2 * QTest testcase for migration
4 * Copyright (c) 2016-2018 Red Hat, Inc. and/or its affiliates
5 * based on the vhost-user-test.c that is:
6 * Copyright (c) 2014 Virtual Open Systems Sarl.
8 * This work is licensed under the terms of the GNU GPL, version 2 or later.
9 * See the COPYING file in the top-level directory.
13 #include "qemu/osdep.h"
16 #include "qapi/qmp/qdict.h"
17 #include "qemu/option.h"
18 #include "qemu/range.h"
19 #include "qemu/sockets.h"
20 #include "chardev/char.h"
21 #include "sysemu/sysemu.h"
23 const unsigned start_address
= 1024 * 1024;
24 const unsigned end_address
= 100 * 1024 * 1024;
26 static bool uffd_feature_thread_id
;
28 #if defined(__linux__)
29 #include <sys/syscall.h>
33 #if defined(__linux__) && defined(__NR_userfaultfd) && defined(CONFIG_EVENTFD)
34 #include <sys/eventfd.h>
35 #include <sys/ioctl.h>
36 #include <linux/userfaultfd.h>
38 static bool ufd_version_check(void)
40 struct uffdio_api api_struct
;
43 int ufd
= syscall(__NR_userfaultfd
, O_CLOEXEC
);
46 g_test_message("Skipping test: userfaultfd not available");
50 api_struct
.api
= UFFD_API
;
51 api_struct
.features
= 0;
52 if (ioctl(ufd
, UFFDIO_API
, &api_struct
)) {
53 g_test_message("Skipping test: UFFDIO_API failed");
56 uffd_feature_thread_id
= api_struct
.features
& UFFD_FEATURE_THREAD_ID
;
58 ioctl_mask
= (__u64
)1 << _UFFDIO_REGISTER
|
59 (__u64
)1 << _UFFDIO_UNREGISTER
;
60 if ((api_struct
.ioctls
& ioctl_mask
) != ioctl_mask
) {
61 g_test_message("Skipping test: Missing userfault feature");
69 static bool ufd_version_check(void)
71 g_test_message("Skipping test: Userfault not available (builtdtime)");
77 static const char *tmpfs
;
79 /* A simple PC boot sector that modifies memory (1-100MB) quickly
80 * outputting a 'B' every so often if it's still running.
82 #include "tests/migration/x86-a-b-bootblock.h"
84 static void init_bootfile_x86(const char *bootpath
)
86 FILE *bootfile
= fopen(bootpath
, "wb");
88 g_assert_cmpint(fwrite(x86_bootsect
, 512, 1, bootfile
), ==, 1);
93 * Wait for some output in the serial output file,
94 * we get an 'A' followed by an endless string of 'B's
95 * but on the destination we won't have the A.
97 static void wait_for_serial(const char *side
)
99 char *serialpath
= g_strdup_printf("%s/%s", tmpfs
, side
);
100 FILE *serialfile
= fopen(serialpath
, "r");
101 const char *arch
= qtest_get_arch();
102 int started
= (strcmp(side
, "src_serial") == 0 &&
103 strcmp(arch
, "ppc64") == 0) ? 0 : 1;
107 int readvalue
= fgetc(serialfile
);
110 /* SLOF prints its banner before starting test,
111 * to ignore it, mark the start of the test with '_',
112 * ignore all characters until this marker
119 fseek(serialfile
, 0, SEEK_SET
);
136 started
= (strcmp(side
, "src_serial") == 0 &&
137 strcmp(arch
, "ppc64") == 0) ? 0 : 1;
138 fseek(serialfile
, 0, SEEK_SET
);
143 fprintf(stderr
, "Unexpected %d on %s serial\n", readvalue
, side
);
144 g_assert_not_reached();
150 * Events can get in the way of responses we are actually waiting for.
152 static QDict
*wait_command(QTestState
*who
, const char *command
)
154 const char *event_string
;
157 response
= qtest_qmp(who
, command
);
159 while (qdict_haskey(response
, "event")) {
160 /* OK, it was an event */
161 event_string
= qdict_get_str(response
, "event");
162 if (!strcmp(event_string
, "STOP")) {
165 qobject_unref(response
);
166 response
= qtest_qmp_receive(who
);
172 * Note: caller is responsible to free the returned object via
173 * qobject_unref() after use
175 static QDict
*migrate_query(QTestState
*who
)
177 QDict
*rsp
, *rsp_return
;
179 rsp
= wait_command(who
, "{ 'execute': 'query-migrate' }");
180 rsp_return
= qdict_get_qdict(rsp
, "return");
181 g_assert(rsp_return
);
182 qobject_ref(rsp_return
);
189 * Note: caller is responsible to free the returned object via
192 static gchar
*migrate_query_status(QTestState
*who
)
194 QDict
*rsp_return
= migrate_query(who
);
195 gchar
*status
= g_strdup(qdict_get_str(rsp_return
, "status"));
198 qobject_unref(rsp_return
);
204 * It's tricky to use qemu's migration event capability with qtest,
205 * events suddenly appearing confuse the qmp()/hmp() responses.
208 static uint64_t get_migration_pass(QTestState
*who
)
210 QDict
*rsp_return
, *rsp_ram
;
213 rsp_return
= migrate_query(who
);
214 if (!qdict_haskey(rsp_return
, "ram")) {
218 rsp_ram
= qdict_get_qdict(rsp_return
, "ram");
219 result
= qdict_get_try_int(rsp_ram
, "dirty-sync-count", 0);
221 qobject_unref(rsp_return
);
225 static void read_blocktime(QTestState
*who
)
229 rsp_return
= migrate_query(who
);
230 g_assert(qdict_haskey(rsp_return
, "postcopy-blocktime"));
231 qobject_unref(rsp_return
);
234 static void wait_for_migration_status(QTestState
*who
,
241 status
= migrate_query_status(who
);
242 completed
= strcmp(status
, goal
) == 0;
243 g_assert_cmpstr(status
, !=, "failed");
252 static void wait_for_migration_complete(QTestState
*who
)
254 wait_for_migration_status(who
, "completed");
257 static void wait_for_migration_pass(QTestState
*who
)
259 uint64_t initial_pass
= get_migration_pass(who
);
262 /* Wait for the 1st sync */
263 while (!got_stop
&& !initial_pass
) {
265 initial_pass
= get_migration_pass(who
);
270 pass
= get_migration_pass(who
);
271 } while (pass
== initial_pass
&& !got_stop
);
274 static void check_guests_ram(QTestState
*who
)
276 /* Our ASM test will have been incrementing one byte from each page from
277 * 1MB to <100MB in order.
278 * This gives us a constraint that any page's byte should be equal or less
279 * than the previous pages byte (mod 256); and they should all be equal
280 * except for one transition at the point where we meet the incrementer.
281 * (We're running this with the guest stopped).
286 bool hit_edge
= false;
289 qtest_memread(who
, start_address
, &first_byte
, 1);
290 last_byte
= first_byte
;
292 for (address
= start_address
+ 4096; address
< end_address
; address
+= 4096)
295 qtest_memread(who
, address
, &b
, 1);
296 if (b
!= last_byte
) {
297 if (((b
+ 1) % 256) == last_byte
&& !hit_edge
) {
298 /* This is OK, the guest stopped at the point of
299 * incrementing the previous page but didn't get
304 fprintf(stderr
, "Memory content inconsistency at %x"
305 " first_byte = %x last_byte = %x current = %x"
307 address
, first_byte
, last_byte
, b
, hit_edge
);
316 static void cleanup(const char *filename
)
318 char *path
= g_strdup_printf("%s/%s", tmpfs
, filename
);
324 static void migrate_check_parameter(QTestState
*who
, const char *parameter
,
327 QDict
*rsp
, *rsp_return
;
330 rsp
= wait_command(who
, "{ 'execute': 'query-migrate-parameters' }");
331 rsp_return
= qdict_get_qdict(rsp
, "return");
332 result
= g_strdup_printf("%" PRId64
,
333 qdict_get_try_int(rsp_return
, parameter
, -1));
334 g_assert_cmpstr(result
, ==, value
);
339 static void migrate_set_parameter(QTestState
*who
, const char *parameter
,
345 cmd
= g_strdup_printf("{ 'execute': 'migrate-set-parameters',"
346 "'arguments': { '%s': %s } }",
348 rsp
= qtest_qmp(who
, cmd
);
350 g_assert(qdict_haskey(rsp
, "return"));
352 migrate_check_parameter(who
, parameter
, value
);
355 static void migrate_pause(QTestState
*who
)
359 rsp
= wait_command(who
, "{ 'execute': 'migrate-pause' }");
360 g_assert(qdict_haskey(rsp
, "return"));
364 static void migrate_recover(QTestState
*who
, const char *uri
)
367 gchar
*cmd
= g_strdup_printf(
368 "{ 'execute': 'migrate-recover', "
369 " 'id': 'recover-cmd', "
370 " 'arguments': { 'uri': '%s' } }", uri
);
372 rsp
= wait_command(who
, cmd
);
373 g_assert(qdict_haskey(rsp
, "return"));
378 static void migrate_set_capability(QTestState
*who
, const char *capability
,
384 cmd
= g_strdup_printf("{ 'execute': 'migrate-set-capabilities',"
386 "'capabilities': [ { "
387 "'capability': '%s', 'state': %s } ] } }",
389 rsp
= qtest_qmp(who
, cmd
);
391 g_assert(qdict_haskey(rsp
, "return"));
395 static void migrate(QTestState
*who
, const char *uri
, const char *extra
)
400 cmd
= g_strdup_printf("{ 'execute': 'migrate',"
401 " 'arguments': { 'uri': '%s' %s } }",
402 uri
, extra
? extra
: "");
403 rsp
= qtest_qmp(who
, cmd
);
405 g_assert(qdict_haskey(rsp
, "return"));
409 static void migrate_postcopy_start(QTestState
*from
, QTestState
*to
)
413 rsp
= wait_command(from
, "{ 'execute': 'migrate-start-postcopy' }");
414 g_assert(qdict_haskey(rsp
, "return"));
418 qtest_qmp_eventwait(from
, "STOP");
421 qtest_qmp_eventwait(to
, "RESUME");
424 static int test_migrate_start(QTestState
**from
, QTestState
**to
,
425 const char *uri
, bool hide_stderr
)
427 gchar
*cmd_src
, *cmd_dst
;
428 char *bootpath
= g_strdup_printf("%s/bootsect", tmpfs
);
429 const char *arch
= qtest_get_arch();
430 const char *accel
= "kvm:tcg";
434 if (strcmp(arch
, "i386") == 0 || strcmp(arch
, "x86_64") == 0) {
435 init_bootfile_x86(bootpath
);
436 cmd_src
= g_strdup_printf("-machine accel=%s -m 150M"
437 " -name source,debug-threads=on"
438 " -serial file:%s/src_serial"
439 " -drive file=%s,format=raw",
440 accel
, tmpfs
, bootpath
);
441 cmd_dst
= g_strdup_printf("-machine accel=%s -m 150M"
442 " -name target,debug-threads=on"
443 " -serial file:%s/dest_serial"
444 " -drive file=%s,format=raw"
446 accel
, tmpfs
, bootpath
, uri
);
447 } else if (strcmp(arch
, "ppc64") == 0) {
449 /* On ppc64, the test only works with kvm-hv, but not with kvm-pr
450 * and TCG is touchy due to race conditions on dirty bits
451 * (especially on PPC for some reason)
453 if (access("/sys/module/kvm_hv", F_OK
)) {
454 g_print("Skipping test: kvm_hv not available ");
457 cmd_src
= g_strdup_printf("-machine accel=%s -m 256M"
458 " -name source,debug-threads=on"
459 " -serial file:%s/src_serial"
461 "boot-command=hex .\" _\" begin %x %x "
462 "do i c@ 1 + i c! 1000 +loop .\" B\" 0 "
463 "until'", accel
, tmpfs
, end_address
,
465 cmd_dst
= g_strdup_printf("-machine accel=%s -m 256M"
466 " -name target,debug-threads=on"
467 " -serial file:%s/dest_serial"
471 g_assert_not_reached();
478 tmp
= g_strdup_printf("%s 2>/dev/null", cmd_src
);
482 tmp
= g_strdup_printf("%s 2>/dev/null", cmd_dst
);
487 *from
= qtest_start(cmd_src
);
490 *to
= qtest_init(cmd_dst
);
495 static void test_migrate_end(QTestState
*from
, QTestState
*to
, bool test_dest
)
497 unsigned char dest_byte_a
, dest_byte_b
, dest_byte_c
, dest_byte_d
;
502 qtest_memread(to
, start_address
, &dest_byte_a
, 1);
504 /* Destination still running, wait for a byte to change */
506 qtest_memread(to
, start_address
, &dest_byte_b
, 1);
508 } while (dest_byte_a
== dest_byte_b
);
510 qtest_qmp_discard_response(to
, "{ 'execute' : 'stop'}");
512 /* With it stopped, check nothing changes */
513 qtest_memread(to
, start_address
, &dest_byte_c
, 1);
515 qtest_memread(to
, start_address
, &dest_byte_d
, 1);
516 g_assert_cmpint(dest_byte_c
, ==, dest_byte_d
);
518 check_guests_ram(to
);
524 cleanup("migsocket");
525 cleanup("src_serial");
526 cleanup("dest_serial");
529 static void deprecated_set_downtime(QTestState
*who
, const double value
)
536 cmd
= g_strdup_printf("{ 'execute': 'migrate_set_downtime',"
537 "'arguments': { 'value': %g } }", value
);
538 rsp
= qtest_qmp(who
, cmd
);
540 g_assert(qdict_haskey(rsp
, "return"));
542 result_int
= value
* 1000L;
543 expected
= g_strdup_printf("%" PRId64
, result_int
);
544 migrate_check_parameter(who
, "downtime-limit", expected
);
548 static void deprecated_set_speed(QTestState
*who
, const char *value
)
553 cmd
= g_strdup_printf("{ 'execute': 'migrate_set_speed',"
554 "'arguments': { 'value': %s } }", value
);
555 rsp
= qtest_qmp(who
, cmd
);
557 g_assert(qdict_haskey(rsp
, "return"));
559 migrate_check_parameter(who
, "max-bandwidth", value
);
562 static void test_deprecated(void)
566 from
= qtest_start("");
568 deprecated_set_downtime(from
, 0.12345);
569 deprecated_set_speed(from
, "12345");
574 static int migrate_postcopy_prepare(QTestState
**from_ptr
,
578 char *uri
= g_strdup_printf("unix:%s/migsocket", tmpfs
);
579 QTestState
*from
, *to
;
581 if (test_migrate_start(&from
, &to
, uri
, hide_error
)) {
585 migrate_set_capability(from
, "postcopy-ram", "true");
586 migrate_set_capability(to
, "postcopy-ram", "true");
587 migrate_set_capability(to
, "postcopy-blocktime", "true");
589 /* We want to pick a speed slow enough that the test completes
590 * quickly, but that it doesn't complete precopy even on a slow
591 * machine, so also set the downtime.
593 migrate_set_parameter(from
, "max-bandwidth", "100000000");
594 migrate_set_parameter(from
, "downtime-limit", "1");
596 /* Wait for the first serial output from the source */
597 wait_for_serial("src_serial");
599 migrate(from
, uri
, NULL
);
602 wait_for_migration_pass(from
);
610 static void migrate_postcopy_complete(QTestState
*from
, QTestState
*to
)
612 wait_for_migration_complete(from
);
614 /* Make sure we get at least one "B" on destination */
615 wait_for_serial("dest_serial");
617 if (uffd_feature_thread_id
) {
621 test_migrate_end(from
, to
, true);
624 static void test_postcopy(void)
626 QTestState
*from
, *to
;
628 if (migrate_postcopy_prepare(&from
, &to
, false)) {
631 migrate_postcopy_start(from
, to
);
632 migrate_postcopy_complete(from
, to
);
635 static void test_postcopy_recovery(void)
637 QTestState
*from
, *to
;
640 if (migrate_postcopy_prepare(&from
, &to
, true)) {
644 /* Turn postcopy speed down, 4K/s is slow enough on any machines */
645 migrate_set_parameter(from
, "max-postcopy-bandwidth", "4096");
647 /* Now we start the postcopy */
648 migrate_postcopy_start(from
, to
);
651 * Wait until postcopy is really started; we can only run the
652 * migrate-pause command during a postcopy
654 wait_for_migration_status(from
, "postcopy-active");
657 * Manually stop the postcopy migration. This emulates a network
658 * failure with the migration socket
663 * Wait for destination side to reach postcopy-paused state. The
664 * migrate-recover command can only succeed if destination machine
665 * is in the paused state
667 wait_for_migration_status(to
, "postcopy-paused");
670 * Create a new socket to emulate a new channel that is different
671 * from the broken migration channel; tell the destination to
672 * listen to the new port
674 uri
= g_strdup_printf("unix:%s/migsocket-recover", tmpfs
);
675 migrate_recover(to
, uri
);
678 * Try to rebuild the migration channel using the resume flag and
679 * the newly created channel
681 wait_for_migration_status(from
, "postcopy-paused");
682 migrate(from
, uri
, ", 'resume': true");
685 /* Restore the postcopy bandwidth to unlimited */
686 migrate_set_parameter(from
, "max-postcopy-bandwidth", "0");
688 migrate_postcopy_complete(from
, to
);
691 static void test_baddest(void)
693 QTestState
*from
, *to
;
694 QDict
*rsp
, *rsp_return
;
698 if (test_migrate_start(&from
, &to
, "tcp:0:0", true)) {
701 migrate(from
, "tcp:0:0", NULL
);
703 status
= migrate_query_status(from
);
704 g_assert(!strcmp(status
, "setup") || !(strcmp(status
, "failed")));
705 failed
= !strcmp(status
, "failed");
709 /* Is the machine currently running? */
710 rsp
= wait_command(from
, "{ 'execute': 'query-status' }");
711 g_assert(qdict_haskey(rsp
, "return"));
712 rsp_return
= qdict_get_qdict(rsp
, "return");
713 g_assert(qdict_haskey(rsp_return
, "running"));
714 g_assert(qdict_get_bool(rsp_return
, "running"));
717 test_migrate_end(from
, to
, false);
720 static void test_precopy_unix(void)
722 char *uri
= g_strdup_printf("unix:%s/migsocket", tmpfs
);
723 QTestState
*from
, *to
;
725 if (test_migrate_start(&from
, &to
, uri
, false)) {
729 /* We want to pick a speed slow enough that the test completes
730 * quickly, but that it doesn't complete precopy even on a slow
731 * machine, so also set the downtime.
733 /* 1 ms should make it not converge*/
734 migrate_set_parameter(from
, "downtime-limit", "1");
736 migrate_set_parameter(from
, "max-bandwidth", "1000000000");
738 /* Wait for the first serial output from the source */
739 wait_for_serial("src_serial");
741 migrate(from
, uri
, NULL
);
743 wait_for_migration_pass(from
);
745 /* 300 ms should converge */
746 migrate_set_parameter(from
, "downtime-limit", "300");
749 qtest_qmp_eventwait(from
, "STOP");
752 qtest_qmp_eventwait(to
, "RESUME");
754 wait_for_serial("dest_serial");
755 wait_for_migration_complete(from
);
757 test_migrate_end(from
, to
, true);
761 int main(int argc
, char **argv
)
763 char template[] = "/tmp/migration-test-XXXXXX";
766 g_test_init(&argc
, &argv
, NULL
);
768 if (!ufd_version_check()) {
772 tmpfs
= mkdtemp(template);
774 g_test_message("mkdtemp on path (%s): %s\n", template, strerror(errno
));
778 module_call_init(MODULE_INIT_QOM
);
780 qtest_add_func("/migration/postcopy/unix", test_postcopy
);
781 qtest_add_func("/migration/postcopy/recovery", test_postcopy_recovery
);
782 qtest_add_func("/migration/deprecated", test_deprecated
);
783 qtest_add_func("/migration/bad_dest", test_baddest
);
784 qtest_add_func("/migration/precopy/unix", test_precopy_unix
);
788 g_assert_cmpint(ret
, ==, 0);
792 g_test_message("unable to rmdir: path (%s): %s\n",
793 tmpfs
, strerror(errno
));