]>
git.proxmox.com Git - mirror_qemu.git/blob - tests/migration-test.c
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 "qapi/qmp/qjson.h"
18 #include "qemu/option.h"
19 #include "qemu/range.h"
20 #include "qemu/sockets.h"
21 #include "chardev/char.h"
22 #include "sysemu/sysemu.h"
24 /* TODO actually test the results and get rid of this */
25 #define qtest_qmp_discard_response(...) qobject_unref(qtest_qmp(__VA_ARGS__))
27 const unsigned start_address
= 1024 * 1024;
28 const unsigned end_address
= 100 * 1024 * 1024;
30 static bool uffd_feature_thread_id
;
32 #if defined(__linux__)
33 #include <sys/syscall.h>
37 #if defined(__linux__) && defined(__NR_userfaultfd) && defined(CONFIG_EVENTFD)
38 #include <sys/eventfd.h>
39 #include <sys/ioctl.h>
40 #include <linux/userfaultfd.h>
42 static bool ufd_version_check(void)
44 struct uffdio_api api_struct
;
47 int ufd
= syscall(__NR_userfaultfd
, O_CLOEXEC
);
50 g_test_message("Skipping test: userfaultfd not available");
54 api_struct
.api
= UFFD_API
;
55 api_struct
.features
= 0;
56 if (ioctl(ufd
, UFFDIO_API
, &api_struct
)) {
57 g_test_message("Skipping test: UFFDIO_API failed");
60 uffd_feature_thread_id
= api_struct
.features
& UFFD_FEATURE_THREAD_ID
;
62 ioctl_mask
= (__u64
)1 << _UFFDIO_REGISTER
|
63 (__u64
)1 << _UFFDIO_UNREGISTER
;
64 if ((api_struct
.ioctls
& ioctl_mask
) != ioctl_mask
) {
65 g_test_message("Skipping test: Missing userfault feature");
73 static bool ufd_version_check(void)
75 g_test_message("Skipping test: Userfault not available (builtdtime)");
81 static const char *tmpfs
;
83 /* A simple PC boot sector that modifies memory (1-100MB) quickly
84 * outputting a 'B' every so often if it's still running.
86 #include "tests/migration/x86-a-b-bootblock.h"
88 static void init_bootfile_x86(const char *bootpath
)
90 FILE *bootfile
= fopen(bootpath
, "wb");
92 g_assert_cmpint(fwrite(x86_bootsect
, 512, 1, bootfile
), ==, 1);
97 * Wait for some output in the serial output file,
98 * we get an 'A' followed by an endless string of 'B's
99 * but on the destination we won't have the A.
101 static void wait_for_serial(const char *side
)
103 char *serialpath
= g_strdup_printf("%s/%s", tmpfs
, side
);
104 FILE *serialfile
= fopen(serialpath
, "r");
105 const char *arch
= qtest_get_arch();
106 int started
= (strcmp(side
, "src_serial") == 0 &&
107 strcmp(arch
, "ppc64") == 0) ? 0 : 1;
111 int readvalue
= fgetc(serialfile
);
114 /* SLOF prints its banner before starting test,
115 * to ignore it, mark the start of the test with '_',
116 * ignore all characters until this marker
123 fseek(serialfile
, 0, SEEK_SET
);
140 started
= (strcmp(side
, "src_serial") == 0 &&
141 strcmp(arch
, "ppc64") == 0) ? 0 : 1;
142 fseek(serialfile
, 0, SEEK_SET
);
147 fprintf(stderr
, "Unexpected %d on %s serial\n", readvalue
, side
);
148 g_assert_not_reached();
153 static void stop_cb(void *opaque
, const char *name
, QDict
*data
)
155 if (!strcmp(name
, "STOP")) {
161 * Events can get in the way of responses we are actually waiting for.
164 static QDict
*wait_command(QTestState
*who
, const char *command
, ...)
168 va_start(ap
, command
);
169 qtest_qmp_vsend(who
, command
, ap
);
172 return qtest_qmp_receive_success(who
, stop_cb
, NULL
);
176 * Note: caller is responsible to free the returned object via
177 * qobject_unref() after use
179 static QDict
*migrate_query(QTestState
*who
)
181 return wait_command(who
, "{ 'execute': 'query-migrate' }");
185 * Note: caller is responsible to free the returned object via
188 static gchar
*migrate_query_status(QTestState
*who
)
190 QDict
*rsp_return
= migrate_query(who
);
191 gchar
*status
= g_strdup(qdict_get_str(rsp_return
, "status"));
194 qobject_unref(rsp_return
);
200 * It's tricky to use qemu's migration event capability with qtest,
201 * events suddenly appearing confuse the qmp()/hmp() responses.
204 static uint64_t get_migration_pass(QTestState
*who
)
206 QDict
*rsp_return
, *rsp_ram
;
209 rsp_return
= migrate_query(who
);
210 if (!qdict_haskey(rsp_return
, "ram")) {
214 rsp_ram
= qdict_get_qdict(rsp_return
, "ram");
215 result
= qdict_get_try_int(rsp_ram
, "dirty-sync-count", 0);
217 qobject_unref(rsp_return
);
221 static void read_blocktime(QTestState
*who
)
225 rsp_return
= migrate_query(who
);
226 g_assert(qdict_haskey(rsp_return
, "postcopy-blocktime"));
227 qobject_unref(rsp_return
);
230 static void wait_for_migration_status(QTestState
*who
,
237 status
= migrate_query_status(who
);
238 completed
= strcmp(status
, goal
) == 0;
239 g_assert_cmpstr(status
, !=, "failed");
248 static void wait_for_migration_complete(QTestState
*who
)
250 wait_for_migration_status(who
, "completed");
253 static void wait_for_migration_pass(QTestState
*who
)
255 uint64_t initial_pass
= get_migration_pass(who
);
258 /* Wait for the 1st sync */
259 while (!got_stop
&& !initial_pass
) {
261 initial_pass
= get_migration_pass(who
);
266 pass
= get_migration_pass(who
);
267 } while (pass
== initial_pass
&& !got_stop
);
270 static void check_guests_ram(QTestState
*who
)
272 /* Our ASM test will have been incrementing one byte from each page from
273 * 1MB to <100MB in order.
274 * This gives us a constraint that any page's byte should be equal or less
275 * than the previous pages byte (mod 256); and they should all be equal
276 * except for one transition at the point where we meet the incrementer.
277 * (We're running this with the guest stopped).
282 bool hit_edge
= false;
285 qtest_memread(who
, start_address
, &first_byte
, 1);
286 last_byte
= first_byte
;
288 for (address
= start_address
+ 4096; address
< end_address
; address
+= 4096)
291 qtest_memread(who
, address
, &b
, 1);
292 if (b
!= last_byte
) {
293 if (((b
+ 1) % 256) == last_byte
&& !hit_edge
) {
294 /* This is OK, the guest stopped at the point of
295 * incrementing the previous page but didn't get
301 fprintf(stderr
, "Memory content inconsistency at %x"
302 " first_byte = %x last_byte = %x current = %x"
304 address
, first_byte
, last_byte
, b
, hit_edge
);
312 static void cleanup(const char *filename
)
314 char *path
= g_strdup_printf("%s/%s", tmpfs
, filename
);
320 static void migrate_check_parameter(QTestState
*who
, const char *parameter
,
326 rsp_return
= wait_command(who
,
327 "{ 'execute': 'query-migrate-parameters' }");
328 result
= g_strdup_printf("%" PRId64
,
329 qdict_get_try_int(rsp_return
, parameter
, -1));
330 g_assert_cmpstr(result
, ==, value
);
332 qobject_unref(rsp_return
);
335 static void migrate_set_parameter(QTestState
*who
, const char *parameter
,
341 cmd
= g_strdup_printf("{ 'execute': 'migrate-set-parameters',"
342 "'arguments': { '%s': %s } }",
344 rsp
= qtest_qmp(who
, cmd
);
346 g_assert(qdict_haskey(rsp
, "return"));
348 migrate_check_parameter(who
, parameter
, value
);
351 static void migrate_pause(QTestState
*who
)
355 rsp
= wait_command(who
, "{ 'execute': 'migrate-pause' }");
359 static void migrate_recover(QTestState
*who
, const char *uri
)
363 rsp
= wait_command(who
,
364 "{ 'execute': 'migrate-recover', "
365 " 'id': 'recover-cmd', "
366 " 'arguments': { 'uri': %s } }",
371 static void migrate_set_capability(QTestState
*who
, const char *capability
,
377 cmd
= g_strdup_printf("{ 'execute': 'migrate-set-capabilities',"
379 "'capabilities': [ { "
380 "'capability': '%s', 'state': %s } ] } }",
382 rsp
= qtest_qmp(who
, cmd
);
384 g_assert(qdict_haskey(rsp
, "return"));
389 * Send QMP command "migrate".
390 * Arguments are built from @fmt... (formatted like
391 * qobject_from_jsonf_nofail()) with "uri": @uri spliced in.
394 static void migrate(QTestState
*who
, const char *uri
, const char *fmt
, ...)
400 args
= qdict_from_vjsonf_nofail(fmt
, ap
);
403 g_assert(!qdict_haskey(args
, "uri"));
404 qdict_put_str(args
, "uri", uri
);
406 rsp
= qmp("{ 'execute': 'migrate', 'arguments': %p}", args
);
407 g_assert(qdict_haskey(rsp
, "return"));
411 static void migrate_postcopy_start(QTestState
*from
, QTestState
*to
)
415 rsp
= wait_command(from
, "{ 'execute': 'migrate-start-postcopy' }");
419 qtest_qmp_eventwait(from
, "STOP");
422 qtest_qmp_eventwait(to
, "RESUME");
425 static int test_migrate_start(QTestState
**from
, QTestState
**to
,
426 const char *uri
, bool hide_stderr
)
428 gchar
*cmd_src
, *cmd_dst
;
429 char *bootpath
= g_strdup_printf("%s/bootsect", tmpfs
);
430 const char *arch
= qtest_get_arch();
431 const char *accel
= "kvm:tcg";
435 if (strcmp(arch
, "i386") == 0 || strcmp(arch
, "x86_64") == 0) {
436 init_bootfile_x86(bootpath
);
437 cmd_src
= g_strdup_printf("-machine accel=%s -m 150M"
438 " -name source,debug-threads=on"
439 " -serial file:%s/src_serial"
440 " -drive file=%s,format=raw",
441 accel
, tmpfs
, bootpath
);
442 cmd_dst
= g_strdup_printf("-machine accel=%s -m 150M"
443 " -name target,debug-threads=on"
444 " -serial file:%s/dest_serial"
445 " -drive file=%s,format=raw"
447 accel
, tmpfs
, bootpath
, uri
);
448 } else if (strcmp(arch
, "ppc64") == 0) {
450 /* On ppc64, the test only works with kvm-hv, but not with kvm-pr
451 * and TCG is touchy due to race conditions on dirty bits
452 * (especially on PPC for some reason)
454 if (access("/sys/module/kvm_hv", F_OK
)) {
455 g_print("Skipping test: kvm_hv not available ");
458 cmd_src
= g_strdup_printf("-machine accel=%s -m 256M"
459 " -name source,debug-threads=on"
460 " -serial file:%s/src_serial"
462 "boot-command=hex .\" _\" begin %x %x "
463 "do i c@ 1 + i c! 1000 +loop .\" B\" 0 "
464 "until'", accel
, tmpfs
, end_address
,
466 cmd_dst
= g_strdup_printf("-machine accel=%s -m 256M"
467 " -name target,debug-threads=on"
468 " -serial file:%s/dest_serial"
472 g_assert_not_reached();
479 tmp
= g_strdup_printf("%s 2>/dev/null", cmd_src
);
483 tmp
= g_strdup_printf("%s 2>/dev/null", cmd_dst
);
488 *from
= qtest_start(cmd_src
);
491 *to
= qtest_init(cmd_dst
);
496 static void test_migrate_end(QTestState
*from
, QTestState
*to
, bool test_dest
)
498 unsigned char dest_byte_a
, dest_byte_b
, dest_byte_c
, dest_byte_d
;
503 qtest_memread(to
, start_address
, &dest_byte_a
, 1);
505 /* Destination still running, wait for a byte to change */
507 qtest_memread(to
, start_address
, &dest_byte_b
, 1);
509 } while (dest_byte_a
== dest_byte_b
);
511 qtest_qmp_discard_response(to
, "{ 'execute' : 'stop'}");
513 /* With it stopped, check nothing changes */
514 qtest_memread(to
, start_address
, &dest_byte_c
, 1);
516 qtest_memread(to
, start_address
, &dest_byte_d
, 1);
517 g_assert_cmpint(dest_byte_c
, ==, dest_byte_d
);
519 check_guests_ram(to
);
525 cleanup("migsocket");
526 cleanup("src_serial");
527 cleanup("dest_serial");
530 static void deprecated_set_downtime(QTestState
*who
, const double value
)
537 "{ 'execute': 'migrate_set_downtime',"
538 " 'arguments': { 'value': %f } }", value
);
539 g_assert(qdict_haskey(rsp
, "return"));
541 result_int
= value
* 1000L;
542 expected
= g_strdup_printf("%" PRId64
, result_int
);
543 migrate_check_parameter(who
, "downtime-limit", expected
);
547 static void deprecated_set_speed(QTestState
*who
, const char *value
)
552 cmd
= g_strdup_printf("{ 'execute': 'migrate_set_speed',"
553 "'arguments': { 'value': %s } }", value
);
554 rsp
= qtest_qmp(who
, cmd
);
556 g_assert(qdict_haskey(rsp
, "return"));
558 migrate_check_parameter(who
, "max-bandwidth", value
);
561 static void test_deprecated(void)
565 from
= qtest_start("");
567 deprecated_set_downtime(from
, 0.12345);
568 deprecated_set_speed(from
, "12345");
573 static int migrate_postcopy_prepare(QTestState
**from_ptr
,
577 char *uri
= g_strdup_printf("unix:%s/migsocket", tmpfs
);
578 QTestState
*from
, *to
;
580 if (test_migrate_start(&from
, &to
, uri
, hide_error
)) {
584 migrate_set_capability(from
, "postcopy-ram", "true");
585 migrate_set_capability(to
, "postcopy-ram", "true");
586 migrate_set_capability(to
, "postcopy-blocktime", "true");
588 /* We want to pick a speed slow enough that the test completes
589 * quickly, but that it doesn't complete precopy even on a slow
590 * machine, so also set the downtime.
592 migrate_set_parameter(from
, "max-bandwidth", "100000000");
593 migrate_set_parameter(from
, "downtime-limit", "1");
595 /* Wait for the first serial output from the source */
596 wait_for_serial("src_serial");
598 migrate(from
, uri
, "{}");
601 wait_for_migration_pass(from
);
609 static void migrate_postcopy_complete(QTestState
*from
, QTestState
*to
)
611 wait_for_migration_complete(from
);
613 /* Make sure we get at least one "B" on destination */
614 wait_for_serial("dest_serial");
616 if (uffd_feature_thread_id
) {
620 test_migrate_end(from
, to
, true);
623 static void test_postcopy(void)
625 QTestState
*from
, *to
;
627 if (migrate_postcopy_prepare(&from
, &to
, false)) {
630 migrate_postcopy_start(from
, to
);
631 migrate_postcopy_complete(from
, to
);
634 static void test_postcopy_recovery(void)
636 QTestState
*from
, *to
;
639 if (migrate_postcopy_prepare(&from
, &to
, true)) {
643 /* Turn postcopy speed down, 4K/s is slow enough on any machines */
644 migrate_set_parameter(from
, "max-postcopy-bandwidth", "4096");
646 /* Now we start the postcopy */
647 migrate_postcopy_start(from
, to
);
650 * Wait until postcopy is really started; we can only run the
651 * migrate-pause command during a postcopy
653 wait_for_migration_status(from
, "postcopy-active");
656 * Manually stop the postcopy migration. This emulates a network
657 * failure with the migration socket
662 * Wait for destination side to reach postcopy-paused state. The
663 * migrate-recover command can only succeed if destination machine
664 * is in the paused state
666 wait_for_migration_status(to
, "postcopy-paused");
669 * Create a new socket to emulate a new channel that is different
670 * from the broken migration channel; tell the destination to
671 * listen to the new port
673 uri
= g_strdup_printf("unix:%s/migsocket-recover", tmpfs
);
674 migrate_recover(to
, uri
);
677 * Try to rebuild the migration channel using the resume flag and
678 * the newly created channel
680 wait_for_migration_status(from
, "postcopy-paused");
681 migrate(from
, uri
, "{'resume': true}");
684 /* Restore the postcopy bandwidth to unlimited */
685 migrate_set_parameter(from
, "max-postcopy-bandwidth", "0");
687 migrate_postcopy_complete(from
, to
);
690 static void test_baddest(void)
692 QTestState
*from
, *to
;
697 if (test_migrate_start(&from
, &to
, "tcp:0:0", true)) {
700 migrate(from
, "tcp:0:0", "{}");
702 status
= migrate_query_status(from
);
703 g_assert(!strcmp(status
, "setup") || !(strcmp(status
, "failed")));
704 failed
= !strcmp(status
, "failed");
708 /* Is the machine currently running? */
709 rsp_return
= wait_command(from
, "{ 'execute': 'query-status' }");
710 g_assert(qdict_haskey(rsp_return
, "running"));
711 g_assert(qdict_get_bool(rsp_return
, "running"));
712 qobject_unref(rsp_return
);
714 test_migrate_end(from
, to
, false);
717 static void test_precopy_unix(void)
719 char *uri
= g_strdup_printf("unix:%s/migsocket", tmpfs
);
720 QTestState
*from
, *to
;
722 if (test_migrate_start(&from
, &to
, uri
, false)) {
726 /* We want to pick a speed slow enough that the test completes
727 * quickly, but that it doesn't complete precopy even on a slow
728 * machine, so also set the downtime.
730 /* 1 ms should make it not converge*/
731 migrate_set_parameter(from
, "downtime-limit", "1");
733 migrate_set_parameter(from
, "max-bandwidth", "1000000000");
735 /* Wait for the first serial output from the source */
736 wait_for_serial("src_serial");
738 migrate(from
, uri
, "{}");
740 wait_for_migration_pass(from
);
742 /* 300 ms should converge */
743 migrate_set_parameter(from
, "downtime-limit", "300");
746 qtest_qmp_eventwait(from
, "STOP");
749 qtest_qmp_eventwait(to
, "RESUME");
751 wait_for_serial("dest_serial");
752 wait_for_migration_complete(from
);
754 test_migrate_end(from
, to
, true);
758 int main(int argc
, char **argv
)
760 char template[] = "/tmp/migration-test-XXXXXX";
763 g_test_init(&argc
, &argv
, NULL
);
765 if (!ufd_version_check()) {
769 tmpfs
= mkdtemp(template);
771 g_test_message("mkdtemp on path (%s): %s\n", template, strerror(errno
));
775 module_call_init(MODULE_INIT_QOM
);
777 qtest_add_func("/migration/postcopy/unix", test_postcopy
);
778 qtest_add_func("/migration/postcopy/recovery", test_postcopy_recovery
);
779 qtest_add_func("/migration/deprecated", test_deprecated
);
780 qtest_add_func("/migration/bad_dest", test_baddest
);
781 qtest_add_func("/migration/precopy/unix", test_precopy_unix
);
785 g_assert_cmpint(ret
, ==, 0);
789 g_test_message("unable to rmdir: path (%s): %s\n",
790 tmpfs
, strerror(errno
));