xref: /qemu/tests/qtest/migration/framework.c (revision 3dec966f2798ebee41fc82fe4d1036d907ec51a4)
1 /*
2  * Copyright (c) 2016-2018 Red Hat, Inc. and/or its affiliates
3  *   based on the vhost-user-test.c that is:
4  *      Copyright (c) 2014 Virtual Open Systems Sarl.
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or later.
7  * See the COPYING file in the top-level directory.
8  *
9  */
10 
11 #include "qemu/osdep.h"
12 
13 #include "chardev/char.h"
14 #include "crypto/tlscredspsk.h"
15 #include "libqtest.h"
16 #include "migration/bootfile.h"
17 #include "migration/framework.h"
18 #include "migration/migration-qmp.h"
19 #include "migration/migration-util.h"
20 #include "ppc-util.h"
21 #include "qapi/qmp/qlist.h"
22 #include "qemu/module.h"
23 #include "qemu/option.h"
24 #include "qemu/range.h"
25 #include "qemu/sockets.h"
26 
27 
28 #define QEMU_VM_FILE_MAGIC 0x5145564d
29 #define QEMU_ENV_SRC "QTEST_QEMU_BINARY_SRC"
30 #define QEMU_ENV_DST "QTEST_QEMU_BINARY_DST"
31 
32 unsigned start_address;
33 unsigned end_address;
34 static QTestMigrationState src_state;
35 static QTestMigrationState dst_state;
36 static char *tmpfs;
37 
38 /*
39  * An initial 3 MB offset is used as that corresponds
40  * to ~1 sec of data transfer with our bandwidth setting.
41  */
42 #define MAGIC_OFFSET_BASE (3 * 1024 * 1024)
43 /*
44  * A further 1k is added to ensure we're not a multiple
45  * of TEST_MEM_PAGE_SIZE, thus avoid clash with writes
46  * from the migration guest workload.
47  */
48 #define MAGIC_OFFSET_SHUFFLE 1024
49 #define MAGIC_OFFSET (MAGIC_OFFSET_BASE + MAGIC_OFFSET_SHUFFLE)
50 #define MAGIC_MARKER 0xFEED12345678CAFEULL
51 
52 
53 /*
54  * Wait for some output in the serial output file,
55  * we get an 'A' followed by an endless string of 'B's
56  * but on the destination we won't have the A (unless we enabled suspend/resume)
57  */
58 void wait_for_serial(const char *side)
59 {
60     g_autofree char *serialpath = g_strdup_printf("%s/%s", tmpfs, side);
61     FILE *serialfile = fopen(serialpath, "r");
62 
63     do {
64         int readvalue = fgetc(serialfile);
65 
66         switch (readvalue) {
67         case 'A':
68             /* Fine */
69             break;
70 
71         case 'B':
72             /* It's alive! */
73             fclose(serialfile);
74             return;
75 
76         case EOF:
77             fseek(serialfile, 0, SEEK_SET);
78             usleep(1000);
79             break;
80 
81         default:
82             fprintf(stderr, "Unexpected %d on %s serial\n", readvalue, side);
83             g_assert_not_reached();
84         }
85     } while (true);
86 }
87 
88 void migrate_prepare_for_dirty_mem(QTestState *from)
89 {
90     /*
91      * The guest workflow iterates from start_address to
92      * end_address, writing 1 byte every TEST_MEM_PAGE_SIZE
93      * bytes.
94      *
95      * IOW, if we write to mem at a point which is NOT
96      * a multiple of TEST_MEM_PAGE_SIZE, our write won't
97      * conflict with the migration workflow.
98      *
99      * We put in a marker here, that we'll use to determine
100      * when the data has been transferred to the dst.
101      */
102     qtest_writeq(from, start_address + MAGIC_OFFSET, MAGIC_MARKER);
103 }
104 
105 void migrate_wait_for_dirty_mem(QTestState *from, QTestState *to)
106 {
107     uint64_t watch_address = start_address + MAGIC_OFFSET_BASE;
108     uint64_t marker_address = start_address + MAGIC_OFFSET;
109     uint8_t watch_byte;
110 
111     /*
112      * Wait for the MAGIC_MARKER to get transferred, as an
113      * indicator that a migration pass has made some known
114      * amount of progress.
115      */
116     do {
117         usleep(1000 * 10);
118     } while (qtest_readq(to, marker_address) != MAGIC_MARKER);
119 
120 
121     /* If suspended, src only iterates once, and watch_byte may never change */
122     if (src_state.suspend_me) {
123         return;
124     }
125 
126     /*
127      * Now ensure that already transferred bytes are
128      * dirty again from the guest workload. Note the
129      * guest byte value will wrap around and by chance
130      * match the original watch_byte. This is harmless
131      * as we'll eventually see a different value if we
132      * keep watching
133      */
134     watch_byte = qtest_readb(from, watch_address);
135     do {
136         usleep(1000 * 10);
137     } while (qtest_readb(from, watch_address) == watch_byte);
138 }
139 
140 static void check_guests_ram(QTestState *who)
141 {
142     /*
143      * Our ASM test will have been incrementing one byte from each page from
144      * start_address to < end_address in order. This gives us a constraint
145      * that any page's byte should be equal or less than the previous pages
146      * byte (mod 256); and they should all be equal except for one transition
147      * at the point where we meet the incrementer. (We're running this with
148      * the guest stopped).
149      */
150     unsigned address;
151     uint8_t first_byte;
152     uint8_t last_byte;
153     bool hit_edge = false;
154     int bad = 0;
155 
156     qtest_memread(who, start_address, &first_byte, 1);
157     last_byte = first_byte;
158 
159     for (address = start_address + TEST_MEM_PAGE_SIZE; address < end_address;
160          address += TEST_MEM_PAGE_SIZE)
161     {
162         uint8_t b;
163         qtest_memread(who, address, &b, 1);
164         if (b != last_byte) {
165             if (((b + 1) % 256) == last_byte && !hit_edge) {
166                 /*
167                  * This is OK, the guest stopped at the point of
168                  * incrementing the previous page but didn't get
169                  * to us yet.
170                  */
171                 hit_edge = true;
172                 last_byte = b;
173             } else {
174                 bad++;
175                 if (bad <= 10) {
176                     fprintf(stderr, "Memory content inconsistency at %x"
177                             " first_byte = %x last_byte = %x current = %x"
178                             " hit_edge = %x\n",
179                             address, first_byte, last_byte, b, hit_edge);
180                 }
181             }
182         }
183     }
184     if (bad >= 10) {
185         fprintf(stderr, "and in another %d pages", bad - 10);
186     }
187     g_assert(bad == 0);
188 }
189 
190 static void cleanup(const char *filename)
191 {
192     g_autofree char *path = g_strdup_printf("%s/%s", tmpfs, filename);
193 
194     unlink(path);
195 }
196 
197 static QList *migrate_start_get_qmp_capabilities(const MigrateStart *args)
198 {
199     QList *capabilities = qlist_new();
200 
201     if (args->oob) {
202         qlist_append_str(capabilities, "oob");
203     }
204     return capabilities;
205 }
206 
207 int migrate_start(QTestState **from, QTestState **to, const char *uri,
208                   MigrateStart *args)
209 {
210     /* options for source and target */
211     g_autofree gchar *arch_opts = NULL;
212     g_autofree gchar *cmd_source = NULL;
213     g_autofree gchar *cmd_target = NULL;
214     const gchar *ignore_stderr;
215     g_autofree char *shmem_opts = NULL;
216     g_autofree char *shmem_path = NULL;
217     const char *kvm_opts = NULL;
218     const char *arch = qtest_get_arch();
219     const char *memory_size;
220     const char *machine_alias, *machine_opts = "";
221     g_autofree char *machine = NULL;
222     const char *bootpath;
223     g_autoptr(QList) capabilities = migrate_start_get_qmp_capabilities(args);
224 
225     if (args->use_shmem) {
226         if (!g_file_test("/dev/shm", G_FILE_TEST_IS_DIR)) {
227             g_test_skip("/dev/shm is not supported");
228             return -1;
229         }
230     }
231 
232     dst_state = (QTestMigrationState) { };
233     src_state = (QTestMigrationState) { };
234     bootpath = bootfile_create(arch, tmpfs, args->suspend_me);
235     src_state.suspend_me = args->suspend_me;
236 
237     if (strcmp(arch, "i386") == 0 || strcmp(arch, "x86_64") == 0) {
238         memory_size = "150M";
239 
240         if (g_str_equal(arch, "i386")) {
241             machine_alias = "pc";
242         } else {
243             machine_alias = "q35";
244         }
245         arch_opts = g_strdup_printf(
246             "-drive if=none,id=d0,file=%s,format=raw "
247             "-device ide-hd,drive=d0,secs=1,cyls=1,heads=1", bootpath);
248         start_address = X86_TEST_MEM_START;
249         end_address = X86_TEST_MEM_END;
250     } else if (g_str_equal(arch, "s390x")) {
251         memory_size = "128M";
252         machine_alias = "s390-ccw-virtio";
253         arch_opts = g_strdup_printf("-bios %s", bootpath);
254         start_address = S390_TEST_MEM_START;
255         end_address = S390_TEST_MEM_END;
256     } else if (strcmp(arch, "ppc64") == 0) {
257         memory_size = "256M";
258         start_address = PPC_TEST_MEM_START;
259         end_address = PPC_TEST_MEM_END;
260         machine_alias = "pseries";
261         machine_opts = "vsmt=8";
262         arch_opts = g_strdup_printf(
263             "-nodefaults -machine " PSERIES_DEFAULT_CAPABILITIES " "
264             "-bios %s", bootpath);
265     } else if (strcmp(arch, "aarch64") == 0) {
266         memory_size = "150M";
267         machine_alias = "virt";
268         machine_opts = "gic-version=3";
269         arch_opts = g_strdup_printf("-cpu max -kernel %s", bootpath);
270         start_address = ARM_TEST_MEM_START;
271         end_address = ARM_TEST_MEM_END;
272     } else {
273         g_assert_not_reached();
274     }
275 
276     if (!getenv("QTEST_LOG") && args->hide_stderr) {
277 #ifndef _WIN32
278         ignore_stderr = "2>/dev/null";
279 #else
280         /*
281          * On Windows the QEMU executable is created via CreateProcess() and
282          * IO redirection does not work, so don't bother adding IO redirection
283          * to the command line.
284          */
285         ignore_stderr = "";
286 #endif
287     } else {
288         ignore_stderr = "";
289     }
290 
291     if (args->use_shmem) {
292         shmem_path = g_strdup_printf("/dev/shm/qemu-%d", getpid());
293         shmem_opts = g_strdup_printf(
294             "-object memory-backend-file,id=mem0,size=%s"
295             ",mem-path=%s,share=on -numa node,memdev=mem0",
296             memory_size, shmem_path);
297     }
298 
299     if (args->use_dirty_ring) {
300         kvm_opts = ",dirty-ring-size=4096";
301     }
302 
303     if (!qtest_has_machine(machine_alias)) {
304         g_autofree char *msg = g_strdup_printf("machine %s not supported", machine_alias);
305         g_test_skip(msg);
306         return -1;
307     }
308 
309     machine = resolve_machine_version(machine_alias, QEMU_ENV_SRC,
310                                       QEMU_ENV_DST);
311 
312     g_test_message("Using machine type: %s", machine);
313 
314     cmd_source = g_strdup_printf("-accel kvm%s -accel tcg "
315                                  "-machine %s,%s "
316                                  "-name source,debug-threads=on "
317                                  "-m %s "
318                                  "-serial file:%s/src_serial "
319                                  "%s %s %s %s",
320                                  kvm_opts ? kvm_opts : "",
321                                  machine, machine_opts,
322                                  memory_size, tmpfs,
323                                  arch_opts ? arch_opts : "",
324                                  shmem_opts ? shmem_opts : "",
325                                  args->opts_source ? args->opts_source : "",
326                                  ignore_stderr);
327     if (!args->only_target) {
328         *from = qtest_init_with_env_and_capabilities(QEMU_ENV_SRC, cmd_source,
329                                                      capabilities);
330         qtest_qmp_set_event_callback(*from,
331                                      migrate_watch_for_events,
332                                      &src_state);
333     }
334 
335     cmd_target = g_strdup_printf("-accel kvm%s -accel tcg "
336                                  "-machine %s,%s "
337                                  "-name target,debug-threads=on "
338                                  "-m %s "
339                                  "-serial file:%s/dest_serial "
340                                  "-incoming %s "
341                                  "%s %s %s %s",
342                                  kvm_opts ? kvm_opts : "",
343                                  machine, machine_opts,
344                                  memory_size, tmpfs, uri,
345                                  arch_opts ? arch_opts : "",
346                                  shmem_opts ? shmem_opts : "",
347                                  args->opts_target ? args->opts_target : "",
348                                  ignore_stderr);
349     *to = qtest_init_with_env_and_capabilities(QEMU_ENV_DST, cmd_target,
350                                                capabilities);
351     qtest_qmp_set_event_callback(*to,
352                                  migrate_watch_for_events,
353                                  &dst_state);
354 
355     /*
356      * Remove shmem file immediately to avoid memory leak in test failed case.
357      * It's valid because QEMU has already opened this file
358      */
359     if (args->use_shmem) {
360         unlink(shmem_path);
361     }
362 
363     /*
364      * Always enable migration events.  Libvirt always uses it, let's try
365      * to mimic as closer as that.
366      */
367     migrate_set_capability(*from, "events", true);
368     migrate_set_capability(*to, "events", true);
369 
370     return 0;
371 }
372 
373 void migrate_end(QTestState *from, QTestState *to, bool test_dest)
374 {
375     unsigned char dest_byte_a, dest_byte_b, dest_byte_c, dest_byte_d;
376 
377     qtest_quit(from);
378 
379     if (test_dest) {
380         qtest_memread(to, start_address, &dest_byte_a, 1);
381 
382         /* Destination still running, wait for a byte to change */
383         do {
384             qtest_memread(to, start_address, &dest_byte_b, 1);
385             usleep(1000 * 10);
386         } while (dest_byte_a == dest_byte_b);
387 
388         qtest_qmp_assert_success(to, "{ 'execute' : 'stop'}");
389 
390         /* With it stopped, check nothing changes */
391         qtest_memread(to, start_address, &dest_byte_c, 1);
392         usleep(1000 * 200);
393         qtest_memread(to, start_address, &dest_byte_d, 1);
394         g_assert_cmpint(dest_byte_c, ==, dest_byte_d);
395 
396         check_guests_ram(to);
397     }
398 
399     qtest_quit(to);
400 
401     cleanup("migsocket");
402     cleanup("src_serial");
403     cleanup("dest_serial");
404     cleanup(FILE_TEST_FILENAME);
405 }
406 
407 static int migrate_postcopy_prepare(QTestState **from_ptr,
408                                     QTestState **to_ptr,
409                                     MigrateCommon *args)
410 {
411     QTestState *from, *to;
412 
413     if (migrate_start(&from, &to, "defer", &args->start)) {
414         return -1;
415     }
416 
417     if (args->start_hook) {
418         args->postcopy_data = args->start_hook(from, to);
419     }
420 
421     migrate_set_capability(from, "postcopy-ram", true);
422     migrate_set_capability(to, "postcopy-ram", true);
423     migrate_set_capability(to, "postcopy-blocktime", true);
424 
425     if (args->postcopy_preempt) {
426         migrate_set_capability(from, "postcopy-preempt", true);
427         migrate_set_capability(to, "postcopy-preempt", true);
428     }
429 
430     migrate_ensure_non_converge(from);
431 
432     migrate_prepare_for_dirty_mem(from);
433     qtest_qmp_assert_success(to, "{ 'execute': 'migrate-incoming',"
434                              "  'arguments': { "
435                              "      'channels': [ { 'channel-type': 'main',"
436                              "      'addr': { 'transport': 'socket',"
437                              "                'type': 'inet',"
438                              "                'host': '127.0.0.1',"
439                              "                'port': '0' } } ] } }");
440 
441     /* Wait for the first serial output from the source */
442     wait_for_serial("src_serial");
443     wait_for_suspend(from, &src_state);
444 
445     migrate_qmp(from, to, NULL, NULL, "{}");
446 
447     migrate_wait_for_dirty_mem(from, to);
448 
449     *from_ptr = from;
450     *to_ptr = to;
451 
452     return 0;
453 }
454 
455 static void migrate_postcopy_complete(QTestState *from, QTestState *to,
456                                       MigrateCommon *args)
457 {
458     MigrationTestEnv *env = migration_get_env();
459 
460     wait_for_migration_complete(from);
461 
462     if (args->start.suspend_me) {
463         /* wakeup succeeds only if guest is suspended */
464         qtest_qmp_assert_success(to, "{'execute': 'system_wakeup'}");
465     }
466 
467     /* Make sure we get at least one "B" on destination */
468     wait_for_serial("dest_serial");
469 
470     if (env->uffd_feature_thread_id) {
471         read_blocktime(to);
472     }
473 
474     if (args->end_hook) {
475         args->end_hook(from, to, args->postcopy_data);
476         args->postcopy_data = NULL;
477     }
478 
479     migrate_end(from, to, true);
480 }
481 
482 void test_postcopy_common(MigrateCommon *args)
483 {
484     QTestState *from, *to;
485 
486     if (migrate_postcopy_prepare(&from, &to, args)) {
487         return;
488     }
489     migrate_postcopy_start(from, to, &src_state);
490     migrate_postcopy_complete(from, to, args);
491 }
492 
493 static void wait_for_postcopy_status(QTestState *one, const char *status)
494 {
495     wait_for_migration_status(one, status,
496                               (const char * []) {
497                                   "failed", "active",
498                                   "completed", NULL
499                               });
500 }
501 
502 static void postcopy_recover_fail(QTestState *from, QTestState *to,
503                                   PostcopyRecoveryFailStage stage)
504 {
505 #ifndef _WIN32
506     bool fail_early = (stage == POSTCOPY_FAIL_CHANNEL_ESTABLISH);
507     int ret, pair1[2], pair2[2];
508     char c;
509 
510     g_assert(stage > POSTCOPY_FAIL_NONE && stage < POSTCOPY_FAIL_MAX);
511 
512     /* Create two unrelated socketpairs */
513     ret = qemu_socketpair(PF_LOCAL, SOCK_STREAM, 0, pair1);
514     g_assert_cmpint(ret, ==, 0);
515 
516     ret = qemu_socketpair(PF_LOCAL, SOCK_STREAM, 0, pair2);
517     g_assert_cmpint(ret, ==, 0);
518 
519     /*
520      * Give the guests unpaired ends of the sockets, so they'll all blocked
521      * at reading.  This mimics a wrong channel established.
522      */
523     qtest_qmp_fds_assert_success(from, &pair1[0], 1,
524                                  "{ 'execute': 'getfd',"
525                                  "  'arguments': { 'fdname': 'fd-mig' }}");
526     qtest_qmp_fds_assert_success(to, &pair2[0], 1,
527                                  "{ 'execute': 'getfd',"
528                                  "  'arguments': { 'fdname': 'fd-mig' }}");
529 
530     /*
531      * Write the 1st byte as QEMU_VM_COMMAND (0x8) for the dest socket, to
532      * emulate the 1st byte of a real recovery, but stops from there to
533      * keep dest QEMU in RECOVER.  This is needed so that we can kick off
534      * the recover process on dest QEMU (by triggering the G_IO_IN event).
535      *
536      * NOTE: this trick is not needed on src QEMUs, because src doesn't
537      * rely on an pre-existing G_IO_IN event, so it will always trigger the
538      * upcoming recovery anyway even if it can read nothing.
539      */
540 #define QEMU_VM_COMMAND              0x08
541     c = QEMU_VM_COMMAND;
542     ret = send(pair2[1], &c, 1, 0);
543     g_assert_cmpint(ret, ==, 1);
544 
545     if (stage == POSTCOPY_FAIL_CHANNEL_ESTABLISH) {
546         /*
547          * This will make src QEMU to fail at an early stage when trying to
548          * resume later, where it shouldn't reach RECOVER stage at all.
549          */
550         close(pair1[1]);
551     }
552 
553     migrate_recover(to, "fd:fd-mig");
554     migrate_qmp(from, to, "fd:fd-mig", NULL, "{'resume': true}");
555 
556     /*
557      * Source QEMU has an extra RECOVER_SETUP phase, dest doesn't have it.
558      * Make sure it appears along the way.
559      */
560     migration_event_wait(from, "postcopy-recover-setup");
561 
562     if (fail_early) {
563         /*
564          * When fails at reconnection, src QEMU will automatically goes
565          * back to PAUSED state.  Making sure there is an event in this
566          * case: Libvirt relies on this to detect early reconnection
567          * errors.
568          */
569         migration_event_wait(from, "postcopy-paused");
570     } else {
571         /*
572          * We want to test "fail later" at RECOVER stage here.  Make sure
573          * both QEMU instances will go into RECOVER stage first, then test
574          * kicking them out using migrate-pause.
575          *
576          * Explicitly check the RECOVER event on src, that's what Libvirt
577          * relies on, rather than polling.
578          */
579         migration_event_wait(from, "postcopy-recover");
580         wait_for_postcopy_status(from, "postcopy-recover");
581 
582         /* Need an explicit kick on src QEMU in this case */
583         migrate_pause(from);
584     }
585 
586     /*
587      * For all failure cases, we'll reach such states on both sides now.
588      * Check them.
589      */
590     wait_for_postcopy_status(from, "postcopy-paused");
591     wait_for_postcopy_status(to, "postcopy-recover");
592 
593     /*
594      * Kick dest QEMU out too. This is normally not needed in reality
595      * because when the channel is shutdown it should also happen on src.
596      * However here we used separate socket pairs so we need to do that
597      * explicitly.
598      */
599     migrate_pause(to);
600     wait_for_postcopy_status(to, "postcopy-paused");
601 
602     close(pair1[0]);
603     close(pair2[0]);
604     close(pair2[1]);
605 
606     if (stage != POSTCOPY_FAIL_CHANNEL_ESTABLISH) {
607         close(pair1[1]);
608     }
609 #endif
610 }
611 
612 void test_postcopy_recovery_common(MigrateCommon *args)
613 {
614     QTestState *from, *to;
615     g_autofree char *uri = NULL;
616 
617     /*
618      * Always enable OOB QMP capability for recovery tests, migrate-recover is
619      * executed out-of-band
620      */
621     args->start.oob = true;
622 
623     /* Always hide errors for postcopy recover tests since they're expected */
624     args->start.hide_stderr = true;
625 
626     if (migrate_postcopy_prepare(&from, &to, args)) {
627         return;
628     }
629 
630     /* Turn postcopy speed down, 4K/s is slow enough on any machines */
631     migrate_set_parameter_int(from, "max-postcopy-bandwidth", 4096);
632 
633     /* Now we start the postcopy */
634     migrate_postcopy_start(from, to, &src_state);
635 
636     /*
637      * Wait until postcopy is really started; we can only run the
638      * migrate-pause command during a postcopy
639      */
640     wait_for_migration_status(from, "postcopy-active", NULL);
641 
642     /*
643      * Manually stop the postcopy migration. This emulates a network
644      * failure with the migration socket
645      */
646     migrate_pause(from);
647 
648     /*
649      * Wait for destination side to reach postcopy-paused state.  The
650      * migrate-recover command can only succeed if destination machine
651      * is in the paused state
652      */
653     wait_for_postcopy_status(to, "postcopy-paused");
654     wait_for_postcopy_status(from, "postcopy-paused");
655 
656     if (args->postcopy_recovery_fail_stage) {
657         /*
658          * Test when a wrong socket specified for recover, and then the
659          * ability to kick it out, and continue with a correct socket.
660          */
661         postcopy_recover_fail(from, to, args->postcopy_recovery_fail_stage);
662         /* continue with a good recovery */
663     }
664 
665     /*
666      * Create a new socket to emulate a new channel that is different
667      * from the broken migration channel; tell the destination to
668      * listen to the new port
669      */
670     uri = g_strdup_printf("unix:%s/migsocket-recover", tmpfs);
671     migrate_recover(to, uri);
672 
673     /*
674      * Try to rebuild the migration channel using the resume flag and
675      * the newly created channel
676      */
677     migrate_qmp(from, to, uri, NULL, "{'resume': true}");
678 
679     /* Restore the postcopy bandwidth to unlimited */
680     migrate_set_parameter_int(from, "max-postcopy-bandwidth", 0);
681 
682     migrate_postcopy_complete(from, to, args);
683 }
684 
685 void test_precopy_common(MigrateCommon *args)
686 {
687     QTestState *from, *to;
688     void *data_hook = NULL;
689 
690     if (migrate_start(&from, &to, args->listen_uri, &args->start)) {
691         return;
692     }
693 
694     if (args->start_hook) {
695         data_hook = args->start_hook(from, to);
696     }
697 
698     /* Wait for the first serial output from the source */
699     if (args->result == MIG_TEST_SUCCEED) {
700         wait_for_serial("src_serial");
701         wait_for_suspend(from, &src_state);
702     }
703 
704     if (args->live) {
705         migrate_ensure_non_converge(from);
706         migrate_prepare_for_dirty_mem(from);
707     } else {
708         /*
709          * Testing non-live migration, we allow it to run at
710          * full speed to ensure short test case duration.
711          * For tests expected to fail, we don't need to
712          * change anything.
713          */
714         if (args->result == MIG_TEST_SUCCEED) {
715             qtest_qmp_assert_success(from, "{ 'execute' : 'stop'}");
716             wait_for_stop(from, &src_state);
717             migrate_ensure_converge(from);
718         }
719     }
720 
721     if (args->result == MIG_TEST_QMP_ERROR) {
722         migrate_qmp_fail(from, args->connect_uri, args->connect_channels, "{}");
723         goto finish;
724     }
725 
726     migrate_qmp(from, to, args->connect_uri, args->connect_channels, "{}");
727 
728     if (args->result != MIG_TEST_SUCCEED) {
729         bool allow_active = args->result == MIG_TEST_FAIL;
730         wait_for_migration_fail(from, allow_active);
731 
732         if (args->result == MIG_TEST_FAIL_DEST_QUIT_ERR) {
733             qtest_set_expected_status(to, EXIT_FAILURE);
734         }
735     } else {
736         if (args->live) {
737             /*
738              * For initial iteration(s) we must do a full pass,
739              * but for the final iteration, we need only wait
740              * for some dirty mem before switching to converge
741              */
742             while (args->iterations > 1) {
743                 wait_for_migration_pass(from, &src_state);
744                 args->iterations--;
745             }
746             migrate_wait_for_dirty_mem(from, to);
747 
748             migrate_ensure_converge(from);
749 
750             /*
751              * We do this first, as it has a timeout to stop us
752              * hanging forever if migration didn't converge
753              */
754             wait_for_migration_complete(from);
755 
756             wait_for_stop(from, &src_state);
757 
758         } else {
759             wait_for_migration_complete(from);
760             /*
761              * Must wait for dst to finish reading all incoming
762              * data on the socket before issuing 'cont' otherwise
763              * it'll be ignored
764              */
765             wait_for_migration_complete(to);
766 
767             qtest_qmp_assert_success(to, "{ 'execute' : 'cont'}");
768         }
769 
770         wait_for_resume(to, &dst_state);
771 
772         if (args->start.suspend_me) {
773             /* wakeup succeeds only if guest is suspended */
774             qtest_qmp_assert_success(to, "{'execute': 'system_wakeup'}");
775         }
776 
777         wait_for_serial("dest_serial");
778     }
779 
780 finish:
781     if (args->end_hook) {
782         args->end_hook(from, to, data_hook);
783     }
784 
785     migrate_end(from, to, args->result == MIG_TEST_SUCCEED);
786 }
787 
788 static void file_dirty_offset_region(void)
789 {
790     g_autofree char *path = g_strdup_printf("%s/%s", tmpfs, FILE_TEST_FILENAME);
791     size_t size = FILE_TEST_OFFSET;
792     g_autofree char *data = g_new0(char, size);
793 
794     memset(data, FILE_TEST_MARKER, size);
795     g_assert(g_file_set_contents(path, data, size, NULL));
796 }
797 
798 static void file_check_offset_region(void)
799 {
800     g_autofree char *path = g_strdup_printf("%s/%s", tmpfs, FILE_TEST_FILENAME);
801     size_t size = FILE_TEST_OFFSET;
802     g_autofree char *expected = g_new0(char, size);
803     g_autofree char *actual = NULL;
804     uint64_t *stream_start;
805 
806     /*
807      * Ensure the skipped offset region's data has not been touched
808      * and the migration stream starts at the right place.
809      */
810 
811     memset(expected, FILE_TEST_MARKER, size);
812 
813     g_assert(g_file_get_contents(path, &actual, NULL, NULL));
814     g_assert(!memcmp(actual, expected, size));
815 
816     stream_start = (uint64_t *)(actual + size);
817     g_assert_cmpint(cpu_to_be64(*stream_start) >> 32, ==, QEMU_VM_FILE_MAGIC);
818 }
819 
820 void test_file_common(MigrateCommon *args, bool stop_src)
821 {
822     QTestState *from, *to;
823     void *data_hook = NULL;
824     bool check_offset = false;
825 
826     if (migrate_start(&from, &to, args->listen_uri, &args->start)) {
827         return;
828     }
829 
830     /*
831      * File migration is never live. We can keep the source VM running
832      * during migration, but the destination will not be running
833      * concurrently.
834      */
835     g_assert_false(args->live);
836 
837     if (g_strrstr(args->connect_uri, "offset=")) {
838         check_offset = true;
839         /*
840          * This comes before the start_hook because it's equivalent to
841          * a management application creating the file and writing to
842          * it so hooks should expect the file to be already present.
843          */
844         file_dirty_offset_region();
845     }
846 
847     if (args->start_hook) {
848         data_hook = args->start_hook(from, to);
849     }
850 
851     migrate_ensure_converge(from);
852     wait_for_serial("src_serial");
853 
854     if (stop_src) {
855         qtest_qmp_assert_success(from, "{ 'execute' : 'stop'}");
856         wait_for_stop(from, &src_state);
857     }
858 
859     if (args->result == MIG_TEST_QMP_ERROR) {
860         migrate_qmp_fail(from, args->connect_uri, NULL, "{}");
861         goto finish;
862     }
863 
864     migrate_qmp(from, to, args->connect_uri, NULL, "{}");
865     wait_for_migration_complete(from);
866 
867     /*
868      * We need to wait for the source to finish before starting the
869      * destination.
870      */
871     migrate_incoming_qmp(to, args->connect_uri, "{}");
872     wait_for_migration_complete(to);
873 
874     if (stop_src) {
875         qtest_qmp_assert_success(to, "{ 'execute' : 'cont'}");
876     }
877     wait_for_resume(to, &dst_state);
878 
879     wait_for_serial("dest_serial");
880 
881     if (check_offset) {
882         file_check_offset_region();
883     }
884 
885 finish:
886     if (args->end_hook) {
887         args->end_hook(from, to, data_hook);
888     }
889 
890     migrate_end(from, to, args->result == MIG_TEST_SUCCEED);
891 }
892 
893 void *migrate_hook_start_precopy_tcp_multifd_common(QTestState *from,
894                                                     QTestState *to,
895                                                     const char *method)
896 {
897     migrate_set_parameter_int(from, "multifd-channels", 16);
898     migrate_set_parameter_int(to, "multifd-channels", 16);
899 
900     migrate_set_parameter_str(from, "multifd-compression", method);
901     migrate_set_parameter_str(to, "multifd-compression", method);
902 
903     migrate_set_capability(from, "multifd", true);
904     migrate_set_capability(to, "multifd", true);
905 
906     /* Start incoming migration from the 1st socket */
907     migrate_incoming_qmp(to, "tcp:127.0.0.1:0", "{}");
908 
909     return NULL;
910 }
911 
912 QTestMigrationState *get_src(void)
913 {
914     return &src_state;
915 }
916 
917 MigrationTestEnv *migration_get_env(void)
918 {
919     static MigrationTestEnv *env;
920     g_autoptr(GError) err = NULL;
921 
922     if (env) {
923         return env;
924     }
925 
926     env = g_new0(MigrationTestEnv, 1);
927     env->qemu_src = getenv(QEMU_ENV_SRC);
928     env->qemu_dst = getenv(QEMU_ENV_DST);
929 
930     /*
931      * The default QTEST_QEMU_BINARY must always be provided because
932      * that is what helpers use to query the accel type and
933      * architecture.
934      */
935     if (env->qemu_src && env->qemu_dst) {
936         g_test_message("Only one of %s, %s is allowed",
937                        QEMU_ENV_SRC, QEMU_ENV_DST);
938         exit(1);
939     }
940 
941     env->has_kvm = qtest_has_accel("kvm");
942     env->has_tcg = qtest_has_accel("tcg");
943 
944     if (!env->has_tcg && !env->has_kvm) {
945         g_test_skip("No KVM or TCG accelerator available");
946         return env;
947     }
948 
949     env->has_dirty_ring = kvm_dirty_ring_supported();
950     env->has_uffd = ufd_version_check(&env->uffd_feature_thread_id);
951     env->arch = qtest_get_arch();
952     env->is_x86 = !strcmp(env->arch, "i386") || !strcmp(env->arch, "x86_64");
953 
954     env->tmpfs = g_dir_make_tmp("migration-test-XXXXXX", &err);
955     if (!env->tmpfs) {
956         g_test_message("Can't create temporary directory in %s: %s",
957                        g_get_tmp_dir(), err->message);
958     }
959     g_assert(env->tmpfs);
960 
961     tmpfs = env->tmpfs;
962 
963     return env;
964 }
965 
966 int migration_env_clean(MigrationTestEnv *env)
967 {
968     char *tmpfs;
969     int ret = 0;
970 
971     if (!env) {
972         return ret;
973     }
974 
975     bootfile_delete();
976 
977     tmpfs = env->tmpfs;
978     ret = rmdir(tmpfs);
979     if (ret != 0) {
980         g_test_message("unable to rmdir: path (%s): %s",
981                        tmpfs, strerror(errno));
982     }
983     g_free(tmpfs);
984 
985     return ret;
986 }
987