xref: /qemu/tests/qtest/migration/framework.c (revision 69ee0189d7977cfbb1b2c7a27393d8b9fb661b20)
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/error.h"
22 #include "qobject/qjson.h"
23 #include "qobject/qlist.h"
24 #include "qemu/module.h"
25 #include "qemu/option.h"
26 #include "qemu/range.h"
27 #include "qemu/sockets.h"
28 
29 
30 #define QEMU_VM_FILE_MAGIC 0x5145564d
31 #define QEMU_ENV_SRC "QTEST_QEMU_BINARY_SRC"
32 #define QEMU_ENV_DST "QTEST_QEMU_BINARY_DST"
33 #define MULTIFD_TEST_CHANNELS 4
34 
35 unsigned start_address;
36 unsigned end_address;
37 static QTestMigrationState src_state;
38 static QTestMigrationState dst_state;
39 static char *tmpfs;
40 
41 /*
42  * An initial 3 MB offset is used as that corresponds
43  * to ~1 sec of data transfer with our bandwidth setting.
44  */
45 #define MAGIC_OFFSET_BASE (3 * 1024 * 1024)
46 /*
47  * A further 1k is added to ensure we're not a multiple
48  * of TEST_MEM_PAGE_SIZE, thus avoid clash with writes
49  * from the migration guest workload.
50  */
51 #define MAGIC_OFFSET_SHUFFLE 1024
52 #define MAGIC_OFFSET (MAGIC_OFFSET_BASE + MAGIC_OFFSET_SHUFFLE)
53 #define MAGIC_MARKER 0xFEED12345678CAFEULL
54 
55 
56 /*
57  * Wait for some output in the serial output file,
58  * we get an 'A' followed by an endless string of 'B's
59  * but on the destination we won't have the A (unless we enabled suspend/resume)
60  */
wait_for_serial(const char * side)61 void wait_for_serial(const char *side)
62 {
63     g_autofree char *serialpath = g_strdup_printf("%s/%s", tmpfs, side);
64     FILE *serialfile = fopen(serialpath, "r");
65 
66     do {
67         int readvalue = fgetc(serialfile);
68 
69         switch (readvalue) {
70         case 'A':
71             /* Fine */
72             break;
73 
74         case 'B':
75             /* It's alive! */
76             fclose(serialfile);
77             return;
78 
79         case EOF:
80             fseek(serialfile, 0, SEEK_SET);
81             usleep(1000);
82             break;
83 
84         default:
85             fprintf(stderr, "Unexpected %d on %s serial\n", readvalue, side);
86             g_assert_not_reached();
87         }
88     } while (true);
89 }
90 
migrate_prepare_for_dirty_mem(QTestState * from)91 void migrate_prepare_for_dirty_mem(QTestState *from)
92 {
93     /*
94      * The guest workflow iterates from start_address to
95      * end_address, writing 1 byte every TEST_MEM_PAGE_SIZE
96      * bytes.
97      *
98      * IOW, if we write to mem at a point which is NOT
99      * a multiple of TEST_MEM_PAGE_SIZE, our write won't
100      * conflict with the migration workflow.
101      *
102      * We put in a marker here, that we'll use to determine
103      * when the data has been transferred to the dst.
104      */
105     qtest_writeq(from, start_address + MAGIC_OFFSET, MAGIC_MARKER);
106 }
107 
migrate_wait_for_dirty_mem(QTestState * from,QTestState * to)108 void migrate_wait_for_dirty_mem(QTestState *from, QTestState *to)
109 {
110     uint64_t watch_address = start_address + MAGIC_OFFSET_BASE;
111     uint64_t marker_address = start_address + MAGIC_OFFSET;
112     uint8_t watch_byte;
113 
114     /*
115      * Wait for the MAGIC_MARKER to get transferred, as an
116      * indicator that a migration pass has made some known
117      * amount of progress.
118      */
119     do {
120         usleep(1000 * 10);
121     } while (qtest_readq(to, marker_address) != MAGIC_MARKER);
122 
123 
124     /* If suspended, src only iterates once, and watch_byte may never change */
125     if (src_state.suspend_me) {
126         return;
127     }
128 
129     /*
130      * Now ensure that already transferred bytes are
131      * dirty again from the guest workload. Note the
132      * guest byte value will wrap around and by chance
133      * match the original watch_byte. This is harmless
134      * as we'll eventually see a different value if we
135      * keep watching
136      */
137     watch_byte = qtest_readb(from, watch_address);
138     do {
139         usleep(1000 * 10);
140     } while (qtest_readb(from, watch_address) == watch_byte);
141 }
142 
check_guests_ram(QTestState * who)143 static void check_guests_ram(QTestState *who)
144 {
145     /*
146      * Our ASM test will have been incrementing one byte from each page from
147      * start_address to < end_address in order. This gives us a constraint
148      * that any page's byte should be equal or less than the previous pages
149      * byte (mod 256); and they should all be equal except for one transition
150      * at the point where we meet the incrementer. (We're running this with
151      * the guest stopped).
152      */
153     unsigned address;
154     uint8_t first_byte;
155     uint8_t last_byte;
156     bool hit_edge = false;
157     int bad = 0;
158 
159     qtest_memread(who, start_address, &first_byte, 1);
160     last_byte = first_byte;
161 
162     for (address = start_address + TEST_MEM_PAGE_SIZE; address < end_address;
163          address += TEST_MEM_PAGE_SIZE)
164     {
165         uint8_t b;
166         qtest_memread(who, address, &b, 1);
167         if (b != last_byte) {
168             if (((b + 1) % 256) == last_byte && !hit_edge) {
169                 /*
170                  * This is OK, the guest stopped at the point of
171                  * incrementing the previous page but didn't get
172                  * to us yet.
173                  */
174                 hit_edge = true;
175                 last_byte = b;
176             } else {
177                 bad++;
178                 if (bad <= 10) {
179                     fprintf(stderr, "Memory content inconsistency at %x"
180                             " first_byte = %x last_byte = %x current = %x"
181                             " hit_edge = %x\n",
182                             address, first_byte, last_byte, b, hit_edge);
183                 }
184             }
185         }
186     }
187     if (bad >= 10) {
188         fprintf(stderr, "and in another %d pages", bad - 10);
189     }
190     g_assert(bad == 0);
191 }
192 
cleanup(const char * filename)193 static void cleanup(const char *filename)
194 {
195     g_autofree char *path = g_strdup_printf("%s/%s", tmpfs, filename);
196 
197     unlink(path);
198 }
199 
migrate_start_get_qmp_capabilities(const MigrateStart * args)200 static QList *migrate_start_get_qmp_capabilities(const MigrateStart *args)
201 {
202     QList *capabilities = NULL;
203 
204     if (args->oob) {
205         capabilities = qlist_new();
206         qlist_append_str(capabilities, "oob");
207     }
208     return capabilities;
209 }
210 
migrate_start_set_capabilities(QTestState * from,QTestState * to,MigrateStart * args)211 static void migrate_start_set_capabilities(QTestState *from, QTestState *to,
212                                            MigrateStart *args)
213 {
214     /*
215      * MigrationCapability_lookup and MIGRATION_CAPABILITY_ constants
216      * are from qapi-types-migration.h.
217      */
218     for (uint8_t i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
219         if (!args->caps[i]) {
220             continue;
221         }
222         if (from) {
223             migrate_set_capability(from,
224                             MigrationCapability_lookup.array[i], true);
225         }
226         if (to) {
227             migrate_set_capability(to,
228                             MigrationCapability_lookup.array[i], true);
229         }
230     }
231 
232     /*
233      * Always enable migration events.  Libvirt always uses it, let's try
234      * to mimic as closer as that.
235      */
236     migrate_set_capability(from, "events", true);
237     if (!args->defer_target_connect) {
238         migrate_set_capability(to, "events", true);
239     }
240 
241     /*
242      * Default number of channels should be fine for most
243      * tests. Individual tests can override by calling
244      * migrate_set_parameter() directly.
245      */
246     if (args->caps[MIGRATION_CAPABILITY_MULTIFD]) {
247         migrate_set_parameter_int(from, "multifd-channels",
248                                   MULTIFD_TEST_CHANNELS);
249         migrate_set_parameter_int(to, "multifd-channels",
250                                   MULTIFD_TEST_CHANNELS);
251     }
252 
253     return;
254 }
255 
migrate_start(QTestState ** from,QTestState ** to,const char * uri,MigrateStart * args)256 int migrate_start(QTestState **from, QTestState **to, const char *uri,
257                   MigrateStart *args)
258 {
259     /* options for source and target */
260     g_autofree gchar *arch_opts = NULL;
261     g_autofree gchar *cmd_source = NULL;
262     g_autofree gchar *cmd_target = NULL;
263     const gchar *ignore_stderr;
264     g_autofree char *shmem_opts = NULL;
265     g_autofree char *shmem_path = NULL;
266     const char *kvm_opts = NULL;
267     const char *arch = qtest_get_arch();
268     const char *memory_size;
269     const char *machine_alias, *machine_opts = "";
270     g_autofree char *machine = NULL;
271     const char *bootpath;
272     g_autoptr(QList) capabilities = migrate_start_get_qmp_capabilities(args);
273     g_autofree char *memory_backend = NULL;
274     const char *events;
275 
276     if (args->use_shmem) {
277         if (!g_file_test("/dev/shm", G_FILE_TEST_IS_DIR)) {
278             g_test_skip("/dev/shm is not supported");
279             return -1;
280         }
281     }
282 
283     dst_state = (QTestMigrationState) { };
284     src_state = (QTestMigrationState) { };
285     bootpath = bootfile_create(arch, tmpfs, args->suspend_me);
286     src_state.suspend_me = args->suspend_me;
287 
288     if (strcmp(arch, "i386") == 0 || strcmp(arch, "x86_64") == 0) {
289         memory_size = "150M";
290 
291         if (g_str_equal(arch, "i386")) {
292             machine_alias = "pc";
293         } else {
294             machine_alias = "q35";
295         }
296         arch_opts = g_strdup_printf(
297             "-drive if=none,id=d0,file=%s,format=raw "
298             "-device ide-hd,drive=d0,secs=1,cyls=1,heads=1", bootpath);
299         start_address = X86_TEST_MEM_START;
300         end_address = X86_TEST_MEM_END;
301     } else if (g_str_equal(arch, "s390x")) {
302         memory_size = "128M";
303         machine_alias = "s390-ccw-virtio";
304         arch_opts = g_strdup_printf("-bios %s", bootpath);
305         start_address = S390_TEST_MEM_START;
306         end_address = S390_TEST_MEM_END;
307     } else if (strcmp(arch, "ppc64") == 0) {
308         memory_size = "256M";
309         start_address = PPC_TEST_MEM_START;
310         end_address = PPC_TEST_MEM_END;
311         machine_alias = "pseries";
312         machine_opts = "vsmt=8";
313         arch_opts = g_strdup_printf(
314             "-nodefaults -machine " PSERIES_DEFAULT_CAPABILITIES " "
315             "-bios %s", bootpath);
316     } else if (strcmp(arch, "aarch64") == 0) {
317         memory_size = "150M";
318         machine_alias = "virt";
319         machine_opts = "gic-version=3";
320         arch_opts = g_strdup_printf("-cpu max -kernel %s", bootpath);
321         start_address = ARM_TEST_MEM_START;
322         end_address = ARM_TEST_MEM_END;
323     } else {
324         g_assert_not_reached();
325     }
326 
327     if (!getenv("QTEST_LOG") && args->hide_stderr) {
328 #ifndef _WIN32
329         ignore_stderr = "2>/dev/null";
330 #else
331         /*
332          * On Windows the QEMU executable is created via CreateProcess() and
333          * IO redirection does not work, so don't bother adding IO redirection
334          * to the command line.
335          */
336         ignore_stderr = "";
337 #endif
338     } else {
339         ignore_stderr = "";
340     }
341 
342     if (args->use_shmem) {
343         shmem_path = g_strdup_printf("/dev/shm/qemu-%d", getpid());
344         shmem_opts = g_strdup_printf(
345             "-object memory-backend-file,id=mem0,size=%s"
346             ",mem-path=%s,share=on -numa node,memdev=mem0",
347             memory_size, shmem_path);
348     }
349 
350     if (args->memory_backend) {
351         memory_backend = g_strdup_printf(args->memory_backend, memory_size);
352     } else {
353         memory_backend = g_strdup_printf("-m %s ", memory_size);
354     }
355 
356     if (args->use_dirty_ring) {
357         kvm_opts = ",dirty-ring-size=4096";
358     }
359 
360     if (!qtest_has_machine(machine_alias)) {
361         g_autofree char *msg = g_strdup_printf("machine %s not supported", machine_alias);
362         g_test_skip(msg);
363         return -1;
364     }
365 
366     machine = resolve_machine_version(machine_alias, QEMU_ENV_SRC,
367                                       QEMU_ENV_DST);
368 
369     g_test_message("Using machine type: %s", machine);
370 
371     cmd_source = g_strdup_printf("-accel kvm%s -accel tcg "
372                                  "-machine %s,%s "
373                                  "-name source,debug-threads=on "
374                                  "%s "
375                                  "-serial file:%s/src_serial "
376                                  "%s %s %s %s",
377                                  kvm_opts ? kvm_opts : "",
378                                  machine, machine_opts,
379                                  memory_backend, tmpfs,
380                                  arch_opts ? arch_opts : "",
381                                  shmem_opts ? shmem_opts : "",
382                                  args->opts_source ? args->opts_source : "",
383                                  ignore_stderr);
384     if (!args->only_target) {
385         *from = qtest_init_ext(QEMU_ENV_SRC, cmd_source, capabilities, true);
386         qtest_qmp_set_event_callback(*from,
387                                      migrate_watch_for_events,
388                                      &src_state);
389     }
390 
391     /*
392      * If the monitor connection is deferred, enable events on the command line
393      * so none are missed.  This is for testing only, do not set migration
394      * options like this in general.
395      */
396     events = args->defer_target_connect ? "-global migration.x-events=on" : "";
397 
398     cmd_target = g_strdup_printf("-accel kvm%s -accel tcg "
399                                  "-machine %s,%s "
400                                  "-name target,debug-threads=on "
401                                  "%s "
402                                  "-serial file:%s/dest_serial "
403                                  "-incoming %s "
404                                  "%s %s %s %s %s",
405                                  kvm_opts ? kvm_opts : "",
406                                  machine, machine_opts,
407                                  memory_backend, tmpfs, uri,
408                                  events,
409                                  arch_opts ? arch_opts : "",
410                                  shmem_opts ? shmem_opts : "",
411                                  args->opts_target ? args->opts_target : "",
412                                  ignore_stderr);
413     *to = qtest_init_ext(QEMU_ENV_DST, cmd_target, capabilities,
414                          !args->defer_target_connect);
415     qtest_qmp_set_event_callback(*to,
416                                  migrate_watch_for_events,
417                                  &dst_state);
418 
419     /*
420      * Remove shmem file immediately to avoid memory leak in test failed case.
421      * It's valid because QEMU has already opened this file
422      */
423     if (args->use_shmem) {
424         unlink(shmem_path);
425     }
426 
427     migrate_start_set_capabilities(*from, *to, args);
428 
429     return 0;
430 }
431 
migrate_end(QTestState * from,QTestState * to,bool test_dest)432 void migrate_end(QTestState *from, QTestState *to, bool test_dest)
433 {
434     unsigned char dest_byte_a, dest_byte_b, dest_byte_c, dest_byte_d;
435 
436     qtest_quit(from);
437 
438     if (test_dest) {
439         qtest_memread(to, start_address, &dest_byte_a, 1);
440 
441         /* Destination still running, wait for a byte to change */
442         do {
443             qtest_memread(to, start_address, &dest_byte_b, 1);
444             usleep(1000 * 10);
445         } while (dest_byte_a == dest_byte_b);
446 
447         qtest_qmp_assert_success(to, "{ 'execute' : 'stop'}");
448 
449         /* With it stopped, check nothing changes */
450         qtest_memread(to, start_address, &dest_byte_c, 1);
451         usleep(1000 * 200);
452         qtest_memread(to, start_address, &dest_byte_d, 1);
453         g_assert_cmpint(dest_byte_c, ==, dest_byte_d);
454 
455         check_guests_ram(to);
456     }
457 
458     qtest_quit(to);
459 
460     cleanup("migsocket");
461     cleanup("cpr.sock");
462     cleanup("src_serial");
463     cleanup("dest_serial");
464     cleanup(FILE_TEST_FILENAME);
465 }
466 
migrate_postcopy_prepare(QTestState ** from_ptr,QTestState ** to_ptr,MigrateCommon * args)467 static int migrate_postcopy_prepare(QTestState **from_ptr,
468                                     QTestState **to_ptr,
469                                     MigrateCommon *args)
470 {
471     QTestState *from, *to;
472 
473     /* set postcopy capabilities */
474     args->start.caps[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME] = true;
475     args->start.caps[MIGRATION_CAPABILITY_POSTCOPY_RAM] = true;
476 
477     if (migrate_start(&from, &to, "defer", &args->start)) {
478         return -1;
479     }
480 
481     if (args->start_hook) {
482         args->postcopy_data = args->start_hook(from, to);
483     }
484 
485     migrate_ensure_non_converge(from);
486     migrate_prepare_for_dirty_mem(from);
487     qtest_qmp_assert_success(to, "{ 'execute': 'migrate-incoming',"
488                              "  'arguments': { "
489                              "      'channels': [ { 'channel-type': 'main',"
490                              "      'addr': { 'transport': 'socket',"
491                              "                'type': 'inet',"
492                              "                'host': '127.0.0.1',"
493                              "                'port': '0' } } ] } }");
494 
495     /* Wait for the first serial output from the source */
496     wait_for_serial("src_serial");
497     wait_for_suspend(from, &src_state);
498 
499     migrate_qmp(from, to, NULL, NULL, "{}");
500 
501     migrate_wait_for_dirty_mem(from, to);
502 
503     *from_ptr = from;
504     *to_ptr = to;
505 
506     return 0;
507 }
508 
migrate_postcopy_complete(QTestState * from,QTestState * to,MigrateCommon * args)509 static void migrate_postcopy_complete(QTestState *from, QTestState *to,
510                                       MigrateCommon *args)
511 {
512     MigrationTestEnv *env = migration_get_env();
513 
514     wait_for_migration_complete(from);
515 
516     if (args->start.suspend_me) {
517         /* wakeup succeeds only if guest is suspended */
518         qtest_qmp_assert_success(to, "{'execute': 'system_wakeup'}");
519     }
520 
521     /* Make sure we get at least one "B" on destination */
522     wait_for_serial("dest_serial");
523 
524     if (env->uffd_feature_thread_id) {
525         read_blocktime(to);
526     }
527 
528     if (args->end_hook) {
529         args->end_hook(from, to, args->postcopy_data);
530         args->postcopy_data = NULL;
531     }
532 
533     migrate_end(from, to, true);
534 }
535 
test_postcopy_common(MigrateCommon * args)536 void test_postcopy_common(MigrateCommon *args)
537 {
538     QTestState *from, *to;
539 
540     if (migrate_postcopy_prepare(&from, &to, args)) {
541         return;
542     }
543     migrate_postcopy_start(from, to, &src_state);
544     migrate_postcopy_complete(from, to, args);
545 }
546 
wait_for_postcopy_status(QTestState * one,const char * status)547 static void wait_for_postcopy_status(QTestState *one, const char *status)
548 {
549     wait_for_migration_status(one, status,
550                               (const char * []) {
551                                   "failed", "active",
552                                   "completed", NULL
553                               });
554 }
555 
postcopy_recover_fail(QTestState * from,QTestState * to,PostcopyRecoveryFailStage stage)556 static void postcopy_recover_fail(QTestState *from, QTestState *to,
557                                   PostcopyRecoveryFailStage stage)
558 {
559 #ifndef _WIN32
560     bool fail_early = (stage == POSTCOPY_FAIL_CHANNEL_ESTABLISH);
561     int ret, pair1[2], pair2[2];
562     char c;
563 
564     g_assert(stage > POSTCOPY_FAIL_NONE && stage < POSTCOPY_FAIL_MAX);
565 
566     /* Create two unrelated socketpairs */
567     ret = qemu_socketpair(PF_LOCAL, SOCK_STREAM, 0, pair1);
568     g_assert_cmpint(ret, ==, 0);
569 
570     ret = qemu_socketpair(PF_LOCAL, SOCK_STREAM, 0, pair2);
571     g_assert_cmpint(ret, ==, 0);
572 
573     /*
574      * Give the guests unpaired ends of the sockets, so they'll all blocked
575      * at reading.  This mimics a wrong channel established.
576      */
577     qtest_qmp_fds_assert_success(from, &pair1[0], 1,
578                                  "{ 'execute': 'getfd',"
579                                  "  'arguments': { 'fdname': 'fd-mig' }}");
580     qtest_qmp_fds_assert_success(to, &pair2[0], 1,
581                                  "{ 'execute': 'getfd',"
582                                  "  'arguments': { 'fdname': 'fd-mig' }}");
583 
584     /*
585      * Write the 1st byte as QEMU_VM_COMMAND (0x8) for the dest socket, to
586      * emulate the 1st byte of a real recovery, but stops from there to
587      * keep dest QEMU in RECOVER.  This is needed so that we can kick off
588      * the recover process on dest QEMU (by triggering the G_IO_IN event).
589      *
590      * NOTE: this trick is not needed on src QEMUs, because src doesn't
591      * rely on an pre-existing G_IO_IN event, so it will always trigger the
592      * upcoming recovery anyway even if it can read nothing.
593      */
594 #define QEMU_VM_COMMAND              0x08
595     c = QEMU_VM_COMMAND;
596     ret = send(pair2[1], &c, 1, 0);
597     g_assert_cmpint(ret, ==, 1);
598 
599     if (stage == POSTCOPY_FAIL_CHANNEL_ESTABLISH) {
600         /*
601          * This will make src QEMU to fail at an early stage when trying to
602          * resume later, where it shouldn't reach RECOVER stage at all.
603          */
604         close(pair1[1]);
605     }
606 
607     migrate_recover(to, "fd:fd-mig");
608     migrate_qmp(from, to, "fd:fd-mig", NULL, "{'resume': true}");
609 
610     /*
611      * Source QEMU has an extra RECOVER_SETUP phase, dest doesn't have it.
612      * Make sure it appears along the way.
613      */
614     migration_event_wait(from, "postcopy-recover-setup");
615 
616     if (fail_early) {
617         /*
618          * When fails at reconnection, src QEMU will automatically goes
619          * back to PAUSED state.  Making sure there is an event in this
620          * case: Libvirt relies on this to detect early reconnection
621          * errors.
622          */
623         migration_event_wait(from, "postcopy-paused");
624     } else {
625         /*
626          * We want to test "fail later" at RECOVER stage here.  Make sure
627          * both QEMU instances will go into RECOVER stage first, then test
628          * kicking them out using migrate-pause.
629          *
630          * Explicitly check the RECOVER event on src, that's what Libvirt
631          * relies on, rather than polling.
632          */
633         migration_event_wait(from, "postcopy-recover");
634         wait_for_postcopy_status(from, "postcopy-recover");
635 
636         /* Need an explicit kick on src QEMU in this case */
637         migrate_pause(from);
638     }
639 
640     /*
641      * For all failure cases, we'll reach such states on both sides now.
642      * Check them.
643      */
644     wait_for_postcopy_status(from, "postcopy-paused");
645     wait_for_postcopy_status(to, "postcopy-recover");
646 
647     /*
648      * Kick dest QEMU out too. This is normally not needed in reality
649      * because when the channel is shutdown it should also happen on src.
650      * However here we used separate socket pairs so we need to do that
651      * explicitly.
652      */
653     migrate_pause(to);
654     wait_for_postcopy_status(to, "postcopy-paused");
655 
656     close(pair1[0]);
657     close(pair2[0]);
658     close(pair2[1]);
659 
660     if (stage != POSTCOPY_FAIL_CHANNEL_ESTABLISH) {
661         close(pair1[1]);
662     }
663 #endif
664 }
665 
test_postcopy_recovery_common(MigrateCommon * args)666 void test_postcopy_recovery_common(MigrateCommon *args)
667 {
668     QTestState *from, *to;
669     g_autofree char *uri = NULL;
670 
671     /*
672      * Always enable OOB QMP capability for recovery tests, migrate-recover is
673      * executed out-of-band
674      */
675     args->start.oob = true;
676 
677     /* Always hide errors for postcopy recover tests since they're expected */
678     args->start.hide_stderr = true;
679 
680     if (migrate_postcopy_prepare(&from, &to, args)) {
681         return;
682     }
683 
684     /* Turn postcopy speed down, 4K/s is slow enough on any machines */
685     migrate_set_parameter_int(from, "max-postcopy-bandwidth", 4096);
686 
687     /* Now we start the postcopy */
688     migrate_postcopy_start(from, to, &src_state);
689 
690     /*
691      * Wait until postcopy is really started; we can only run the
692      * migrate-pause command during a postcopy
693      */
694     wait_for_migration_status(from, "postcopy-active", NULL);
695 
696     /*
697      * Manually stop the postcopy migration. This emulates a network
698      * failure with the migration socket
699      */
700     migrate_pause(from);
701 
702     /*
703      * Wait for destination side to reach postcopy-paused state.  The
704      * migrate-recover command can only succeed if destination machine
705      * is in the paused state
706      */
707     wait_for_postcopy_status(to, "postcopy-paused");
708     wait_for_postcopy_status(from, "postcopy-paused");
709 
710     if (args->postcopy_recovery_fail_stage) {
711         /*
712          * Test when a wrong socket specified for recover, and then the
713          * ability to kick it out, and continue with a correct socket.
714          */
715         postcopy_recover_fail(from, to, args->postcopy_recovery_fail_stage);
716         /* continue with a good recovery */
717     }
718 
719     /*
720      * Create a new socket to emulate a new channel that is different
721      * from the broken migration channel; tell the destination to
722      * listen to the new port
723      */
724     uri = g_strdup_printf("unix:%s/migsocket-recover", tmpfs);
725     migrate_recover(to, uri);
726 
727     /*
728      * Try to rebuild the migration channel using the resume flag and
729      * the newly created channel
730      */
731     migrate_qmp(from, to, uri, NULL, "{'resume': true}");
732 
733     /* Restore the postcopy bandwidth to unlimited */
734     migrate_set_parameter_int(from, "max-postcopy-bandwidth", 0);
735 
736     migrate_postcopy_complete(from, to, args);
737 }
738 
test_precopy_common(MigrateCommon * args)739 void test_precopy_common(MigrateCommon *args)
740 {
741     QTestState *from, *to;
742     void *data_hook = NULL;
743     QObject *in_channels = NULL;
744     QObject *out_channels = NULL;
745 
746     g_assert(!args->cpr_channel || args->connect_channels);
747 
748     if (migrate_start(&from, &to, args->listen_uri, &args->start)) {
749         return;
750     }
751 
752     if (args->start_hook) {
753         data_hook = args->start_hook(from, to);
754     }
755 
756     /* Wait for the first serial output from the source */
757     if (args->result == MIG_TEST_SUCCEED) {
758         wait_for_serial("src_serial");
759         wait_for_suspend(from, &src_state);
760     }
761 
762     if (args->live) {
763         migrate_ensure_non_converge(from);
764         migrate_prepare_for_dirty_mem(from);
765     } else {
766         /*
767          * Testing non-live migration, we allow it to run at
768          * full speed to ensure short test case duration.
769          * For tests expected to fail, we don't need to
770          * change anything.
771          */
772         if (args->result == MIG_TEST_SUCCEED) {
773             qtest_qmp_assert_success(from, "{ 'execute' : 'stop'}");
774             wait_for_stop(from, &src_state);
775             migrate_ensure_converge(from);
776         }
777     }
778 
779     /*
780      * The cpr channel must be included in outgoing channels, but not in
781      * migrate-incoming channels.
782      */
783     if (args->connect_channels) {
784         if (args->start.defer_target_connect &&
785             !strcmp(args->listen_uri, "defer")) {
786             in_channels = qobject_from_json(args->connect_channels,
787                                             &error_abort);
788         }
789         out_channels = qobject_from_json(args->connect_channels, &error_abort);
790 
791         if (args->cpr_channel) {
792             QList *channels_list = qobject_to(QList, out_channels);
793             QObject *obj = migrate_str_to_channel(args->cpr_channel);
794 
795             qlist_append(channels_list, obj);
796         }
797     }
798 
799     if (args->result == MIG_TEST_QMP_ERROR) {
800         migrate_qmp_fail(from, args->connect_uri, out_channels, "{}");
801         goto finish;
802     }
803 
804     migrate_qmp(from, to, args->connect_uri, out_channels, "{}");
805 
806     if (args->start.defer_target_connect) {
807         qtest_connect(to);
808         qtest_qmp_handshake(to, NULL);
809         if (!strcmp(args->listen_uri, "defer")) {
810             migrate_incoming_qmp(to, args->connect_uri, in_channels, "{}");
811         }
812     }
813 
814     if (args->result != MIG_TEST_SUCCEED) {
815         bool allow_active = args->result == MIG_TEST_FAIL;
816         wait_for_migration_fail(from, allow_active);
817 
818         if (args->result == MIG_TEST_FAIL_DEST_QUIT_ERR) {
819             qtest_set_expected_status(to, EXIT_FAILURE);
820         }
821     } else {
822         if (args->live) {
823             /*
824              * For initial iteration(s) we must do a full pass,
825              * but for the final iteration, we need only wait
826              * for some dirty mem before switching to converge
827              */
828             while (args->iterations > 1) {
829                 wait_for_migration_pass(from, &src_state);
830                 args->iterations--;
831             }
832             migrate_wait_for_dirty_mem(from, to);
833 
834             migrate_ensure_converge(from);
835 
836             /*
837              * We do this first, as it has a timeout to stop us
838              * hanging forever if migration didn't converge
839              */
840             wait_for_migration_complete(from);
841 
842             wait_for_stop(from, &src_state);
843 
844         } else {
845             wait_for_migration_complete(from);
846             /*
847              * Must wait for dst to finish reading all incoming
848              * data on the socket before issuing 'cont' otherwise
849              * it'll be ignored
850              */
851             wait_for_migration_complete(to);
852 
853             qtest_qmp_assert_success(to, "{ 'execute' : 'cont'}");
854         }
855 
856         wait_for_resume(to, &dst_state);
857 
858         if (args->start.suspend_me) {
859             /* wakeup succeeds only if guest is suspended */
860             qtest_qmp_assert_success(to, "{'execute': 'system_wakeup'}");
861         }
862 
863         wait_for_serial("dest_serial");
864     }
865 
866 finish:
867     if (args->end_hook) {
868         args->end_hook(from, to, data_hook);
869     }
870 
871     migrate_end(from, to, args->result == MIG_TEST_SUCCEED);
872 }
873 
file_dirty_offset_region(void)874 static void file_dirty_offset_region(void)
875 {
876     g_autofree char *path = g_strdup_printf("%s/%s", tmpfs, FILE_TEST_FILENAME);
877     size_t size = FILE_TEST_OFFSET;
878     g_autofree char *data = g_new0(char, size);
879 
880     memset(data, FILE_TEST_MARKER, size);
881     g_assert(g_file_set_contents(path, data, size, NULL));
882 }
883 
file_check_offset_region(void)884 static void file_check_offset_region(void)
885 {
886     g_autofree char *path = g_strdup_printf("%s/%s", tmpfs, FILE_TEST_FILENAME);
887     size_t size = FILE_TEST_OFFSET;
888     g_autofree char *expected = g_new0(char, size);
889     g_autofree char *actual = NULL;
890     uint64_t *stream_start;
891 
892     /*
893      * Ensure the skipped offset region's data has not been touched
894      * and the migration stream starts at the right place.
895      */
896 
897     memset(expected, FILE_TEST_MARKER, size);
898 
899     g_assert(g_file_get_contents(path, &actual, NULL, NULL));
900     g_assert(!memcmp(actual, expected, size));
901 
902     stream_start = (uint64_t *)(actual + size);
903     g_assert_cmpint(cpu_to_be64(*stream_start) >> 32, ==, QEMU_VM_FILE_MAGIC);
904 }
905 
test_file_common(MigrateCommon * args,bool stop_src)906 void test_file_common(MigrateCommon *args, bool stop_src)
907 {
908     QTestState *from, *to;
909     void *data_hook = NULL;
910     bool check_offset = false;
911 
912     if (migrate_start(&from, &to, args->listen_uri, &args->start)) {
913         return;
914     }
915 
916     /*
917      * File migration is never live. We can keep the source VM running
918      * during migration, but the destination will not be running
919      * concurrently.
920      */
921     g_assert_false(args->live);
922 
923     if (g_strrstr(args->connect_uri, "offset=")) {
924         check_offset = true;
925         /*
926          * This comes before the start_hook because it's equivalent to
927          * a management application creating the file and writing to
928          * it so hooks should expect the file to be already present.
929          */
930         file_dirty_offset_region();
931     }
932 
933     if (args->start_hook) {
934         data_hook = args->start_hook(from, to);
935     }
936 
937     migrate_ensure_converge(from);
938     wait_for_serial("src_serial");
939 
940     if (stop_src) {
941         qtest_qmp_assert_success(from, "{ 'execute' : 'stop'}");
942         wait_for_stop(from, &src_state);
943     }
944 
945     if (args->result == MIG_TEST_QMP_ERROR) {
946         migrate_qmp_fail(from, args->connect_uri, NULL, "{}");
947         goto finish;
948     }
949 
950     migrate_qmp(from, to, args->connect_uri, NULL, "{}");
951     wait_for_migration_complete(from);
952 
953     /*
954      * We need to wait for the source to finish before starting the
955      * destination.
956      */
957     migrate_incoming_qmp(to, args->connect_uri, NULL, "{}");
958     wait_for_migration_complete(to);
959 
960     if (stop_src) {
961         qtest_qmp_assert_success(to, "{ 'execute' : 'cont'}");
962     }
963     wait_for_resume(to, &dst_state);
964 
965     wait_for_serial("dest_serial");
966 
967     if (check_offset) {
968         file_check_offset_region();
969     }
970 
971 finish:
972     if (args->end_hook) {
973         args->end_hook(from, to, data_hook);
974     }
975 
976     migrate_end(from, to, args->result == MIG_TEST_SUCCEED);
977 }
978 
migrate_hook_start_precopy_tcp_multifd_common(QTestState * from,QTestState * to,const char * method)979 void *migrate_hook_start_precopy_tcp_multifd_common(QTestState *from,
980                                                     QTestState *to,
981                                                     const char *method)
982 {
983     migrate_set_parameter_str(from, "multifd-compression", method);
984     migrate_set_parameter_str(to, "multifd-compression", method);
985 
986     /* Start incoming migration from the 1st socket */
987     migrate_incoming_qmp(to, "tcp:127.0.0.1:0", NULL, "{}");
988 
989     return NULL;
990 }
991 
get_src(void)992 QTestMigrationState *get_src(void)
993 {
994     return &src_state;
995 }
996 
migration_get_env(void)997 MigrationTestEnv *migration_get_env(void)
998 {
999     static MigrationTestEnv *env;
1000     g_autoptr(GError) err = NULL;
1001 
1002     if (env) {
1003         return env;
1004     }
1005 
1006     env = g_new0(MigrationTestEnv, 1);
1007     env->qemu_src = getenv(QEMU_ENV_SRC);
1008     env->qemu_dst = getenv(QEMU_ENV_DST);
1009 
1010     /*
1011      * The default QTEST_QEMU_BINARY must always be provided because
1012      * that is what helpers use to query the accel type and
1013      * architecture.
1014      */
1015     if (env->qemu_src && env->qemu_dst) {
1016         g_test_message("Only one of %s, %s is allowed",
1017                        QEMU_ENV_SRC, QEMU_ENV_DST);
1018         exit(1);
1019     }
1020 
1021     env->has_kvm = qtest_has_accel("kvm");
1022     env->has_tcg = qtest_has_accel("tcg");
1023 
1024     if (!env->has_tcg && !env->has_kvm) {
1025         g_test_skip("No KVM or TCG accelerator available");
1026         return env;
1027     }
1028 
1029     env->has_dirty_ring = kvm_dirty_ring_supported();
1030     env->has_uffd = ufd_version_check(&env->uffd_feature_thread_id);
1031     env->arch = qtest_get_arch();
1032     env->is_x86 = !strcmp(env->arch, "i386") || !strcmp(env->arch, "x86_64");
1033 
1034     env->tmpfs = g_dir_make_tmp("migration-test-XXXXXX", &err);
1035     if (!env->tmpfs) {
1036         g_test_message("Can't create temporary directory in %s: %s",
1037                        g_get_tmp_dir(), err->message);
1038     }
1039     g_assert(env->tmpfs);
1040 
1041     tmpfs = env->tmpfs;
1042 
1043     return env;
1044 }
1045 
migration_env_clean(MigrationTestEnv * env)1046 int migration_env_clean(MigrationTestEnv *env)
1047 {
1048     char *tmpfs;
1049     int ret = 0;
1050 
1051     if (!env) {
1052         return ret;
1053     }
1054 
1055     bootfile_delete();
1056 
1057     tmpfs = env->tmpfs;
1058     ret = rmdir(tmpfs);
1059     if (ret != 0) {
1060         g_test_message("unable to rmdir: path (%s): %s",
1061                        tmpfs, strerror(errno));
1062     }
1063     g_free(tmpfs);
1064 
1065     return ret;
1066 }
1067