qtest_start(cmd_src);
 423     g_free(cmd_src);
 424
 425     *to = qtest_init(cmd_dst);
 426     g_free(cmd_dst);
 427 }
 428
 429 static void test_migrate_end(QTestState *from, QTestState *to, bool test_dest)
 430 {
 431     unsigned char dest_byte_a, dest_byte_b, dest_byte_c, dest_byte_d;
 432
 433     qtest_quit(from);
 434
 435     if (test_dest) {
 436         qtest_memread(to, start_address, &dest_byte_a, 1);
 437
 438         /* Destination still running, wait for a byte to change */
 439         do {
 440             qtest_memread(to, start_address, &dest_byte_b, 1);
 441             usleep(1000 * 10);
 442         } while (dest_byte_a == dest_byte_b);
 443
 444         qtest_qmp_discard_response(to, "{ 'execute' : 'stop'}");
 445
 446         /* With it stopped, check nothing changes */
 447         qtest_memread(to, start_address, &dest_byte_c, 1);
 448         usleep(1000 * 200);
 449         qtest_memread(to, start_address, &dest_byte_d, 1);
 450         g_assert_cmpint(dest_byte_c, ==, dest_byte_d);
 451
 452         check_guests_ram(to);
 453     }
 454
 455     qtest_quit(to);
 456
 457     cleanup("bootsect");
 458     cleanup("migsocket");
 459     cleanup("src_serial");
 460     cleanup("dest_serial");
 461 }
 462
 463 static void deprecated_set_downtime(QTestState *who, const double value)
 464 {
 465     QDict *rsp;
 466     gchar *cmd;
 467     char *expected;
 468     int64_t result_int;
 469
 470     cmd = g_strdup_printf("{ 'execute': 'migrate_set_downtime',"
 471                           "'arguments': { 'value': %g } }", value);
 472     rsp = qtest_qmp(who, cmd);
 473     g_free(cmd);
 474     g_assert(qdict_haskey(rsp, "return"));
 475     qobject_unref(rsp);
 476     result_int = value * 1000L;
 477     expected = g_strdup_printf("%" PRId64, result_int);
 478     migrate_check_parameter(who, "downtime-limit", expected);
 479     g_free(expected);
 480 }
 481
 482 static void deprecated_set_speed(QTestState *who, const char *value)
 483 {
 484     QDict *rsp;
 485     gchar *cmd;
 486
 487     cmd = g_strdup_printf("{ 'execute': 'migrate_set_speed',"
 488                           "'arguments': { 'value': %s } }", value);
 489     rsp = qtest_qmp(who, cmd);
 490     g_free(cmd);
 491     g_assert(qdict_haskey(rsp, "return"));
 492     qobject_unref(rsp);
 493     migrate_check_parameter(who, "max-bandwidth", value);
 494 }
 495
 496 static void test_deprecated(void)
 497 {
 498     QTestState *from;
 499
 500     from = qtest_start("");
 501
 502     deprecated_set_downtime(from, 0.12345);
 503     deprecated_set_speed(from, "12345");
 504
 505     qtest_quit(from);
 506 }
 507
 508 static void test_postcopy(void)
 509 {
 510     char *uri = g_strdup_printf("unix:%s/migsocket", tmpfs);
 511     QTestState *from, *to;
 512
 513     test_migrate_start(&from, &to, uri, false);
 514
 515     migrate_set_capability(from, "postcopy-ram", "true");
 516     migrate_set_capability(to, "postcopy-ram", "true");
 517     migrate_set_capability(to, "postcopy-blocktime", "true");
 518
 519     /* We want to pick a speed slow enough that the test completes
 520      * quickly, but that it doesn't complete precopy even on a slow
 521      * machine, so also set the downtime.
 522      */
 523     migrate_set_parameter(from, "max-bandwidth", "100000000");
 524     migrate_set_parameter(from, "downtime-limit", "1");
 525
 526     /* Wait for the first serial output from the source */
 527     wait_for_serial("src_serial");
 528
 529     migrate(from, uri);
 530
 531     wait_for_migration_pass(from);
 532
 533     migrate_start_postcopy(from);
 534
 535     if (!got_stop) {
 536         qtest_qmp_eventwait(from, "STOP");
 537     }
 538
 539     qtest_qmp_eventwait(to, "RESUME");
 540
 541     wait_for_serial("dest_serial");
 542     wait_for_migration_complete(from);
 543
 544     if (uffd_feature_thread_id) {
 545         read_blocktime(to);
 546     }
 547     g_free(uri);
 548
 549     test_migrate_end(from, to, true);
 550 }
 551
 552 static void test_baddest(void)
 553 {
 554     QTestState *from, *to;
 555     QDict *rsp, *rsp_return;
 556     const char *status;
 557     bool failed;
 558
 559     test_migrate_start(&from, &to, "tcp:0:0", true);
 560     migrate(from, "tcp:0:0");
 561     do {
 562         rsp = wait_command(from, "{ 'execute': 'query-migrate' }");
 563         rsp_return = qdict_get_qdict(rsp, "return");
 564
 565         status = qdict_get_str(rsp_return, "status");
 566
 567         g_assert(!strcmp(status, "setup") || !(strcmp(status, "failed")));
 568         failed = !strcmp(status, "failed");
 569         qobject_unref(rsp);
 570     } while (!failed);
 571
 572     /* Is the machine currently running? */
 573     rsp = wait_command(from, "{ 'execute': 'query-status' }");
 574     g_assert(qdict_haskey(rsp, "return"));
 575     rsp_return = qdict_get_qdict(rsp, "return");
 576     g_assert(qdict_haskey(rsp_return, "running"));
 577     g_assert(qdict_get_bool(rsp_return, "running"));
 578     qobject_unref(rsp);
 579
 580     test_migrate_end(from, to, false);
 581 }
 582
 583 static void test_precopy_unix(void)
 584 {
 585     char *uri = g_strdup_printf("unix:%s/migsocket", tmpfs);
 586     QTestState *from, *to;
 587
 588     test_migrate_start(&from, &to, uri, false);
 589
 590     /* We want to pick a speed slow enough that the test completes
 591      * quickly, but that it doesn't complete precopy even on a slow
 592      * machine, so also set the downtime.
 593      */
 594     /* 1 ms should make it not converge*/
 595     migrate_set_parameter(from, "downtime-limit", "1");
 596     /* 1GB/s */
 597     migrate_set_parameter(from, "max-bandwidth", "1000000000");
 598
 599     /* Wait for the first serial output from the source */
 600     wait_for_serial("src_serial");
 601
 602     migrate(from, uri);
 603
 604     wait_for_migration_pass(from);
 605
 606     /* 300 ms should converge */
 607     migrate_set_parameter(from, "downtime-limit", "300");
 608
 609     if (!got_stop) {
 610         qtest_qmp_eventwait(from, "STOP");
 611     }
 612
 613     qtest_qmp_eventwait(to, "RESUME");
 614
 615     wait_for_serial("dest_serial");
 616     wait_for_migration_complete(from);
 617
 618     test_migrate_end(from, to, true);
 619     g_free(uri);
 620 }
 621
 622 int main(int argc, char **argv)
 623 {
 624     char template[] = "/tmp/migration-test-XXXXXX";
 625     int ret;
 626
 627     g_test_init(&argc, &argv, NULL);
 628
 629     if (!ufd_version_check()) {
 630         return 0;
 631     }
 632
 633     tmpfs = mkdtemp(template);
 634     if (!tmpfs) {
 635         g_test_message("mkdtemp on path (%s): %s\n", template, strerror(errno));
 636     }
 637     g_assert(tmpfs);
 638
 639     module_call_init(MODULE_INIT_QOM);
 640
 641     qtest_add_func("/migration/postcopy/unix", test_postcopy);
 642     qtest_add_func("/migration/deprecated", test_deprecated);
 643     qtest_add_func("/migration/bad_dest", test_baddest);
 644     qtest_add_func("/migration/precopy/unix", test_precopy_unix);
 645
 646     ret = g_test_run();
 647
 648     g_assert_cmpint(ret, ==, 0);
 649
 650     ret = rmdir(tmpfs);
 651     if (ret != 0) {
 652         g_test_message("unable to rmdir: path (%s): %s\n",
 653                        tmpfs, strerror(errno));
 654     }
 655
 656     return ret;
 657 }