xref: /qemu/tests/qtest/migration/framework.c (revision e1803dabdcbd26f2cf72e586f22b03e2e90d6110)
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 int migrate_start(QTestState **from, QTestState **to, const char *uri,
198                   MigrateStart *args)
199 {
200     g_autofree gchar *arch_source = NULL;
201     g_autofree gchar *arch_target = NULL;
202     /* options for source and target */
203     g_autofree gchar *arch_opts = NULL;
204     g_autofree gchar *cmd_source = NULL;
205     g_autofree gchar *cmd_target = NULL;
206     const gchar *ignore_stderr;
207     g_autofree char *shmem_opts = NULL;
208     g_autofree char *shmem_path = NULL;
209     const char *kvm_opts = NULL;
210     const char *arch = qtest_get_arch();
211     const char *memory_size;
212     const char *machine_alias, *machine_opts = "";
213     g_autofree char *machine = NULL;
214     const char *bootpath;
215 
216     if (args->use_shmem) {
217         if (!g_file_test("/dev/shm", G_FILE_TEST_IS_DIR)) {
218             g_test_skip("/dev/shm is not supported");
219             return -1;
220         }
221     }
222 
223     dst_state = (QTestMigrationState) { };
224     src_state = (QTestMigrationState) { };
225     bootpath = bootfile_create(arch, tmpfs, args->suspend_me);
226     src_state.suspend_me = args->suspend_me;
227 
228     if (strcmp(arch, "i386") == 0 || strcmp(arch, "x86_64") == 0) {
229         memory_size = "150M";
230 
231         if (g_str_equal(arch, "i386")) {
232             machine_alias = "pc";
233         } else {
234             machine_alias = "q35";
235         }
236         arch_opts = g_strdup_printf(
237             "-drive if=none,id=d0,file=%s,format=raw "
238             "-device ide-hd,drive=d0,secs=1,cyls=1,heads=1", bootpath);
239         start_address = X86_TEST_MEM_START;
240         end_address = X86_TEST_MEM_END;
241     } else if (g_str_equal(arch, "s390x")) {
242         memory_size = "128M";
243         machine_alias = "s390-ccw-virtio";
244         arch_opts = g_strdup_printf("-bios %s", bootpath);
245         start_address = S390_TEST_MEM_START;
246         end_address = S390_TEST_MEM_END;
247     } else if (strcmp(arch, "ppc64") == 0) {
248         memory_size = "256M";
249         start_address = PPC_TEST_MEM_START;
250         end_address = PPC_TEST_MEM_END;
251         machine_alias = "pseries";
252         machine_opts = "vsmt=8";
253         arch_opts = g_strdup_printf(
254             "-nodefaults -machine " PSERIES_DEFAULT_CAPABILITIES " "
255             "-bios %s", bootpath);
256     } else if (strcmp(arch, "aarch64") == 0) {
257         memory_size = "150M";
258         machine_alias = "virt";
259         machine_opts = "gic-version=3";
260         arch_opts = g_strdup_printf("-cpu max -kernel %s", bootpath);
261         start_address = ARM_TEST_MEM_START;
262         end_address = ARM_TEST_MEM_END;
263     } else {
264         g_assert_not_reached();
265     }
266 
267     if (!getenv("QTEST_LOG") && args->hide_stderr) {
268 #ifndef _WIN32
269         ignore_stderr = "2>/dev/null";
270 #else
271         /*
272          * On Windows the QEMU executable is created via CreateProcess() and
273          * IO redirection does not work, so don't bother adding IO redirection
274          * to the command line.
275          */
276         ignore_stderr = "";
277 #endif
278     } else {
279         ignore_stderr = "";
280     }
281 
282     if (args->use_shmem) {
283         shmem_path = g_strdup_printf("/dev/shm/qemu-%d", getpid());
284         shmem_opts = g_strdup_printf(
285             "-object memory-backend-file,id=mem0,size=%s"
286             ",mem-path=%s,share=on -numa node,memdev=mem0",
287             memory_size, shmem_path);
288     }
289 
290     if (args->use_dirty_ring) {
291         kvm_opts = ",dirty-ring-size=4096";
292     }
293 
294     if (!qtest_has_machine(machine_alias)) {
295         g_autofree char *msg = g_strdup_printf("machine %s not supported", machine_alias);
296         g_test_skip(msg);
297         return -1;
298     }
299 
300     machine = resolve_machine_version(machine_alias, QEMU_ENV_SRC,
301                                       QEMU_ENV_DST);
302 
303     g_test_message("Using machine type: %s", machine);
304 
305     cmd_source = g_strdup_printf("-accel kvm%s -accel tcg "
306                                  "-machine %s,%s "
307                                  "-name source,debug-threads=on "
308                                  "-m %s "
309                                  "-serial file:%s/src_serial "
310                                  "%s %s %s %s %s",
311                                  kvm_opts ? kvm_opts : "",
312                                  machine, machine_opts,
313                                  memory_size, tmpfs,
314                                  arch_opts ? arch_opts : "",
315                                  arch_source ? arch_source : "",
316                                  shmem_opts ? shmem_opts : "",
317                                  args->opts_source ? args->opts_source : "",
318                                  ignore_stderr);
319     if (!args->only_target) {
320         *from = qtest_init_with_env(QEMU_ENV_SRC, cmd_source);
321         qtest_qmp_set_event_callback(*from,
322                                      migrate_watch_for_events,
323                                      &src_state);
324     }
325 
326     cmd_target = g_strdup_printf("-accel kvm%s -accel tcg "
327                                  "-machine %s,%s "
328                                  "-name target,debug-threads=on "
329                                  "-m %s "
330                                  "-serial file:%s/dest_serial "
331                                  "-incoming %s "
332                                  "%s %s %s %s %s",
333                                  kvm_opts ? kvm_opts : "",
334                                  machine, machine_opts,
335                                  memory_size, tmpfs, uri,
336                                  arch_opts ? arch_opts : "",
337                                  arch_target ? arch_target : "",
338                                  shmem_opts ? shmem_opts : "",
339                                  args->opts_target ? args->opts_target : "",
340                                  ignore_stderr);
341     *to = qtest_init_with_env(QEMU_ENV_DST, cmd_target);
342     qtest_qmp_set_event_callback(*to,
343                                  migrate_watch_for_events,
344                                  &dst_state);
345 
346     /*
347      * Remove shmem file immediately to avoid memory leak in test failed case.
348      * It's valid because QEMU has already opened this file
349      */
350     if (args->use_shmem) {
351         unlink(shmem_path);
352     }
353 
354     /*
355      * Always enable migration events.  Libvirt always uses it, let's try
356      * to mimic as closer as that.
357      */
358     migrate_set_capability(*from, "events", true);
359     migrate_set_capability(*to, "events", true);
360 
361     return 0;
362 }
363 
364 void migrate_end(QTestState *from, QTestState *to, bool test_dest)
365 {
366     unsigned char dest_byte_a, dest_byte_b, dest_byte_c, dest_byte_d;
367 
368     qtest_quit(from);
369 
370     if (test_dest) {
371         qtest_memread(to, start_address, &dest_byte_a, 1);
372 
373         /* Destination still running, wait for a byte to change */
374         do {
375             qtest_memread(to, start_address, &dest_byte_b, 1);
376             usleep(1000 * 10);
377         } while (dest_byte_a == dest_byte_b);
378 
379         qtest_qmp_assert_success(to, "{ 'execute' : 'stop'}");
380 
381         /* With it stopped, check nothing changes */
382         qtest_memread(to, start_address, &dest_byte_c, 1);
383         usleep(1000 * 200);
384         qtest_memread(to, start_address, &dest_byte_d, 1);
385         g_assert_cmpint(dest_byte_c, ==, dest_byte_d);
386 
387         check_guests_ram(to);
388     }
389 
390     qtest_quit(to);
391 
392     cleanup("migsocket");
393     cleanup("src_serial");
394     cleanup("dest_serial");
395     cleanup(FILE_TEST_FILENAME);
396 }
397 
398 static int migrate_postcopy_prepare(QTestState **from_ptr,
399                                     QTestState **to_ptr,
400                                     MigrateCommon *args)
401 {
402     QTestState *from, *to;
403 
404     if (migrate_start(&from, &to, "defer", &args->start)) {
405         return -1;
406     }
407 
408     if (args->start_hook) {
409         args->postcopy_data = args->start_hook(from, to);
410     }
411 
412     migrate_set_capability(from, "postcopy-ram", true);
413     migrate_set_capability(to, "postcopy-ram", true);
414     migrate_set_capability(to, "postcopy-blocktime", true);
415 
416     if (args->postcopy_preempt) {
417         migrate_set_capability(from, "postcopy-preempt", true);
418         migrate_set_capability(to, "postcopy-preempt", true);
419     }
420 
421     migrate_ensure_non_converge(from);
422 
423     migrate_prepare_for_dirty_mem(from);
424     qtest_qmp_assert_success(to, "{ 'execute': 'migrate-incoming',"
425                              "  'arguments': { "
426                              "      'channels': [ { 'channel-type': 'main',"
427                              "      'addr': { 'transport': 'socket',"
428                              "                'type': 'inet',"
429                              "                'host': '127.0.0.1',"
430                              "                'port': '0' } } ] } }");
431 
432     /* Wait for the first serial output from the source */
433     wait_for_serial("src_serial");
434     wait_for_suspend(from, &src_state);
435 
436     migrate_qmp(from, to, NULL, NULL, "{}");
437 
438     migrate_wait_for_dirty_mem(from, to);
439 
440     *from_ptr = from;
441     *to_ptr = to;
442 
443     return 0;
444 }
445 
446 static void migrate_postcopy_complete(QTestState *from, QTestState *to,
447                                       MigrateCommon *args)
448 {
449     MigrationTestEnv *env = migration_get_env();
450 
451     wait_for_migration_complete(from);
452 
453     if (args->start.suspend_me) {
454         /* wakeup succeeds only if guest is suspended */
455         qtest_qmp_assert_success(to, "{'execute': 'system_wakeup'}");
456     }
457 
458     /* Make sure we get at least one "B" on destination */
459     wait_for_serial("dest_serial");
460 
461     if (env->uffd_feature_thread_id) {
462         read_blocktime(to);
463     }
464 
465     if (args->end_hook) {
466         args->end_hook(from, to, args->postcopy_data);
467         args->postcopy_data = NULL;
468     }
469 
470     migrate_end(from, to, true);
471 }
472 
473 void test_postcopy_common(MigrateCommon *args)
474 {
475     QTestState *from, *to;
476 
477     if (migrate_postcopy_prepare(&from, &to, args)) {
478         return;
479     }
480     migrate_postcopy_start(from, to, &src_state);
481     migrate_postcopy_complete(from, to, args);
482 }
483 
484 static void wait_for_postcopy_status(QTestState *one, const char *status)
485 {
486     wait_for_migration_status(one, status,
487                               (const char * []) {
488                                   "failed", "active",
489                                   "completed", NULL
490                               });
491 }
492 
493 static void postcopy_recover_fail(QTestState *from, QTestState *to,
494                                   PostcopyRecoveryFailStage stage)
495 {
496 #ifndef _WIN32
497     bool fail_early = (stage == POSTCOPY_FAIL_CHANNEL_ESTABLISH);
498     int ret, pair1[2], pair2[2];
499     char c;
500 
501     g_assert(stage > POSTCOPY_FAIL_NONE && stage < POSTCOPY_FAIL_MAX);
502 
503     /* Create two unrelated socketpairs */
504     ret = qemu_socketpair(PF_LOCAL, SOCK_STREAM, 0, pair1);
505     g_assert_cmpint(ret, ==, 0);
506 
507     ret = qemu_socketpair(PF_LOCAL, SOCK_STREAM, 0, pair2);
508     g_assert_cmpint(ret, ==, 0);
509 
510     /*
511      * Give the guests unpaired ends of the sockets, so they'll all blocked
512      * at reading.  This mimics a wrong channel established.
513      */
514     qtest_qmp_fds_assert_success(from, &pair1[0], 1,
515                                  "{ 'execute': 'getfd',"
516                                  "  'arguments': { 'fdname': 'fd-mig' }}");
517     qtest_qmp_fds_assert_success(to, &pair2[0], 1,
518                                  "{ 'execute': 'getfd',"
519                                  "  'arguments': { 'fdname': 'fd-mig' }}");
520 
521     /*
522      * Write the 1st byte as QEMU_VM_COMMAND (0x8) for the dest socket, to
523      * emulate the 1st byte of a real recovery, but stops from there to
524      * keep dest QEMU in RECOVER.  This is needed so that we can kick off
525      * the recover process on dest QEMU (by triggering the G_IO_IN event).
526      *
527      * NOTE: this trick is not needed on src QEMUs, because src doesn't
528      * rely on an pre-existing G_IO_IN event, so it will always trigger the
529      * upcoming recovery anyway even if it can read nothing.
530      */
531 #define QEMU_VM_COMMAND              0x08
532     c = QEMU_VM_COMMAND;
533     ret = send(pair2[1], &c, 1, 0);
534     g_assert_cmpint(ret, ==, 1);
535 
536     if (stage == POSTCOPY_FAIL_CHANNEL_ESTABLISH) {
537         /*
538          * This will make src QEMU to fail at an early stage when trying to
539          * resume later, where it shouldn't reach RECOVER stage at all.
540          */
541         close(pair1[1]);
542     }
543 
544     migrate_recover(to, "fd:fd-mig");
545     migrate_qmp(from, to, "fd:fd-mig", NULL, "{'resume': true}");
546 
547     /*
548      * Source QEMU has an extra RECOVER_SETUP phase, dest doesn't have it.
549      * Make sure it appears along the way.
550      */
551     migration_event_wait(from, "postcopy-recover-setup");
552 
553     if (fail_early) {
554         /*
555          * When fails at reconnection, src QEMU will automatically goes
556          * back to PAUSED state.  Making sure there is an event in this
557          * case: Libvirt relies on this to detect early reconnection
558          * errors.
559          */
560         migration_event_wait(from, "postcopy-paused");
561     } else {
562         /*
563          * We want to test "fail later" at RECOVER stage here.  Make sure
564          * both QEMU instances will go into RECOVER stage first, then test
565          * kicking them out using migrate-pause.
566          *
567          * Explicitly check the RECOVER event on src, that's what Libvirt
568          * relies on, rather than polling.
569          */
570         migration_event_wait(from, "postcopy-recover");
571         wait_for_postcopy_status(from, "postcopy-recover");
572 
573         /* Need an explicit kick on src QEMU in this case */
574         migrate_pause(from);
575     }
576 
577     /*
578      * For all failure cases, we'll reach such states on both sides now.
579      * Check them.
580      */
581     wait_for_postcopy_status(from, "postcopy-paused");
582     wait_for_postcopy_status(to, "postcopy-recover");
583 
584     /*
585      * Kick dest QEMU out too. This is normally not needed in reality
586      * because when the channel is shutdown it should also happen on src.
587      * However here we used separate socket pairs so we need to do that
588      * explicitly.
589      */
590     migrate_pause(to);
591     wait_for_postcopy_status(to, "postcopy-paused");
592 
593     close(pair1[0]);
594     close(pair2[0]);
595     close(pair2[1]);
596 
597     if (stage != POSTCOPY_FAIL_CHANNEL_ESTABLISH) {
598         close(pair1[1]);
599     }
600 #endif
601 }
602 
603 void test_postcopy_recovery_common(MigrateCommon *args)
604 {
605     QTestState *from, *to;
606     g_autofree char *uri = NULL;
607 
608     /* Always hide errors for postcopy recover tests since they're expected */
609     args->start.hide_stderr = true;
610 
611     if (migrate_postcopy_prepare(&from, &to, args)) {
612         return;
613     }
614 
615     /* Turn postcopy speed down, 4K/s is slow enough on any machines */
616     migrate_set_parameter_int(from, "max-postcopy-bandwidth", 4096);
617 
618     /* Now we start the postcopy */
619     migrate_postcopy_start(from, to, &src_state);
620 
621     /*
622      * Wait until postcopy is really started; we can only run the
623      * migrate-pause command during a postcopy
624      */
625     wait_for_migration_status(from, "postcopy-active", NULL);
626 
627     /*
628      * Manually stop the postcopy migration. This emulates a network
629      * failure with the migration socket
630      */
631     migrate_pause(from);
632 
633     /*
634      * Wait for destination side to reach postcopy-paused state.  The
635      * migrate-recover command can only succeed if destination machine
636      * is in the paused state
637      */
638     wait_for_postcopy_status(to, "postcopy-paused");
639     wait_for_postcopy_status(from, "postcopy-paused");
640 
641     if (args->postcopy_recovery_fail_stage) {
642         /*
643          * Test when a wrong socket specified for recover, and then the
644          * ability to kick it out, and continue with a correct socket.
645          */
646         postcopy_recover_fail(from, to, args->postcopy_recovery_fail_stage);
647         /* continue with a good recovery */
648     }
649 
650     /*
651      * Create a new socket to emulate a new channel that is different
652      * from the broken migration channel; tell the destination to
653      * listen to the new port
654      */
655     uri = g_strdup_printf("unix:%s/migsocket-recover", tmpfs);
656     migrate_recover(to, uri);
657 
658     /*
659      * Try to rebuild the migration channel using the resume flag and
660      * the newly created channel
661      */
662     migrate_qmp(from, to, uri, NULL, "{'resume': true}");
663 
664     /* Restore the postcopy bandwidth to unlimited */
665     migrate_set_parameter_int(from, "max-postcopy-bandwidth", 0);
666 
667     migrate_postcopy_complete(from, to, args);
668 }
669 
670 void test_precopy_common(MigrateCommon *args)
671 {
672     QTestState *from, *to;
673     void *data_hook = NULL;
674 
675     if (migrate_start(&from, &to, args->listen_uri, &args->start)) {
676         return;
677     }
678 
679     if (args->start_hook) {
680         data_hook = args->start_hook(from, to);
681     }
682 
683     /* Wait for the first serial output from the source */
684     if (args->result == MIG_TEST_SUCCEED) {
685         wait_for_serial("src_serial");
686         wait_for_suspend(from, &src_state);
687     }
688 
689     if (args->live) {
690         migrate_ensure_non_converge(from);
691         migrate_prepare_for_dirty_mem(from);
692     } else {
693         /*
694          * Testing non-live migration, we allow it to run at
695          * full speed to ensure short test case duration.
696          * For tests expected to fail, we don't need to
697          * change anything.
698          */
699         if (args->result == MIG_TEST_SUCCEED) {
700             qtest_qmp_assert_success(from, "{ 'execute' : 'stop'}");
701             wait_for_stop(from, &src_state);
702             migrate_ensure_converge(from);
703         }
704     }
705 
706     if (args->result == MIG_TEST_QMP_ERROR) {
707         migrate_qmp_fail(from, args->connect_uri, args->connect_channels, "{}");
708         goto finish;
709     }
710 
711     migrate_qmp(from, to, args->connect_uri, args->connect_channels, "{}");
712 
713     if (args->result != MIG_TEST_SUCCEED) {
714         bool allow_active = args->result == MIG_TEST_FAIL;
715         wait_for_migration_fail(from, allow_active);
716 
717         if (args->result == MIG_TEST_FAIL_DEST_QUIT_ERR) {
718             qtest_set_expected_status(to, EXIT_FAILURE);
719         }
720     } else {
721         if (args->live) {
722             /*
723              * For initial iteration(s) we must do a full pass,
724              * but for the final iteration, we need only wait
725              * for some dirty mem before switching to converge
726              */
727             while (args->iterations > 1) {
728                 wait_for_migration_pass(from, &src_state);
729                 args->iterations--;
730             }
731             migrate_wait_for_dirty_mem(from, to);
732 
733             migrate_ensure_converge(from);
734 
735             /*
736              * We do this first, as it has a timeout to stop us
737              * hanging forever if migration didn't converge
738              */
739             wait_for_migration_complete(from);
740 
741             wait_for_stop(from, &src_state);
742 
743         } else {
744             wait_for_migration_complete(from);
745             /*
746              * Must wait for dst to finish reading all incoming
747              * data on the socket before issuing 'cont' otherwise
748              * it'll be ignored
749              */
750             wait_for_migration_complete(to);
751 
752             qtest_qmp_assert_success(to, "{ 'execute' : 'cont'}");
753         }
754 
755         wait_for_resume(to, &dst_state);
756 
757         if (args->start.suspend_me) {
758             /* wakeup succeeds only if guest is suspended */
759             qtest_qmp_assert_success(to, "{'execute': 'system_wakeup'}");
760         }
761 
762         wait_for_serial("dest_serial");
763     }
764 
765 finish:
766     if (args->end_hook) {
767         args->end_hook(from, to, data_hook);
768     }
769 
770     migrate_end(from, to, args->result == MIG_TEST_SUCCEED);
771 }
772 
773 static void file_dirty_offset_region(void)
774 {
775     g_autofree char *path = g_strdup_printf("%s/%s", tmpfs, FILE_TEST_FILENAME);
776     size_t size = FILE_TEST_OFFSET;
777     g_autofree char *data = g_new0(char, size);
778 
779     memset(data, FILE_TEST_MARKER, size);
780     g_assert(g_file_set_contents(path, data, size, NULL));
781 }
782 
783 static void file_check_offset_region(void)
784 {
785     g_autofree char *path = g_strdup_printf("%s/%s", tmpfs, FILE_TEST_FILENAME);
786     size_t size = FILE_TEST_OFFSET;
787     g_autofree char *expected = g_new0(char, size);
788     g_autofree char *actual = NULL;
789     uint64_t *stream_start;
790 
791     /*
792      * Ensure the skipped offset region's data has not been touched
793      * and the migration stream starts at the right place.
794      */
795 
796     memset(expected, FILE_TEST_MARKER, size);
797 
798     g_assert(g_file_get_contents(path, &actual, NULL, NULL));
799     g_assert(!memcmp(actual, expected, size));
800 
801     stream_start = (uint64_t *)(actual + size);
802     g_assert_cmpint(cpu_to_be64(*stream_start) >> 32, ==, QEMU_VM_FILE_MAGIC);
803 }
804 
805 void test_file_common(MigrateCommon *args, bool stop_src)
806 {
807     QTestState *from, *to;
808     void *data_hook = NULL;
809     bool check_offset = false;
810 
811     if (migrate_start(&from, &to, args->listen_uri, &args->start)) {
812         return;
813     }
814 
815     /*
816      * File migration is never live. We can keep the source VM running
817      * during migration, but the destination will not be running
818      * concurrently.
819      */
820     g_assert_false(args->live);
821 
822     if (g_strrstr(args->connect_uri, "offset=")) {
823         check_offset = true;
824         /*
825          * This comes before the start_hook because it's equivalent to
826          * a management application creating the file and writing to
827          * it so hooks should expect the file to be already present.
828          */
829         file_dirty_offset_region();
830     }
831 
832     if (args->start_hook) {
833         data_hook = args->start_hook(from, to);
834     }
835 
836     migrate_ensure_converge(from);
837     wait_for_serial("src_serial");
838 
839     if (stop_src) {
840         qtest_qmp_assert_success(from, "{ 'execute' : 'stop'}");
841         wait_for_stop(from, &src_state);
842     }
843 
844     if (args->result == MIG_TEST_QMP_ERROR) {
845         migrate_qmp_fail(from, args->connect_uri, NULL, "{}");
846         goto finish;
847     }
848 
849     migrate_qmp(from, to, args->connect_uri, NULL, "{}");
850     wait_for_migration_complete(from);
851 
852     /*
853      * We need to wait for the source to finish before starting the
854      * destination.
855      */
856     migrate_incoming_qmp(to, args->connect_uri, "{}");
857     wait_for_migration_complete(to);
858 
859     if (stop_src) {
860         qtest_qmp_assert_success(to, "{ 'execute' : 'cont'}");
861     }
862     wait_for_resume(to, &dst_state);
863 
864     wait_for_serial("dest_serial");
865 
866     if (check_offset) {
867         file_check_offset_region();
868     }
869 
870 finish:
871     if (args->end_hook) {
872         args->end_hook(from, to, data_hook);
873     }
874 
875     migrate_end(from, to, args->result == MIG_TEST_SUCCEED);
876 }
877 
878 void *migrate_hook_start_precopy_tcp_multifd_common(QTestState *from,
879                                                     QTestState *to,
880                                                     const char *method)
881 {
882     migrate_set_parameter_int(from, "multifd-channels", 16);
883     migrate_set_parameter_int(to, "multifd-channels", 16);
884 
885     migrate_set_parameter_str(from, "multifd-compression", method);
886     migrate_set_parameter_str(to, "multifd-compression", method);
887 
888     migrate_set_capability(from, "multifd", true);
889     migrate_set_capability(to, "multifd", true);
890 
891     /* Start incoming migration from the 1st socket */
892     migrate_incoming_qmp(to, "tcp:127.0.0.1:0", "{}");
893 
894     return NULL;
895 }
896 
897 QTestMigrationState *get_src(void)
898 {
899     return &src_state;
900 }
901 
902 MigrationTestEnv *migration_get_env(void)
903 {
904     static MigrationTestEnv *env;
905     g_autoptr(GError) err = NULL;
906 
907     if (env) {
908         return env;
909     }
910 
911     env = g_new0(MigrationTestEnv, 1);
912     env->qemu_src = getenv(QEMU_ENV_SRC);
913     env->qemu_dst = getenv(QEMU_ENV_DST);
914 
915     /*
916      * The default QTEST_QEMU_BINARY must always be provided because
917      * that is what helpers use to query the accel type and
918      * architecture.
919      */
920     if (env->qemu_src && env->qemu_dst) {
921         g_test_message("Only one of %s, %s is allowed",
922                        QEMU_ENV_SRC, QEMU_ENV_DST);
923         exit(1);
924     }
925 
926     env->has_kvm = qtest_has_accel("kvm");
927     env->has_tcg = qtest_has_accel("tcg");
928 
929     if (!env->has_tcg && !env->has_kvm) {
930         g_test_skip("No KVM or TCG accelerator available");
931         return env;
932     }
933 
934     env->has_dirty_ring = kvm_dirty_ring_supported();
935     env->has_uffd = ufd_version_check(&env->uffd_feature_thread_id);
936     env->arch = qtest_get_arch();
937     env->is_x86 = !strcmp(env->arch, "i386") || !strcmp(env->arch, "x86_64");
938 
939     env->tmpfs = g_dir_make_tmp("migration-test-XXXXXX", &err);
940     if (!env->tmpfs) {
941         g_test_message("Can't create temporary directory in %s: %s",
942                        g_get_tmp_dir(), err->message);
943     }
944     g_assert(env->tmpfs);
945 
946     tmpfs = env->tmpfs;
947 
948     return env;
949 }
950 
951 int migration_env_clean(MigrationTestEnv *env)
952 {
953     char *tmpfs;
954     int ret = 0;
955 
956     if (!env) {
957         return ret;
958     }
959 
960     bootfile_delete();
961 
962     tmpfs = env->tmpfs;
963     ret = rmdir(tmpfs);
964     if (ret != 0) {
965         g_test_message("unable to rmdir: path (%s): %s",
966                        tmpfs, strerror(errno));
967     }
968     g_free(tmpfs);
969 
970     return ret;
971 }
972