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 int 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          * and TCG is touchy due to race conditions on dirty bits
 390          * (especially on PPC for some reason)
 391          */
 392         if (access("/sys/module/kvm_hv", F_OK)) {
 393             g_print("Skipping test: kvm_hv not available ");
 394             return -1;
 395         }
 396         cmd_src = g_strdup_printf("-machine accel=%s -m 256M"
 397                                   " -name source,debug-threads=on"
 398                                   " -serial file:%s/src_serial"
 399                                   " -prom-env '"
 400                                   "boot-command=hex .\" _\" begin %x %x "
 401                                   "do i c@ 1 + i c! 1000 +loop .\" B\" 0 "
 402                                   "until'",  accel, tmpfs, end_address,
 403                                   start_address);
 404         cmd_dst = g_strdup_printf("-machine accel=%s -m 256M"
 405                                   " -name target,debug-threads=on"
 406                                   " -serial file:%s/dest_serial"
 407                                   " -incoming %s",
 408                                   accel, tmpfs, uri);
 409     } else {
 410         g_assert_not_reached();
 411     }
 412
 413     g_free(bootpath);
 414
 415     if (hide_stderr) {
 416         gchar *tmp;
 417         tmp = g_strdup_printf("%s 2>/dev/null", cmd_src);
 418         g_free(cmd_src);
 419         cmd_src = tmp;
 420
 421         tmp = g_strdup_printf("%s 2>/dev/null", cmd_dst);
 422         g_free(cmd_dst);
 423         cmd_dst = tmp;
 424     }
 425
 426     *from = qtest_start(cmd_src);
 427     g_free(cmd_src);
 428
 429     *to = qtest_init(cmd_dst);
 430     g_free(cmd_dst);
 431     return 0;
 432 }
 433
 434 static void test_migrate_end(QTestState *from, QTestState *to, bool test_dest)
 435 {
 436     unsigned char dest_byte_a, dest_byte_b, dest_byte_c, dest_byte_d;
 437
 438     qtest_quit(from);
 439
 440     if (test_dest) {
 441         qtest_memread(to, start_address, &dest_byte_a, 1);
 442
 443         /* Destination still running, wait for a byte to change */
 444         do {
 445             qtest_memread(to, start_address, &dest_byte_b, 1);
 446             usleep(1000 * 10);
 447         } while (dest_byte_a == dest_byte_b);
 448
 449         qtest_qmp_discard_response(to, "{ 'execute' : 'stop'}");
 450
 451         /* With it stopped, check nothing changes */
 452         qtest_memread(to, start_address, &dest_byte_c, 1);
 453         usleep(1000 * 200);
 454         qtest_memread(to, start_address, &dest_byte_d, 1);
 455         g_assert_cmpint(dest_byte_c, ==, dest_byte_d);
 456
 457         check_guests_ram(to);
 458     }
 459
 460     qtest_quit(to);
 461
 462     cleanup("bootsect");
 463     cleanup("migsocket");
 464     cleanup("src_serial");
 465     cleanup("dest_serial");
 466 }
 467
 468 static void deprecated_set_downtime(QTestState *who, const double value)
 469 {
 470     QDict *rsp;
 471     gchar *cmd;
 472     char *expected;
 473     int64_t result_int;
 474
 475     cmd = g_strdup_printf("{ 'execute': 'migrate_set_downtime',"
 476                           "'arguments': { 'value': %g } }", value);
 477     rsp = qtest_qmp(who, cmd);
 478     g_free(cmd);
 479     g_assert(qdict_haskey(rsp, "return"));
 480     qobject_unref(rsp);
 481     result_int = value * 1000L;
 482     expected = g_strdup_printf("%" PRId64, result_int);
 483     migrate_check_parameter(who, "downtime-limit", expected);
 484     g_free(expected);
 485 }
 486
 487 static void deprecated_set_speed(QTestState *who, const char *value)
 488 {
 489     QDict *rsp;
 490     gchar *cmd;
 491
 492     cmd = g_strdup_printf("{ 'execute': 'migrate_set_speed',"
 493                           "'arguments': { 'value': %s } }", value);
 494     rsp = qtest_qmp(who, cmd);
 495     g_free(cmd);
 496     g_assert(qdict_haskey(rsp, "return"));
 497     qobject_unref(rsp);
 498     migrate_check_parameter(who, "max-bandwidth", value);
 499 }
 500
 501 static void test_deprecated(void)
 502 {
 503     QTestState *from;
 504
 505     from = qtest_start("");
 506
 507     deprecated_set_downtime(from, 0.12345);
 508     deprecated_set_speed(from, "12345");
 509
 510     qtest_quit(from);
 511 }
 512
 513 static void test_postcopy(void)
 514 {
 515     char *uri = g_strdup_printf("unix:%s/migsocket", tmpfs);
 516     QTestState *from, *to;
 517
 518     if (test_migrate_start(&from, &to, uri, false)) {
 519         return;
 520     }
 521
 522     migrate_set_capability(from, "postcopy-ram", "true");
 523     migrate_set_capability(to, "postcopy-ram", "true");
 524     migrate_set_capability(to, "postcopy-blocktime", "true");
 525
 526     /* We want to pick a speed slow enough that the test completes
 527      * quickly, but that it doesn't complete precopy even on a slow
 528      * machine, so also set the downtime.
 529      */
 530     migrate_set_parameter(from, "max-bandwidth", "100000000");
 531     migrate_set_parameter(from, "downtime-limit", "1");
 532
 533     /* Wait for the first serial output from the source */
 534     wait_for_serial("src_serial");
 535
 536     migrate(from, uri);
 537
 538     wait_for_migration_pass(from);
 539
 540     migrate_start_postcopy(from);
 541
 542     if (!got_stop) {
 543         qtest_qmp_eventwait(from, "STOP");
 544     }
 545
 546     qtest_qmp_eventwait(to, "RESUME");
 547
 548     wait_for_serial("dest_serial");
 549     wait_for_migration_complete(from);
 550
 551     if (uffd_feature_thread_id) {
 552         read_blocktime(to);
 553     }
 554     g_free(uri);
 555
 556     test_migrate_end(from, to, true);
 557 }
 558
 559 static void test_baddest(void)
 560 {
 561     QTestState *from, *to;
 562     QDict *rsp, *rsp_return;
 563     const char *status;
 564     bool failed;
 565
 566     if (test_migrate_start(&from, &to, "tcp:0:0", true)) {
 567         return;
 568     }
 569     migrate(from, "tcp:0:0");
 570     do {
 571         rsp = wait_command(from, "{ 'execute': 'query-migrate' }");
 572         rsp_return = qdict_get_qdict(rsp, "return");
 573
 574         status = qdict_get_str(rsp_return, "status");
 575
 576         g_assert(!strcmp(status, "setup") || !(strcmp(status, "failed")));
 577         failed = !strcmp(status, "failed");
 578         qobject_unref(rsp);
 579     } while (!failed);
 580
 581     /* Is the machine currently running? */
 582     rsp = wait_command(from, "{ 'execute': 'query-status' }");
 583     g_assert(qdict_haskey(rsp, "return"));
 584     rsp_return = qdict_get_qdict(rsp, "return");
 585     g_assert(qdict_haskey(rsp_return, "running"));
 586     g_assert(qdict_get_bool(rsp_return, "running"));
 587     qobject_unref(rsp);
 588
 589     test_migrate_end(from, to, false);
 590 }
 591
 592 static void test_precopy_unix(void)
 593 {
 594     char *uri = g_strdup_printf("unix:%s/migsocket", tmpfs);
 595     QTestState *from, *to;
 596
 597     if (test_migrate_start(&from, &to, uri, false)) {
 598         return;
 599     }
 600
 601     /* We want to pick a speed slow enough that the test completes
 602      * quickly, but that it doesn't complete precopy even on a slow
 603      * machine, so also set the downtime.
 604      */
 605     /* 1 ms should make it not converge*/
 606     migrate_set_parameter(from, "downtime-limit", "1");
 607     /* 1GB/s */
 608     migrate_set_parameter(from, "max-bandwidth", "1000000000");
 609
 610     /* Wait for the first serial output from the source */
 611     wait_for_serial("src_serial");
 612
 613     migrate(from, uri);
 614
 615     wait_for_migration_pass(from);
 616
 617     /* 300 ms should converge */
 618     migrate_set_parameter(from, "downtime-limit", "300");
 619
 620     if (!got_stop) {
 621         qtest_qmp_eventwait(from, "STOP");
 622     }
 623
 624     qtest_qmp_eventwait(to, "RESUME");
 625
 626     wait_for_serial("dest_serial");
 627     wait_for_migration_complete(from);
 628
 629     test_migrate_end(from, to, true);
 630     g_free(uri);
 631 }
 632
 633 int main(int argc, char **argv)
 634 {
 635     char template[] = "/tmp/migration-test-XXXXXX";
 636     int ret;
 637
 638     g_test_init(&argc, &argv, NULL);
 639
 640     if (!ufd_version_check()) {
 641         return 0;
 642     }
 643
 644     tmpfs = mkdtemp(template);
 645     if (!tmpfs) {
 646         g_test_message("mkdtemp on path (%s): %s\n", template, strerror(errno));
 647     }
 648     g_assert(tmpfs);
 649
 650     module_call_init(MODULE_INIT_QOM);
 651
 652     qtest_add_func("/migration/postcopy/unix", test_postcopy);
 653     qtest_add_func("/migration/deprecated", test_deprecated);
 654     qtest_add_func("/migration/bad_dest", test_baddest);
 655     qtest_add_func("/migration/precopy/unix", test_precopy_unix);
 656
 657     ret = g_test_run();
 658
 659     g_assert_cmpint(ret, ==, 0);
 660
 661     ret = rmdir(tmpfs);
 662     if (ret != 0) {
 663         g_test_message("unable to rmdir: path (%s): %s\n",
 664                        tmpfs, strerror(errno));
 665     }
 666
 667     return ret;
 668 }