savevm.c · samitouri/QOSamiQemu · siGit Code & Deploy

Code Commits Issues Pulls Sessions CI/CD

master

c 3,808 lines 110 KB

Raw

1	/*
2	* QEMU System Emulator
3	*
4	* Copyright (c) 2003-2008 Fabrice Bellard
5	* Copyright (c) 2009-2015 Red Hat Inc
6	*
7	* Authors:
8	* Juan Quintela <quintela@redhat.com>
9	*
10	* Permission is hereby granted, free of charge, to any person obtaining a copy
11	* of this software and associated documentation files (the "Software"), to deal
12	* in the Software without restriction, including without limitation the rights
13	* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
14	* copies of the Software, and to permit persons to whom the Software is
15	* furnished to do so, subject to the following conditions:
16	*
17	* The above copyright notice and this permission notice shall be included in
18	* all copies or substantial portions of the Software.
19	*
20	* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
21	* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
22	* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
23	* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
24	* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
25	* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26	* THE SOFTWARE.
27	*/
28
29	#include "qemu/osdep.h"
30	#include "hw/core/boards.h"
31	#include "net/net.h"
32	#include "migration.h"
33	#include "migration/snapshot.h"
34	#include "migration-stats.h"
35	#include "migration/vmstate.h"
36	#include "migration/misc.h"
37	#include "migration/register.h"
38	#include "migration/global_state.h"
39	#include "migration/channel-block.h"
40	#include "multifd.h"
41	#include "ram.h"
42	#include "qemu-file.h"
43	#include "savevm.h"
44	#include "postcopy-ram.h"
45	#include "qapi/error.h"
46	#include "qapi/qapi-commands-migration.h"
47	#include "qapi/clone-visitor.h"
48	#include "qapi/qapi-builtin-visit.h"
49	#include "qemu/error-report.h"
50	#include "system/cpus.h"
51	#include "system/memory.h"
52	#include "exec/target_page.h"
53	#include "exec/page-vary.h"
54	#include "trace.h"
55	#include "qemu/iov.h"
56	#include "qemu/job.h"
57	#include "qemu/main-loop.h"
58	#include "block/snapshot.h"
59	#include "block/thread-pool.h"
60	#include "qemu/cutils.h"
61	#include "io/channel-buffer.h"
62	#include "io/channel-file.h"
63	#include "system/replay.h"
64	#include "system/runstate.h"
65	#include "system/system.h"
66	#include "system/xen.h"
67	#include "migration/colo.h"
68	#include "qemu/bitmap.h"
69	#include "net/announce.h"
70	#include "qemu/yank.h"
71	#include "yank_functions.h"
72	#include "system/qtest.h"
73	#include "options.h"
74
75	const unsigned int postcopy_ram_discard_version;
76
77	/* Subcommands for QEMU_VM_COMMAND */
78	enum qemu_vm_cmd {
79	MIG_CMD_INVALID = 0, /* Must be 0 */
80	MIG_CMD_OPEN_RETURN_PATH, /* Tell the dest to open the Return path */
81	MIG_CMD_PING, /* Request a PONG on the RP */
82
83	MIG_CMD_POSTCOPY_ADVISE, /* Prior to any page transfers, just
84	warn we might want to do PC */
85	MIG_CMD_POSTCOPY_LISTEN, /* Start listening for incoming
86	pages as it's running. */
87	MIG_CMD_POSTCOPY_RUN, /* Start execution */
88
89	MIG_CMD_POSTCOPY_RAM_DISCARD, /* A list of pages to discard that
90	were previously sent during
91	precopy but are dirty. */
92	MIG_CMD_PACKAGED, /* Send a wrapped stream within this stream */
93	MIG_CMD_DEPRECATED_0, /* Prior to 10.2, used as MIG_CMD_ENABLE_COLO */
94	MIG_CMD_POSTCOPY_RESUME, /* resume postcopy on dest */
95	MIG_CMD_RECV_BITMAP, /* Request for recved bitmap on dst */
96	MIG_CMD_SWITCHOVER_START, /* Switchover start notification */
97	MIG_CMD_MAX
98	};
99
100	#define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX
101	static struct mig_cmd_args {
102	ssize_t len; /* -1 = variable */
103	const char *name;
104	} mig_cmd_args[] = {
105	[MIG_CMD_INVALID] = { .len = -1, .name = "INVALID" },
106	[MIG_CMD_OPEN_RETURN_PATH] = { .len = 0, .name = "OPEN_RETURN_PATH" },
107	[MIG_CMD_PING] = { .len = sizeof(uint32_t), .name = "PING" },
108	[MIG_CMD_POSTCOPY_ADVISE] = { .len = -1, .name = "POSTCOPY_ADVISE" },
109	[MIG_CMD_POSTCOPY_LISTEN] = { .len = 0, .name = "POSTCOPY_LISTEN" },
110	[MIG_CMD_POSTCOPY_RUN] = { .len = 0, .name = "POSTCOPY_RUN" },
111	[MIG_CMD_POSTCOPY_RAM_DISCARD] = {
112	.len = -1, .name = "POSTCOPY_RAM_DISCARD" },
113	[MIG_CMD_POSTCOPY_RESUME] = { .len = 0, .name = "POSTCOPY_RESUME" },
114	[MIG_CMD_PACKAGED] = { .len = 4, .name = "PACKAGED" },
115	[MIG_CMD_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" },
116	[MIG_CMD_SWITCHOVER_START] = { .len = 0, .name = "SWITCHOVER_START" },
117	[MIG_CMD_MAX] = { .len = -1, .name = "MAX" },
118	};
119
120	/* Note for MIG_CMD_POSTCOPY_ADVISE:
121	* The format of arguments is depending on postcopy mode:
122	* - postcopy RAM only
123	* uint64_t host page size
124	* uint64_t target page size
125	*
126	* - postcopy RAM and postcopy dirty bitmaps
127	* format is the same as for postcopy RAM only
128	*
129	* - postcopy dirty bitmaps only
130	* Nothing. Command length field is 0.
131	*
132	* Be careful: adding a new postcopy entity with some other parameters should
133	* not break format self-description ability. Good way is to introduce some
134	* generic extendable format with an exception for two old entities.
135	*/
136
137	/***********************************************************/
138	/* Optional load threads pool support */
139
140	static void qemu_loadvm_thread_pool_create(MigrationIncomingState *mis)
141	{
142	assert(!mis->load_threads);
143	mis->load_threads = thread_pool_new();
144	mis->load_threads_abort = false;
145	}
146
147	static void qemu_loadvm_thread_pool_destroy(MigrationIncomingState *mis)
148	{
149	qatomic_set(&mis->load_threads_abort, true);
150
151	bql_unlock(); /* Load threads might be waiting for BQL */
152	g_clear_pointer(&mis->load_threads, thread_pool_free);
153	bql_lock();
154	}
155
156	static bool qemu_loadvm_thread_pool_wait(MigrationState *s,
157	MigrationIncomingState *mis)
158	{
159	bql_unlock(); /* Let load threads do work requiring BQL */
160	thread_pool_wait(mis->load_threads);
161	bql_lock();
162
163	return !migrate_has_error(s);
164	}
165
166	/***********************************************************/
167	/* savevm/loadvm support */
168
169	static QEMUFile qemu_fopen_bdrv(BlockDriverState bs, int is_writable)
170	{
171	if (is_writable) {
172	return qemu_file_new_output(QIO_CHANNEL(qio_channel_block_new(bs)));
173	} else {
174	return qemu_file_new_input(QIO_CHANNEL(qio_channel_block_new(bs)));
175	}
176	}
177
178
179	/* QEMUFile timer support.
180	* Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
181	*/
182
183	void timer_put(QEMUFile f, QEMUTimer ts)
184	{
185	uint64_t expire_time;
186
187	expire_time = timer_expire_time_ns(ts);
188	qemu_put_be64(f, expire_time);
189	}
190
191	void timer_get(QEMUFile f, QEMUTimer ts)
192	{
193	uint64_t expire_time;
194
195	expire_time = qemu_get_be64(f);
196	if (expire_time != -1) {
197	timer_mod_ns(ts, expire_time);
198	} else {
199	timer_del(ts);
200	}
201	}
202
203
204	/* VMState timer support.
205	* Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
206	*/
207
208	static bool load_timer(QEMUFile f, void pv, size_t size,
209	const VMStateField field, Error *errp)
210	{
211	QEMUTimer *v = pv;
212	timer_get(f, v);
213	return true;
214	}
215
216	static bool save_timer(QEMUFile f, void pv, size_t size,
217	const VMStateField field, JSONWriter vmdesc,
218	Error **errp)
219	{
220	QEMUTimer *v = pv;
221	timer_put(f, v);
222
223	return true;
224	}
225
226	const VMStateInfo vmstate_info_timer = {
227	.name = "timer",
228	.load = load_timer,
229	.save = save_timer,
230	};
231
232
233	typedef struct CompatEntry {
234	char idstr[256];
235	int instance_id;
236	} CompatEntry;
237
238	typedef struct SaveStateEntry {
239	QTAILQ_ENTRY(SaveStateEntry) entry;
240	char idstr[256];
241	uint32_t instance_id;
242	int alias_id;
243	int version_id;
244	/* version id read from the stream */
245	int load_version_id;
246	int section_id;
247	/* section id read from the stream */
248	int load_section_id;
249	const SaveVMHandlers *ops;
250	const VMStateDescription *vmsd;
251	void *opaque;
252	CompatEntry *compat;
253	} SaveStateEntry;
254
255	typedef struct SaveState {
256	QTAILQ_HEAD(, SaveStateEntry) handlers;
257	SaveStateEntry *handler_pri_head[MIG_PRI_MAX + 1];
258	int global_section_id;
259	uint32_t len;
260	const char *name;
261	uint32_t target_page_bits;
262	uint32_t caps_count;
263	MigrationCapability *capabilities;
264	QemuUUID uuid;
265	} SaveState;
266
267	static SaveState savevm_state = {
268	.handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
269	.handler_pri_head = { [0 ... MIG_PRI_MAX] = NULL },
270	.global_section_id = 0,
271	};
272
273	static SaveStateEntry find_se(const char idstr, uint32_t instance_id);
274
275	static bool should_validate_capability(int capability)
276	{
277	assert(capability >= 0 && capability < MIGRATION_CAPABILITY__MAX);
278	/* Validate only new capabilities to keep compatibility. */
279	switch (capability) {
280	case MIGRATION_CAPABILITY_X_IGNORE_SHARED:
281	case MIGRATION_CAPABILITY_MAPPED_RAM:
282	return true;
283	default:
284	return false;
285	}
286	}
287
288	static uint32_t get_validatable_capabilities_count(void)
289	{
290	MigrationState *s = migrate_get_current();
291	uint32_t result = 0;
292	int i;
293	for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
294	if (should_validate_capability(i) && s->capabilities[i]) {
295	result++;
296	}
297	}
298	return result;
299	}
300
301	static bool configuration_pre_save(void opaque, Error *errp)
302	{
303	SaveState *state = opaque;
304	const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
305	MigrationState *s = migrate_get_current();
306	int i, j;
307
308	state->len = strlen(current_name);
309	state->name = current_name;
310	state->target_page_bits = qemu_target_page_bits();
311
312	state->caps_count = get_validatable_capabilities_count();
313	state->capabilities = g_renew(MigrationCapability, state->capabilities,
314	state->caps_count);
315	for (i = j = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
316	if (should_validate_capability(i) && s->capabilities[i]) {
317	state->capabilities[j++] = i;
318	}
319	}
320	state->uuid = qemu_uuid;
321
322	return true;
323	}
324
325	static void configuration_post_save(void *opaque)
326	{
327	SaveState *state = opaque;
328
329	g_free(state->capabilities);
330	state->capabilities = NULL;
331	state->caps_count = 0;
332	}
333
334	static bool configuration_pre_load(void opaque, Error *errp)
335	{
336	SaveState *state = opaque;
337
338	/* If there is no target-page-bits subsection it means the source
339	* predates the variable-target-page-bits support and is using the
340	* minimum possible value for this CPU.
341	*/
342	state->target_page_bits = migration_legacy_page_bits();
343	return true;
344	}
345
346	static bool configuration_validate_capabilities(SaveState *state)
347	{
348	bool ret = true;
349	MigrationState *s = migrate_get_current();
350	DECLARE_BITMAP(source_caps_bm, MIGRATION_CAPABILITY__MAX);
351	int i;
352
353	bitmap_zero(source_caps_bm, MIGRATION_CAPABILITY__MAX);
354	for (i = 0; i < state->caps_count; i++) {
355	MigrationCapability capability = state->capabilities[i];
356	set_bit(capability, source_caps_bm);
357	}
358
359	for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
360	bool source_state, target_state;
361	if (!should_validate_capability(i)) {
362	continue;
363	}
364	source_state = test_bit(i, source_caps_bm);
365	target_state = s->capabilities[i];
366	if (source_state != target_state) {
367	error_report("Capability %s is %s, but received capability is %s",
368	MigrationCapability_str(i),
369	target_state ? "on" : "off",
370	source_state ? "on" : "off");
371	ret = false;
372	/* Don't break here to report all failed capabilities */
373	}
374	}
375
376	return ret;
377	}
378
379	static bool configuration_post_load(void opaque, int version_id, Error *errp)
380	{
381	SaveState *state = opaque;
382	const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
383	bool ok = true;
384
385	if (strncmp(state->name, current_name, state->len) != 0) {
386	error_setg(errp,
387	"Machine type received is '%.*s' and local is '%s'",
388	(int) state->len, state->name, current_name);
389	ok = false;
390	goto out;
391	}
392
393	if (state->target_page_bits != qemu_target_page_bits()) {
394	error_setg(errp,
395	"Received TARGET_PAGE_BITS is %d but local is %d",
396	state->target_page_bits, qemu_target_page_bits());
397	ok = false;
398	goto out;
399	}
400
401	if (!configuration_validate_capabilities(state)) {
402	error_setg(errp, "Failed to validate capabilities");
403	ok = false;
404	goto out;
405	}
406
407	out:
408	g_free((void *)state->name);
409	state->name = NULL;
410	state->len = 0;
411	g_free(state->capabilities);
412	state->capabilities = NULL;
413	state->caps_count = 0;
414
415	return ok;
416	}
417
418	static bool load_capability(QEMUFile f, void pv, size_t size,
419	const VMStateField *field,
420	Error **errp)
421	{
422	MigrationCapability *capability = pv;
423	char capability_str[UINT8_MAX + 1];
424	uint8_t len;
425	int i;
426
427	len = qemu_get_byte(f);
428	qemu_get_buffer(f, (uint8_t *)capability_str, len);
429	capability_str[len] = '\0';
430	for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
431	if (!strcmp(MigrationCapability_str(i), capability_str)) {
432	*capability = i;
433	return true;
434	}
435	}
436	error_setg(errp, "Received unknown capability %s", capability_str);
437	return false;
438	}
439
440	static bool save_capability(QEMUFile f, void pv, size_t size,
441	const VMStateField field, JSONWriter vmdesc,
442	Error **errp)
443	{
444	MigrationCapability *capability = pv;
445	const char capability_str = MigrationCapability_str(capability);
446	size_t len = strlen(capability_str);
447	assert(len <= UINT8_MAX);
448
449	qemu_put_byte(f, len);
450	qemu_put_buffer(f, (uint8_t *)capability_str, len);
451	return true;
452	}
453
454	static const VMStateInfo vmstate_info_capability = {
455	.name = "capability",
456	.load = load_capability,
457	.save = save_capability,
458	};
459
460	/* The target-page-bits subsection is present only if the
461	* target page size is not the same as the default (ie the
462	* minimum page size for a variable-page-size guest CPU).
463	* If it is present then it contains the actual target page
464	* bits for the machine, and migration will fail if the
465	* two ends don't agree about it.
466	*/
467	static bool vmstate_target_page_bits_needed(void *opaque)
468	{
469	return qemu_target_page_bits() > migration_legacy_page_bits();
470	}
471
472	static const VMStateDescription vmstate_target_page_bits = {
473	.name = "configuration/target-page-bits",
474	.version_id = 1,
475	.minimum_version_id = 1,
476	.needed = vmstate_target_page_bits_needed,
477	.fields = (const VMStateField[]) {
478	VMSTATE_UINT32(target_page_bits, SaveState),
479	VMSTATE_END_OF_LIST()
480	}
481	};
482
483	static bool vmstate_capabilites_needed(void *opaque)
484	{
485	return get_validatable_capabilities_count() > 0;
486	}
487
488	static const VMStateDescription vmstate_capabilites = {
489	.name = "configuration/capabilities",
490	.version_id = 1,
491	.minimum_version_id = 1,
492	.needed = vmstate_capabilites_needed,
493	.fields = (const VMStateField[]) {
494	VMSTATE_UINT32_V(caps_count, SaveState, 1),
495	VMSTATE_VARRAY_UINT32_ALLOC(capabilities, SaveState, caps_count, 1,
496	vmstate_info_capability,
497	MigrationCapability),
498	VMSTATE_END_OF_LIST()
499	}
500	};
501
502	static bool vmstate_uuid_needed(void *opaque)
503	{
504	return qemu_uuid_set && migrate_validate_uuid();
505	}
506
507	static int vmstate_uuid_post_load(void *opaque, int version_id)
508	{
509	SaveState *state = opaque;
510	char uuid_src[UUID_STR_LEN];
511	char uuid_dst[UUID_STR_LEN];
512
513	if (!qemu_uuid_set) {
514	/*
515	* It's warning because user might not know UUID in some cases,
516	* e.g. load an old snapshot
517	*/
518	qemu_uuid_unparse(&state->uuid, uuid_src);
519	warn_report("UUID is received %s, but local uuid isn't set",
520	uuid_src);
521	return 0;
522	}
523	if (!qemu_uuid_is_equal(&state->uuid, &qemu_uuid)) {
524	qemu_uuid_unparse(&state->uuid, uuid_src);
525	qemu_uuid_unparse(&qemu_uuid, uuid_dst);
526	error_report("UUID received is %s and local is %s", uuid_src, uuid_dst);
527	return -EINVAL;
528	}
529	return 0;
530	}
531
532	static const VMStateDescription vmstate_uuid = {
533	.name = "configuration/uuid",
534	.version_id = 1,
535	.minimum_version_id = 1,
536	.needed = vmstate_uuid_needed,
537	.post_load = vmstate_uuid_post_load,
538	.fields = (const VMStateField[]) {
539	VMSTATE_UINT8_ARRAY_V(uuid.data, SaveState, sizeof(QemuUUID), 1),
540	VMSTATE_END_OF_LIST()
541	}
542	};
543
544	static const VMStateDescription vmstate_configuration = {
545	.name = "configuration",
546	.version_id = 1,
547	.pre_load_errp = configuration_pre_load,
548	.post_load_errp = configuration_post_load,
549	.pre_save_errp = configuration_pre_save,
550	.post_save = configuration_post_save,
551	.fields = (const VMStateField[]) {
552	VMSTATE_UINT32(len, SaveState),
553	VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len),
554	VMSTATE_END_OF_LIST()
555	},
556	.subsections = (const VMStateDescription * const []) {
557	&vmstate_target_page_bits,
558	&vmstate_capabilites,
559	&vmstate_uuid,
560	NULL
561	}
562	};
563
564	static void dump_vmstate_vmsd(FILE *out_file,
565	const VMStateDescription *vmsd, int indent,
566	bool is_subsection);
567
568	static void dump_vmstate_vmsf(FILE out_file, const VMStateField field,
569	int indent)
570	{
571	fprintf(out_file, "%*s{\n", indent, "");
572	indent += 2;
573	fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
574	fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
575	field->version_id);
576	fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
577	field->field_exists ? "true" : "false");
578	if (field->flags & VMS_ARRAY) {
579	fprintf(out_file, "%*s\"num\": %d,\n", indent, "", field->num);
580	}
581	fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
582	if (field->vmsd != NULL) {
583	fprintf(out_file, ",\n");
584	dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
585	}
586	fprintf(out_file, "\n%*s}", indent - 2, "");
587	}
588
589	static void dump_vmstate_vmss(FILE *out_file,
590	const VMStateDescription *subsection,
591	int indent)
592	{
593	if (subsection != NULL) {
594	dump_vmstate_vmsd(out_file, subsection, indent, true);
595	}
596	}
597
598	static void dump_vmstate_vmsd(FILE *out_file,
599	const VMStateDescription *vmsd, int indent,
600	bool is_subsection)
601	{
602	if (is_subsection) {
603	fprintf(out_file, "%*s{\n", indent, "");
604	} else {
605	fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
606	}
607	indent += 2;
608	fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
609	fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
610	vmsd->version_id);
611	fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
612	vmsd->minimum_version_id);
613	if (vmsd->fields != NULL) {
614	const VMStateField *field = vmsd->fields;
615	bool first;
616
617	fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
618	first = true;
619	while (field->name != NULL) {
620	if (field->flags & VMS_MUST_EXIST) {
621	/* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
622	field++;
623	continue;
624	}
625	if (!first) {
626	fprintf(out_file, ",\n");
627	}
628	dump_vmstate_vmsf(out_file, field, indent + 2);
629	field++;
630	first = false;
631	}
632	assert(field->flags == VMS_END);
633	fprintf(out_file, "\n%*s]", indent, "");
634	}
635	if (vmsd->subsections != NULL) {
636	const VMStateDescription * const *subsection = vmsd->subsections;
637	bool first;
638
639	fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
640	first = true;
641	while (*subsection != NULL) {
642	if (!first) {
643	fprintf(out_file, ",\n");
644	}
645	dump_vmstate_vmss(out_file, *subsection, indent + 2);
646	subsection++;
647	first = false;
648	}
649	fprintf(out_file, "\n%*s]", indent, "");
650	}
651	fprintf(out_file, "\n%*s}", indent - 2, "");
652	}
653
654	static void dump_machine_type(FILE *out_file)
655	{
656	MachineClass *mc;
657
658	mc = MACHINE_GET_CLASS(current_machine);
659
660	fprintf(out_file, " \"vmschkmachine\": {\n");
661	fprintf(out_file, " \"Name\": \"%s\"\n", mc->name);
662	fprintf(out_file, " },\n");
663	}
664
665	void dump_vmstate_json_to_file(FILE *out_file)
666	{
667	GSList list, elt;
668	bool first;
669
670	fprintf(out_file, "{\n");
671	dump_machine_type(out_file);
672
673	first = true;
674	list = object_class_get_list(TYPE_DEVICE, true);
675	for (elt = list; elt; elt = elt->next) {
676	DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
677	TYPE_DEVICE);
678	const char *name;
679	int indent = 2;
680
681	if (!dc->vmsd) {
682	continue;
683	}
684
685	if (!first) {
686	fprintf(out_file, ",\n");
687	}
688	name = object_class_get_name(OBJECT_CLASS(dc));
689	fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
690	indent += 2;
691	fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
692	fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
693	dc->vmsd->version_id);
694	fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
695	dc->vmsd->minimum_version_id);
696
697	dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
698
699	fprintf(out_file, "\n%*s}", indent - 2, "");
700	first = false;
701	}
702	fprintf(out_file, "\n}\n");
703	fclose(out_file);
704	g_slist_free(list);
705	}
706
707	static uint32_t calculate_new_instance_id(const char *idstr)
708	{
709	SaveStateEntry *se;
710	uint32_t instance_id = 0;
711
712	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
713	if (strcmp(idstr, se->idstr) == 0
714	&& instance_id <= se->instance_id) {
715	instance_id = se->instance_id + 1;
716	}
717	}
718	/* Make sure we never loop over without being noticed */
719	assert(instance_id != VMSTATE_INSTANCE_ID_ANY);
720	return instance_id;
721	}
722
723	static int calculate_compat_instance_id(const char *idstr)
724	{
725	SaveStateEntry *se;
726	int instance_id = 0;
727
728	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
729	if (!se->compat) {
730	continue;
731	}
732
733	if (strcmp(idstr, se->compat->idstr) == 0
734	&& instance_id <= se->compat->instance_id) {
735	instance_id = se->compat->instance_id + 1;
736	}
737	}
738	return instance_id;
739	}
740
741	static inline MigrationPriority save_state_priority(SaveStateEntry *se)
742	{
743	if (se->vmsd && se->vmsd->priority) {
744	return se->vmsd->priority;
745	}
746	return MIG_PRI_DEFAULT;
747	}
748
749	static void savevm_state_handler_insert(SaveStateEntry *nse)
750	{
751	MigrationPriority priority = save_state_priority(nse);
752	SaveStateEntry *se;
753	int i;
754
755	assert(priority <= MIG_PRI_MAX);
756
757	/*
758	* This should never happen otherwise migration will probably fail
759	* silently somewhere because we can be wrongly applying one
760	* object properties upon another one. Bail out ASAP.
761	*/
762	if (find_se(nse->idstr, nse->instance_id)) {
763	error_report("%s: Detected duplicate SaveStateEntry: "
764	"id=%s, instance_id=0x%"PRIx32, __func__,
765	nse->idstr, nse->instance_id);
766	exit(EXIT_FAILURE);
767	}
768
769	for (i = priority - 1; i >= 0; i--) {
770	se = savevm_state.handler_pri_head[i];
771	if (se != NULL) {
772	assert(save_state_priority(se) < priority);
773	break;
774	}
775	}
776
777	if (i >= 0) {
778	QTAILQ_INSERT_BEFORE(se, nse, entry);
779	} else {
780	QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry);
781	}
782
783	if (savevm_state.handler_pri_head[priority] == NULL) {
784	savevm_state.handler_pri_head[priority] = nse;
785	}
786	}
787
788	static void savevm_state_handler_remove(SaveStateEntry *se)
789	{
790	SaveStateEntry *next;
791	MigrationPriority priority = save_state_priority(se);
792
793	if (se == savevm_state.handler_pri_head[priority]) {
794	next = QTAILQ_NEXT(se, entry);
795	if (next != NULL && save_state_priority(next) == priority) {
796	savevm_state.handler_pri_head[priority] = next;
797	} else {
798	savevm_state.handler_pri_head[priority] = NULL;
799	}
800	}
801	QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
802	}
803
804	/* TODO: Individual devices generally have very little idea about the rest
805	of the system, so instance_id should be removed/replaced.
806	Meanwhile pass -1 as instance_id if you do not already have a clearly
807	distinguishing id for all instances of your device class. */
808	int register_savevm_live(const char *idstr,
809	uint32_t instance_id,
810	int version_id,
811	const SaveVMHandlers *ops,
812	void *opaque)
813	{
814	SaveStateEntry *se;
815
816	se = g_new0(SaveStateEntry, 1);
817	se->version_id = version_id;
818	se->section_id = savevm_state.global_section_id++;
819	se->ops = ops;
820	se->opaque = opaque;
821	se->vmsd = NULL;
822
823	pstrcat(se->idstr, sizeof(se->idstr), idstr);
824
825	if (instance_id == VMSTATE_INSTANCE_ID_ANY) {
826	se->instance_id = calculate_new_instance_id(se->idstr);
827	} else {
828	se->instance_id = instance_id;
829	}
830	assert(!se->compat \|\| se->instance_id == 0);
831	savevm_state_handler_insert(se);
832	return 0;
833	}
834
835	void unregister_savevm(VMStateIf obj, const char idstr, void *opaque)
836	{
837	SaveStateEntry se, new_se;
838	char id[256] = "";
839
840	if (obj) {
841	char *oid = vmstate_if_get_id(obj);
842	if (oid) {
843	pstrcpy(id, sizeof(id), oid);
844	pstrcat(id, sizeof(id), "/");
845	g_free(oid);
846	}
847	}
848	pstrcat(id, sizeof(id), idstr);
849
850	QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
851	if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
852	savevm_state_handler_remove(se);
853	g_free(se->compat);
854	g_free(se);
855	}
856	}
857	}
858
859	/*
860	* Perform some basic checks on vmsd's at registration
861	* time.
862	*/
863	static void vmstate_check(const VMStateDescription *vmsd)
864	{
865	const VMStateField *field = vmsd->fields;
866	const VMStateDescription * const *subsection = vmsd->subsections;
867
868	if (field) {
869	while (field->name) {
870	if (field->flags & VMS_ARRAY_OF_POINTER) {
871	if (field->flags & VMS_ARRAY_OF_POINTER_AUTO_ALLOC) {
872	/*
873	* Size must be provided because dest QEMU needs that
874	* info to know what to allocate
875	*/
876	assert(field->size != 0 \|\| field->size_indirect.size != 0);
877	} else {
878	/*
879	* Otherwise size info isn't useful (because it's
880	* always the size of host pointer), detect accidental
881	* setup of sizes in this case.
882	*/
883	assert(field->size == 0 && field->size_indirect.size == 0);
884	}
885	/*
886	* VMS_ARRAY_OF_POINTER must be used only together with one
887	* of VMS_(V)ARRAY flags.
888	*/
889	assert(field->flags & (VMS_ARRAY \| VMS_VARRAY));
890	}
891
892	if (field->flags & VMS_ARRAY_OF_POINTER_AUTO_ALLOC) {
893	assert(field->flags & VMS_ARRAY_OF_POINTER);
894	}
895
896	/*
897	* The VMS*ARRAY flags and VMS_VBUFFER affect allocation,
898	* they must have the proper fields set and no other
899	* vmstate types can set those fields, otherwise it won't
900	* be picked-up due to the missing flag.
901	*/
902
903	if (field->flags & (VMS_ARRAY \| VMS_VARRAY)) {
904	assert(field->num > 0 \|\| field->num_indirect.size != 0);
905	} else {
906	assert(field->num == 0 && field->num_indirect.size == 0);
907	}
908
909	if (field->flags & VMS_VBUFFER) {
910	assert(field->size_indirect.size != 0);
911	} else {
912	assert(field->size_indirect.size == 0);
913	}
914
915	if (field->flags & (VMS_STRUCT \| VMS_VSTRUCT)) {
916	/* Recurse to sub structures */
917	vmstate_check(field->vmsd);
918	}
919	/* Carry on */
920	field++;
921	}
922	/* Check for the end of field list canary */
923	if (field->flags != VMS_END) {
924	error_report("VMSTATE not ending with VMS_END: %s", vmsd->name);
925	g_assert_not_reached();
926	}
927	}
928
929	while (subsection && *subsection) {
930	/*
931	* The name of a subsection should start with the name of the
932	* current object.
933	*/
934	assert(!strncmp(vmsd->name, (*subsection)->name, strlen(vmsd->name)));
935	vmstate_check(*subsection);
936	subsection++;
937	}
938	}
939
940
941	int vmstate_register_with_alias_id(VMStateIf *obj, uint32_t instance_id,
942	const VMStateDescription *vmsd,
943	void *opaque, int alias_id,
944	int required_for_version,
945	Error **errp)
946	{
947	SaveStateEntry *se;
948
949	/* If this triggers, alias support can be dropped for the vmsd. */
950	assert(alias_id == -1 \|\| required_for_version >= vmsd->minimum_version_id);
951
952	se = g_new0(SaveStateEntry, 1);
953	se->version_id = vmsd->version_id;
954	se->section_id = savevm_state.global_section_id++;
955	se->opaque = opaque;
956	se->vmsd = vmsd;
957	se->alias_id = alias_id;
958
959	if (obj) {
960	char *id = vmstate_if_get_id(obj);
961	if (id) {
962	if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
963	sizeof(se->idstr)) {
964	error_setg(errp, "Path too long for VMState (%s)", id);
965	g_free(id);
966	g_free(se);
967
968	return -1;
969	}
970	g_free(id);
971
972	se->compat = g_new0(CompatEntry, 1);
973	pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
974	se->compat->instance_id = instance_id == VMSTATE_INSTANCE_ID_ANY ?
975	calculate_compat_instance_id(vmsd->name) : instance_id;
976	instance_id = VMSTATE_INSTANCE_ID_ANY;
977	}
978	}
979	pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
980
981	if (instance_id == VMSTATE_INSTANCE_ID_ANY) {
982	se->instance_id = calculate_new_instance_id(se->idstr);
983	} else {
984	se->instance_id = instance_id;
985	}
986
987	/* Perform a recursive sanity check during the test runs */
988	if (qtest_enabled()) {
989	vmstate_check(vmsd);
990	}
991	assert(!se->compat \|\| se->instance_id == 0);
992	savevm_state_handler_insert(se);
993	return 0;
994	}
995
996	void vmstate_unregister(VMStateIf obj, const VMStateDescription vmsd,
997	void *opaque)
998	{
999	SaveStateEntry se, new_se;
1000
1001	QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
1002	if (se->vmsd == vmsd && se->opaque == opaque) {
1003	savevm_state_handler_remove(se);
1004	g_free(se->compat);
1005	g_free(se);
1006	}
1007	}
1008	}
1009
1010	static int vmstate_load(QEMUFile f, SaveStateEntry se, Error **errp)
1011	{
1012	int ret;
1013	trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
1014	if (!se->vmsd) { /* Old style */
1015	ret = se->ops->load_state(f, se->opaque, se->load_version_id);
1016	if (ret < 0) {
1017	error_setg(errp, "Failed to load vmstate version_id: %d, ret: %d",
1018	se->load_version_id, ret);
1019	}
1020	return ret;
1021	}
1022
1023	if (!vmstate_load_vmsd(f, se->vmsd, se->opaque, se->load_version_id,
1024	errp)) {
1025	return -EINVAL;
1026	}
1027
1028	return 0;
1029	}
1030
1031	static void vmstate_save_old_style(QEMUFile f, SaveStateEntry se,
1032	JSONWriter *vmdesc)
1033	{
1034	uint64_t old_offset = qemu_file_transferred(f);
1035	se->ops->save_state(f, se->opaque);
1036	uint64_t size = qemu_file_transferred(f) - old_offset;
1037
1038	if (vmdesc) {
1039	json_writer_int64(vmdesc, "size", size);
1040	json_writer_start_array(vmdesc, "fields");
1041	json_writer_start_object(vmdesc, NULL);
1042	json_writer_str(vmdesc, "name", "data");
1043	json_writer_int64(vmdesc, "size", size);
1044	json_writer_str(vmdesc, "type", "buffer");
1045	json_writer_end_object(vmdesc);
1046	json_writer_end_array(vmdesc);
1047	}
1048	}
1049
1050	/*
1051	* Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
1052	*/
1053	static void save_section_header(QEMUFile f, SaveStateEntry se,
1054	uint8_t section_type)
1055	{
1056	qemu_put_byte(f, section_type);
1057	qemu_put_be32(f, se->section_id);
1058
1059	if (section_type == QEMU_VM_SECTION_FULL \|\|
1060	section_type == QEMU_VM_SECTION_START) {
1061	/* ID string */
1062	size_t len = strlen(se->idstr);
1063	qemu_put_byte(f, len);
1064	qemu_put_buffer(f, (uint8_t *)se->idstr, len);
1065
1066	qemu_put_be32(f, se->instance_id);
1067	qemu_put_be32(f, se->version_id);
1068	}
1069	}
1070
1071	/*
1072	* Write a footer onto device sections that catches cases misformatted device
1073	* sections.
1074	*/
1075	static void save_section_footer(QEMUFile f, SaveStateEntry se)
1076	{
1077	if (migrate_get_current()->send_section_footer) {
1078	qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
1079	qemu_put_be32(f, se->section_id);
1080	}
1081	}
1082
1083	static int vmstate_save(QEMUFile f, SaveStateEntry se, JSONWriter *vmdesc,
1084	Error **errp)
1085	{
1086	if ((!se->ops \|\| !se->ops->save_state) && !se->vmsd) {
1087	return 0;
1088	}
1089	if (se->vmsd && !vmstate_section_needed(se->vmsd, se->opaque)) {
1090	trace_savevm_section_skip(se->idstr, se->section_id);
1091	return 0;
1092	}
1093
1094	trace_savevm_section_start(se->idstr, se->section_id);
1095	save_section_header(f, se, QEMU_VM_SECTION_FULL);
1096	if (vmdesc) {
1097	json_writer_start_object(vmdesc, NULL);
1098	json_writer_str(vmdesc, "name", se->idstr);
1099	json_writer_int64(vmdesc, "instance_id", se->instance_id);
1100	}
1101
1102	trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
1103	if (!se->vmsd) {
1104	vmstate_save_old_style(f, se, vmdesc);
1105	} else {
1106	if (!vmstate_save_vmsd(f, se->vmsd, se->opaque, vmdesc, errp)) {
1107	return -EINVAL;
1108	}
1109	}
1110
1111	trace_savevm_section_end(se->idstr, se->section_id, 0);
1112	save_section_footer(f, se);
1113	if (vmdesc) {
1114	json_writer_end_object(vmdesc);
1115	}
1116	return 0;
1117	}
1118
1119	void qemu_savevm_state_end(QEMUFile *f)
1120	{
1121	qemu_put_byte(f, QEMU_VM_EOF);
1122	}
1123
1124	static inline bool qemu_savevm_state_active(SaveStateEntry *se)
1125	{
1126	/* When no is_active() hook, always treat it as ACTIVE */
1127	if (!se->ops->is_active) {
1128	return true;
1129	}
1130
1131	return se->ops->is_active(se->opaque);
1132	}
1133
1134	/**
1135	* qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
1136	* command and associated data.
1137	*
1138	* @f: File to send command on
1139	* @command: Command type to send
1140	* @len: Length of associated data
1141	* @data: Data associated with command.
1142	*/
1143	static void qemu_savevm_command_send(QEMUFile *f,
1144	enum qemu_vm_cmd command,
1145	uint16_t len,
1146	uint8_t *data)
1147	{
1148	trace_savevm_command_send(command, len);
1149	qemu_put_byte(f, QEMU_VM_COMMAND);
1150	qemu_put_be16(f, (uint16_t)command);
1151	qemu_put_be16(f, len);
1152	qemu_put_buffer(f, data, len);
1153	qemu_fflush(f);
1154	}
1155
1156	void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
1157	{
1158	uint32_t buf;
1159
1160	trace_savevm_send_ping(value);
1161	buf = cpu_to_be32(value);
1162	qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
1163	}
1164
1165	void qemu_savevm_send_open_return_path(QEMUFile *f)
1166	{
1167	trace_savevm_send_open_return_path();
1168	qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
1169	}
1170
1171	/* We have a buffer of data to send; we don't want that all to be loaded
1172	* by the command itself, so the command contains just the length of the
1173	* extra buffer that we then send straight after it.
1174	* TODO: Must be a better way to organise that
1175	*
1176	* Returns:
1177	* 0 on success
1178	* -ve on error
1179	*/
1180	int qemu_savevm_send_packaged(QEMUFile f, const uint8_t buf, size_t len)
1181	{
1182	uint32_t tmp;
1183	Error *local_err = NULL;
1184
1185	if (len > MAX_VM_CMD_PACKAGED_SIZE) {
1186	error_setg(&local_err, "%s: Unreasonably large packaged state: %zu",
1187	__func__, len);
1188	migrate_error_propagate(migrate_get_current(),
1189	error_copy(local_err));
1190	error_report_err(local_err);
1191	return -1;
1192	}
1193
1194	tmp = cpu_to_be32(len);
1195
1196	trace_qemu_savevm_send_packaged();
1197	qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
1198
1199	/* We can use async put because the qemufile will be flushed right away */
1200	qemu_put_buffer_async(f, buf, len, false);
1201	qemu_fflush(f);
1202
1203	return 0;
1204	}
1205
1206	/* Send prior to any postcopy transfer */
1207	void qemu_savevm_send_postcopy_advise(QEMUFile *f)
1208	{
1209	if (migrate_postcopy_ram()) {
1210	uint64_t tmp[2];
1211	tmp[0] = cpu_to_be64(ram_pagesize_summary());
1212	tmp[1] = cpu_to_be64(qemu_target_page_size());
1213
1214	trace_qemu_savevm_send_postcopy_advise();
1215	qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE,
1216	16, (uint8_t *)tmp);
1217	} else {
1218	qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 0, NULL);
1219	}
1220	}
1221
1222	/* Sent prior to starting the destination running in postcopy, discard pages
1223	* that have already been sent but redirtied on the source.
1224	* CMD_POSTCOPY_RAM_DISCARD consist of:
1225	* byte version (0)
1226	* byte Length of name field (not including 0)
1227	* n x byte RAM block name
1228	* byte 0 terminator (just for safety)
1229	* n x Byte ranges within the named RAMBlock
1230	* be64 Start of the range
1231	* be64 Length
1232	*
1233	* name: RAMBlock name that these entries are part of
1234	* len: Number of page entries
1235	* start_list: 'len' addresses
1236	* length_list: 'len' addresses
1237	*
1238	*/
1239	void qemu_savevm_send_postcopy_ram_discard(QEMUFile f, const char name,
1240	uint16_t len,
1241	uint64_t *start_list,
1242	uint64_t *length_list)
1243	{
1244	uint8_t *buf;
1245	uint16_t tmplen;
1246	uint16_t t;
1247	size_t name_len = strlen(name);
1248
1249	trace_qemu_savevm_send_postcopy_ram_discard(name, len);
1250	assert(name_len < 256);
1251	buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
1252	buf[0] = postcopy_ram_discard_version;
1253	buf[1] = name_len;
1254	memcpy(buf + 2, name, name_len);
1255	tmplen = 2 + name_len;
1256	buf[tmplen++] = '\0';
1257
1258	for (t = 0; t < len; t++) {
1259	stq_be_p(buf + tmplen, start_list[t]);
1260	tmplen += 8;
1261	stq_be_p(buf + tmplen, length_list[t]);
1262	tmplen += 8;
1263	}
1264	qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
1265	g_free(buf);
1266	}
1267
1268	/* Get the destination into a state where it can receive postcopy data. */
1269	void qemu_savevm_send_postcopy_listen(QEMUFile *f)
1270	{
1271	trace_savevm_send_postcopy_listen();
1272	qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
1273	}
1274
1275	/* Kick the destination into running */
1276	void qemu_savevm_send_postcopy_run(QEMUFile *f)
1277	{
1278	trace_savevm_send_postcopy_run();
1279	qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
1280	}
1281
1282	void qemu_savevm_send_postcopy_resume(QEMUFile *f)
1283	{
1284	trace_savevm_send_postcopy_resume();
1285	qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RESUME, 0, NULL);
1286	}
1287
1288	void qemu_savevm_send_recv_bitmap(QEMUFile f, char block_name)
1289	{
1290	size_t len;
1291	char buf[256];
1292
1293	trace_savevm_send_recv_bitmap(block_name);
1294
1295	buf[0] = len = strlen(block_name);
1296	memcpy(buf + 1, block_name, len);
1297
1298	qemu_savevm_command_send(f, MIG_CMD_RECV_BITMAP, len + 1, (uint8_t *)buf);
1299	}
1300
1301	static void qemu_savevm_send_switchover_start(QEMUFile *f)
1302	{
1303	trace_savevm_send_switchover_start();
1304	qemu_savevm_command_send(f, MIG_CMD_SWITCHOVER_START, 0, NULL);
1305	}
1306
1307	void qemu_savevm_maybe_send_switchover_start(QEMUFile *f)
1308	{
1309	if (migrate_send_switchover_start()) {
1310	qemu_savevm_send_switchover_start(f);
1311	}
1312	}
1313
1314	bool qemu_savevm_state_blocked(Error **errp)
1315	{
1316	SaveStateEntry *se;
1317
1318	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1319	if (se->vmsd && se->vmsd->unmigratable) {
1320	error_setg(errp, "State blocked by non-migratable device '%s'",
1321	se->idstr);
1322	return true;
1323	}
1324	}
1325	return false;
1326	}
1327
1328	void qemu_savevm_non_migratable_list(strList **reasons)
1329	{
1330	SaveStateEntry *se;
1331
1332	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1333	if (se->vmsd && se->vmsd->unmigratable) {
1334	QAPI_LIST_PREPEND(*reasons,
1335	g_strdup_printf("non-migratable device: %s",
1336	se->idstr));
1337	}
1338	}
1339	}
1340
1341	void qemu_savevm_send_header(QEMUFile *f)
1342	{
1343	trace_savevm_state_header();
1344	qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1345	qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1346	}
1347
1348	static void qemu_savevm_send_configuration(MigrationState s, QEMUFile f)
1349	{
1350	JSONWriter *vmdesc = s->vmdesc;
1351	Error *local_err = NULL;
1352
1353	qemu_put_byte(f, QEMU_VM_CONFIGURATION);
1354
1355	if (vmdesc) {
1356	/*
1357	* This starts the main json object and is paired with the
1358	* json_writer_end_object in
1359	* qemu_savevm_state_complete_precopy_non_iterable
1360	*/
1361	json_writer_start_object(vmdesc, NULL);
1362	json_writer_start_object(vmdesc, "configuration");
1363	}
1364
1365	vmstate_save_vmsd(f, &vmstate_configuration, &savevm_state,
1366	vmdesc, &local_err);
1367	if (local_err) {
1368	error_report_err(local_err);
1369	}
1370
1371	if (vmdesc) {
1372	json_writer_end_object(vmdesc);
1373	}
1374	}
1375
1376	void qemu_savevm_state_header(QEMUFile *f)
1377	{
1378	MigrationState *s = migrate_get_current();
1379
1380	qemu_savevm_send_header(f);
1381	if (s->send_configuration) {
1382	qemu_savevm_send_configuration(s, f);
1383	}
1384	}
1385
1386	bool qemu_savevm_state_guest_unplug_pending(void)
1387	{
1388	SaveStateEntry *se;
1389
1390	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1391	if (se->vmsd && se->vmsd->dev_unplug_pending &&
1392	se->vmsd->dev_unplug_pending(se->opaque)) {
1393	return true;
1394	}
1395	}
1396
1397	return false;
1398	}
1399
1400	int qemu_savevm_state_prepare(Error **errp)
1401	{
1402	SaveStateEntry *se;
1403	int ret;
1404
1405	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1406	if (!se->ops \|\| !se->ops->save_prepare) {
1407	continue;
1408	}
1409	if (!qemu_savevm_state_active(se)) {
1410	continue;
1411	}
1412	ret = se->ops->save_prepare(se->opaque, errp);
1413	if (ret < 0) {
1414	return ret;
1415	}
1416	}
1417
1418	return 0;
1419	}
1420
1421	int qemu_savevm_state_non_iterable_early(QEMUFile *f,
1422	JSONWriter *vmdesc,
1423	Error **errp)
1424	{
1425	SaveStateEntry *se;
1426	int ret;
1427
1428	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1429	if (se->vmsd && se->vmsd->early_setup) {
1430	ret = vmstate_save(f, se, vmdesc, errp);
1431	if (ret) {
1432	return ret;
1433	}
1434	}
1435	}
1436
1437	return 0;
1438	}
1439
1440	static int qemu_savevm_state_setup(QEMUFile f, Error *errp)
1441	{
1442	SaveStateEntry *se;
1443	int ret;
1444
1445	trace_savevm_state_setup();
1446
1447	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1448	if (!se->ops \|\| !se->ops->save_setup) {
1449	continue;
1450	}
1451	if (!qemu_savevm_state_active(se)) {
1452	continue;
1453	}
1454	save_section_header(f, se, QEMU_VM_SECTION_START);
1455	ret = se->ops->save_setup(f, se->opaque, errp);
1456	save_section_footer(f, se);
1457	if (ret < 0) {
1458	return ret;
1459	}
1460	}
1461
1462	return 0;
1463	}
1464
1465	int qemu_savevm_state_do_setup(QEMUFile f, Error *errp)
1466	{
1467	ERRP_GUARD();
1468	MigrationState *ms = migrate_get_current();
1469	JSONWriter *vmdesc = ms->vmdesc;
1470	int ret;
1471
1472	if (vmdesc) {
1473	json_writer_int64(vmdesc, "page_size", qemu_target_page_size());
1474	json_writer_start_array(vmdesc, "devices");
1475	}
1476
1477	ret = qemu_savevm_state_non_iterable_early(f, vmdesc, errp);
1478	if (ret) {
1479	return ret;
1480	}
1481
1482	ret = qemu_savevm_state_setup(f, errp);
1483	if (ret) {
1484	return ret;
1485	}
1486
1487	/* TODO: Should we check that errp is set in case of failure ? */
1488	return precopy_notify(PRECOPY_NOTIFY_SETUP, errp);
1489	}
1490
1491	int qemu_savevm_state_resume_prepare(MigrationState *s)
1492	{
1493	SaveStateEntry *se;
1494	int ret;
1495
1496	trace_savevm_state_resume_prepare();
1497
1498	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1499	if (!se->ops \|\| !se->ops->resume_prepare) {
1500	continue;
1501	}
1502	if (!qemu_savevm_state_active(se)) {
1503	continue;
1504	}
1505	ret = se->ops->resume_prepare(s, se->opaque);
1506	if (ret < 0) {
1507	return ret;
1508	}
1509	}
1510
1511	return 0;
1512	}
1513
1514	/*
1515	* this function has three return values:
1516	* negative: there was one error, and we have -errno.
1517	* 0 : We haven't finished, caller have to go again
1518	* 1 : We have finished, we can go to complete phase
1519	*/
1520	int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
1521	{
1522	SaveStateEntry *se;
1523	bool all_finished = true;
1524	int ret;
1525
1526	trace_savevm_state_iterate();
1527	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1528	if (!se->ops \|\| !se->ops->save_live_iterate) {
1529	continue;
1530	}
1531	if (!qemu_savevm_state_active(se)) {
1532	continue;
1533	}
1534	if (se->ops->is_active_iterate &&
1535	!se->ops->is_active_iterate(se->opaque)) {
1536	continue;
1537	}
1538	/*
1539	* In the postcopy phase, any device that doesn't know how to
1540	* do postcopy should have saved it's state in the _complete
1541	* call that's already run, it might get confused if we call
1542	* iterate afterwards.
1543	*/
1544	if (postcopy &&
1545	!(se->ops->has_postcopy && se->ops->has_postcopy(se->opaque))) {
1546	continue;
1547	}
1548	if (migration_rate_exceeded(f)) {
1549	return 0;
1550	}
1551	trace_savevm_section_start(se->idstr, se->section_id);
1552
1553	save_section_header(f, se, QEMU_VM_SECTION_PART);
1554
1555	ret = se->ops->save_live_iterate(f, se->opaque);
1556	trace_savevm_section_end(se->idstr, se->section_id, ret);
1557	save_section_footer(f, se);
1558
1559	if (ret < 0) {
1560	error_report("failed to save SaveStateEntry with id(name): "
1561	"%d(%s): %d",
1562	se->section_id, se->idstr, ret);
1563	qemu_file_set_error(f, ret);
1564	return ret;
1565	} else if (!ret) {
1566	all_finished = false;
1567	}
1568	}
1569	return all_finished;
1570	}
1571
1572	bool should_send_vmdesc(void)
1573	{
1574	MachineState *machine = MACHINE(qdev_get_machine());
1575
1576	return !machine->suppress_vmdesc;
1577	}
1578
1579	static bool qemu_savevm_complete_exists(SaveStateEntry *se)
1580	{
1581	return se->ops && se->ops->save_complete;
1582	}
1583
1584	/*
1585	* Invoke the ->save_complete() if necessary.
1586	* Returns: 0 if skip the current SE or succeeded, <0 if error happened.
1587	*/
1588	static int qemu_savevm_complete(SaveStateEntry se, QEMUFile f)
1589	{
1590	int ret;
1591
1592	if (!qemu_savevm_state_active(se)) {
1593	return 0;
1594	}
1595
1596	trace_savevm_section_start(se->idstr, se->section_id);
1597	save_section_header(f, se, QEMU_VM_SECTION_END);
1598	ret = se->ops->save_complete(f, se->opaque);
1599	trace_savevm_section_end(se->idstr, se->section_id, ret);
1600	save_section_footer(f, se);
1601
1602	if (ret < 0) {
1603	qemu_file_set_error(f, ret);
1604	}
1605
1606	return ret;
1607	}
1608
1609	/*
1610	* Complete saving any postcopy-able devices.
1611	*
1612	* Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1613	* all the other devices, but that happens at the point we switch to postcopy.
1614	*/
1615	void qemu_savevm_state_complete_postcopy(QEMUFile *f)
1616	{
1617	SaveStateEntry *se;
1618
1619	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1620	if (!qemu_savevm_complete_exists(se)) {
1621	continue;
1622	}
1623
1624	if (qemu_savevm_complete(se, f) < 0) {
1625	return;
1626	}
1627	}
1628
1629	qemu_savevm_state_end(f);
1630	qemu_fflush(f);
1631	}
1632
1633	bool qemu_savevm_state_postcopy_prepare(QEMUFile f, Error *errp)
1634	{
1635	SaveStateEntry *se;
1636	bool ret;
1637
1638	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1639	if (!se->ops \|\| !se->ops->save_postcopy_prepare) {
1640	continue;
1641	}
1642
1643	if (!qemu_savevm_state_active(se)) {
1644	continue;
1645	}
1646
1647	trace_savevm_section_start(se->idstr, se->section_id);
1648
1649	save_section_header(f, se, QEMU_VM_SECTION_PART);
1650	ret = se->ops->save_postcopy_prepare(f, se->opaque, errp);
1651	save_section_footer(f, se);
1652
1653	trace_savevm_section_end(se->idstr, se->section_id, ret);
1654
1655	if (!ret) {
1656	assert(*errp);
1657	return false;
1658	}
1659	}
1660
1661	return true;
1662	}
1663
1664	int qemu_savevm_state_complete_precopy_iterable(QEMUFile *f, bool in_postcopy)
1665	{
1666	int64_t start_ts_each, end_ts_each;
1667	SaveStateEntry *se;
1668	bool multifd_device_state = multifd_device_state_supported();
1669
1670	if (multifd_device_state) {
1671	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1672	SaveCompletePrecopyThreadHandler hdlr;
1673
1674	if (!se->ops \|\| (in_postcopy && se->ops->has_postcopy &&
1675	se->ops->has_postcopy(se->opaque)) \|\|
1676	!se->ops->save_complete_precopy_thread) {
1677	continue;
1678	}
1679
1680	hdlr = se->ops->save_complete_precopy_thread;
1681	multifd_spawn_device_state_save_thread(hdlr,
1682	se->idstr, se->instance_id,
1683	se->opaque);
1684	}
1685	}
1686
1687	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1688	if (!qemu_savevm_complete_exists(se)) {
1689	continue;
1690	}
1691
1692	if (in_postcopy && se->ops->has_postcopy &&
1693	se->ops->has_postcopy(se->opaque)) {
1694	/*
1695	* If postcopy will start soon, and if the SE supports
1696	* postcopy, then we can skip the SE for the postcopy phase.
1697	*/
1698	continue;
1699	}
1700
1701	start_ts_each = qemu_clock_get_us(QEMU_CLOCK_REALTIME);
1702	if (qemu_savevm_complete(se, f) < 0) {
1703	goto ret_fail_abort_threads;
1704	}
1705	end_ts_each = qemu_clock_get_us(QEMU_CLOCK_REALTIME);
1706
1707	trace_vmstate_downtime_save("iterable", se->idstr, se->instance_id,
1708	end_ts_each - start_ts_each);
1709	}
1710
1711	if (multifd_device_state) {
1712	if (migrate_has_error(migrate_get_current())) {
1713	multifd_abort_device_state_save_threads();
1714	}
1715
1716	if (!multifd_join_device_state_save_threads()) {
1717	qemu_file_set_error(f, -EINVAL);
1718	return -1;
1719	}
1720	}
1721
1722	trace_vmstate_downtime_checkpoint("src-iterable-saved");
1723
1724	return 0;
1725
1726	ret_fail_abort_threads:
1727	if (multifd_device_state) {
1728	multifd_abort_device_state_save_threads();
1729	multifd_join_device_state_save_threads();
1730	}
1731
1732	return -1;
1733	}
1734
1735	static void qemu_savevm_state_vm_desc(MigrationState s, QEMUFile f)
1736	{
1737	JSONWriter *vmdesc = s->vmdesc;
1738	int vmdesc_len;
1739
1740	if (vmdesc) {
1741	json_writer_end_array(vmdesc);
1742	json_writer_end_object(vmdesc);
1743	vmdesc_len = strlen(json_writer_get(vmdesc));
1744
1745	qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1746	qemu_put_be32(f, vmdesc_len);
1747	qemu_put_buffer(f, (uint8_t *)json_writer_get(vmdesc), vmdesc_len);
1748	}
1749	}
1750
1751	void qemu_savevm_state_end_precopy(MigrationState s, QEMUFile f)
1752	{
1753	qemu_savevm_state_end(f);
1754	qemu_savevm_state_vm_desc(s, f);
1755	}
1756
1757	bool qemu_savevm_state_non_iterable(QEMUFile f, Error *errp)
1758	{
1759	MigrationState *ms = migrate_get_current();
1760	int64_t start_ts_each, end_ts_each;
1761	JSONWriter *vmdesc = ms->vmdesc;
1762	SaveStateEntry *se;
1763
1764	/* Making sure cpu states are synchronized before saving non-iterable */
1765	cpu_synchronize_all_states();
1766
1767	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1768	if (se->vmsd && se->vmsd->early_setup) {
1769	/* Already saved during qemu_savevm_state_setup(). */
1770	continue;
1771	}
1772
1773	start_ts_each = qemu_clock_get_us(QEMU_CLOCK_REALTIME);
1774
1775	if (vmstate_save(f, se, vmdesc, errp) < 0) {
1776	return false;
1777	}
1778
1779	end_ts_each = qemu_clock_get_us(QEMU_CLOCK_REALTIME);
1780	trace_vmstate_downtime_save("non-iterable", se->idstr, se->instance_id,
1781	end_ts_each - start_ts_each);
1782	}
1783
1784	trace_vmstate_downtime_checkpoint("src-non-iterable-saved");
1785
1786	return true;
1787	}
1788
1789	bool qemu_savevm_state_complete_precopy(MigrationState s, Error *errp)
1790	{
1791	ERRP_GUARD();
1792	QEMUFile *f = s->to_dst_file;
1793	int ret;
1794
1795	ret = qemu_savevm_state_complete_precopy_iterable(f, false);
1796	if (ret) {
1797	qemu_file_get_error_obj(f, errp);
1798	error_prepend(errp, "Failed to save iterable device state: ");
1799	return false;
1800	}
1801
1802	if (!qemu_savevm_state_non_iterable(f, errp)) {
1803	return false;
1804	}
1805
1806	qemu_savevm_state_end_precopy(s, f);
1807
1808	ret = qemu_fflush(f);
1809	if (ret) {
1810	qemu_file_get_error_obj(f, errp);
1811	error_prepend(errp, "Failed to flush QEMUFile: ");
1812	return false;
1813	}
1814
1815	return true;
1816	}
1817
1818	static void qemu_savevm_query_pending(MigrationState *s,
1819	MigPendingData *pending, bool exact,
1820	bool final)
1821	{
1822	SaveStateEntry *se;
1823
1824	memset(pending, 0, sizeof(*pending));
1825
1826	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1827	if (!se->ops \|\| !se->ops->save_query_pending) {
1828	continue;
1829	}
1830	if (!qemu_savevm_state_active(se)) {
1831	continue;
1832	}
1833	se->ops->save_query_pending(se->opaque, pending, exact, final);
1834	}
1835
1836	pending->total_bytes = pending->precopy_bytes +
1837	pending->stopcopy_bytes + pending->postcopy_bytes;
1838
1839	/*
1840	* Update system remaining dirty bytes whenever QEMU queries. It will
1841	* make the value to be not as accurate, but should still be pretty
1842	* close to reality when this got invoked frequently while iterating.
1843	*/
1844	mig_stats.dirty_bytes_total = pending->total_bytes;
1845
1846	if (migrate_switchover_ack() && !migrate_switchover_ack_legacy() &&
1847	pending->switchover_ack_pending) {
1848	/*
1849	* NOTE: Currently we rely on per-device protocol to request switchover
1850	* ACK from the device on the destination side.
1851	*/
1852	qatomic_add(&s->switchover_ack_pending_num,
1853	pending->switchover_ack_pending);
1854	}
1855
1856	trace_qemu_savevm_query_pending(
1857	exact, final, pending->precopy_bytes, pending->stopcopy_bytes,
1858	pending->postcopy_bytes, pending->total_bytes,
1859	pending->switchover_ack_pending,
1860	qatomic_read(&s->switchover_ack_pending_num));
1861	}
1862
1863	void qemu_savevm_query_pending_iter(MigrationState s, MigPendingData pending,
1864	bool exact)
1865	{
1866	qemu_savevm_query_pending(s, pending, exact, false);
1867	}
1868
1869	bool qemu_savevm_query_pending_final(MigrationState s, MigPendingData pending,
1870	Error **errp)
1871	{
1872	g_assert(bql_locked());
1873
1874	qemu_savevm_query_pending(s, pending, true, true);
1875
1876	/*
1877	* Switchover-ack requests done after switchover decision are not allowed.
1878	* Fail the migration in this case since we currently don't support going
1879	* back to precopy.
1880	*/
1881	if (migrate_switchover_ack() && !migrate_switchover_ack_legacy() &&
1882	pending->switchover_ack_pending > 0) {
1883	error_setg(errp,
1884	"Switchover ACK was requested by %" PRIu32
1885	" devices during switchover",
1886	pending->switchover_ack_pending);
1887	return false;
1888	}
1889
1890	return true;
1891	}
1892
1893	void qemu_savevm_state_cleanup(void)
1894	{
1895	SaveStateEntry *se;
1896	Error *local_err = NULL;
1897
1898	if (precopy_notify(PRECOPY_NOTIFY_CLEANUP, &local_err)) {
1899	error_report_err(local_err);
1900	}
1901
1902	trace_savevm_state_cleanup();
1903	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1904	if (se->ops && se->ops->save_cleanup) {
1905	se->ops->save_cleanup(se->opaque);
1906	}
1907	}
1908	}
1909
1910	static int qemu_savevm_state(QEMUFile f, Error *errp)
1911	{
1912	int ret;
1913	MigrationState *ms = migrate_get_current();
1914	MigrationStatus status;
1915
1916	if (migration_is_running()) {
1917	error_setg(errp, "There's a migration process in progress");
1918	return -EINVAL;
1919	}
1920
1921	ret = migrate_init(ms, errp);
1922	if (ret) {
1923	return ret;
1924	}
1925	ms->to_dst_file = f;
1926
1927	qemu_savevm_state_header(f);
1928	ret = qemu_savevm_state_do_setup(f, errp);
1929	if (ret) {
1930	goto cleanup;
1931	}
1932
1933	while (qemu_file_get_error(f) == 0) {
1934	if (qemu_savevm_state_iterate(f, false) > 0) {
1935	break;
1936	}
1937	}
1938
1939	ret = qemu_file_get_error(f);
1940	if (ret) {
1941	error_setg_errno(errp, -ret, "Error while writing VM state");
1942	goto cleanup;
1943	}
1944
1945	if (!qemu_savevm_state_complete_precopy(ms, errp)) {
1946	ret = -1;
1947	}
1948	cleanup:
1949	qemu_savevm_state_cleanup();
1950
1951	if (ret != 0) {
1952	status = MIGRATION_STATUS_FAILED;
1953	} else {
1954	status = MIGRATION_STATUS_COMPLETED;
1955	}
1956	migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status);
1957
1958	/* f is outer parameter, it should not stay in global migration state after
1959	* this function finished */
1960	ms->to_dst_file = NULL;
1961
1962	return ret;
1963	}
1964
1965	/* Is a save state entry iterable (e.g. RAM)? */
1966	static bool qemu_savevm_se_iterable(SaveStateEntry *se)
1967	{
1968	return se->ops && se->ops->save_setup;
1969	}
1970
1971	int qemu_save_device_state(QEMUFile f, Error *errp)
1972	{
1973	int ret;
1974
1975	/* Both COLO and Xen never use vmdesc, hence NULL. */
1976	ret = qemu_savevm_state_non_iterable_early(f, NULL, errp);
1977	if (ret) {
1978	return ret;
1979	}
1980
1981	if (!qemu_savevm_state_non_iterable(f, errp)) {
1982	return -1;
1983	}
1984
1985	qemu_savevm_state_end(f);
1986
1987	return 0;
1988	}
1989
1990	static SaveStateEntry find_se(const char idstr, uint32_t instance_id)
1991	{
1992	SaveStateEntry *se;
1993
1994	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1995	if (!strcmp(se->idstr, idstr) &&
1996	(instance_id == se->instance_id \|\|
1997	instance_id == se->alias_id))
1998	return se;
1999	/* Migrating from an older version? */
2000	if (strstr(se->idstr, idstr) && se->compat) {
2001	if (!strcmp(se->compat->idstr, idstr) &&
2002	(instance_id == se->compat->instance_id \|\|
2003	instance_id == se->alias_id))
2004	return se;
2005	}
2006	}
2007	return NULL;
2008	}
2009
2010	enum LoadVMExitCodes {
2011	/* Allow a command to quit all layers of nested loadvm loops */
2012	LOADVM_QUIT = 1,
2013	};
2014
2015	/* ------ incoming postcopy messages ------ */
2016	/* 'advise' arrives before any transfers just to tell us that a postcopy
2017	* might happen - it might be skipped if precopy transferred everything
2018	* quickly.
2019	*/
2020	static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis,
2021	uint16_t len, Error **errp)
2022	{
2023	PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
2024	uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps;
2025	size_t page_size = qemu_target_page_size();
2026
2027	trace_loadvm_postcopy_handle_advise();
2028	if (ps != POSTCOPY_INCOMING_NONE) {
2029	error_setg(errp, "CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)",
2030	ps);
2031	return -1;
2032	}
2033
2034	switch (len) {
2035	case 0:
2036	if (migrate_postcopy_ram()) {
2037	error_setg(errp, "RAM postcopy is enabled but have 0 byte advise");
2038	return -EINVAL;
2039	}
2040	return 0;
2041	case 8 + 8:
2042	if (!migrate_postcopy_ram()) {
2043	error_setg(errp,
2044	"RAM postcopy is disabled but have 16 byte advise");
2045	return -EINVAL;
2046	}
2047	break;
2048	default:
2049	error_setg(errp, "CMD_POSTCOPY_ADVISE invalid length (%d)", len);
2050	return -EINVAL;
2051	}
2052
2053	if (!postcopy_ram_supported_by_host(mis, errp)) {
2054	postcopy_state_set(POSTCOPY_INCOMING_NONE);
2055	return -1;
2056	}
2057
2058	remote_pagesize_summary = qemu_get_be64(mis->from_src_file);
2059	local_pagesize_summary = ram_pagesize_summary();
2060
2061	if (remote_pagesize_summary != local_pagesize_summary) {
2062	/*
2063	* This detects two potential causes of mismatch:
2064	* a) A mismatch in host page sizes
2065	* Some combinations of mismatch are probably possible but it gets
2066	* a bit more complicated. In particular we need to place whole
2067	* host pages on the dest at once, and we need to ensure that we
2068	* handle dirtying to make sure we never end up sending part of
2069	* a hostpage on it's own.
2070	* b) The use of different huge page sizes on source/destination
2071	* a more fine grain test is performed during RAM block migration
2072	* but this test here causes a nice early clear failure, and
2073	* also fails when passed to an older qemu that doesn't
2074	* do huge pages.
2075	*/
2076	error_setg(errp,
2077	"Postcopy needs matching RAM page sizes "
2078	"(s=%" PRIx64 " d=%" PRIx64 ")",
2079	remote_pagesize_summary, local_pagesize_summary);
2080	return -1;
2081	}
2082
2083	remote_tps = qemu_get_be64(mis->from_src_file);
2084	if (remote_tps != page_size) {
2085	/*
2086	* Again, some differences could be dealt with, but for now keep it
2087	* simple.
2088	*/
2089	error_setg(errp,
2090	"Postcopy needs matching target page sizes (s=%d d=%zd)",
2091	(int)remote_tps, page_size);
2092	return -1;
2093	}
2094
2095	if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_ADVISE, errp)) {
2096	return -1;
2097	}
2098
2099	if (ram_postcopy_incoming_init(mis, errp) < 0) {
2100	error_prepend(errp, "Postcopy RAM incoming init failed: ");
2101	return -1;
2102	}
2103
2104	return 0;
2105	}
2106
2107	/* After postcopy we will be told to throw some pages away since they're
2108	* dirty and will have to be demand fetched. Must happen before CPU is
2109	* started.
2110	* There can be 0..many of these messages, each encoding multiple pages.
2111	*/
2112	static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
2113	uint16_t len, Error **errp)
2114	{
2115	int tmp;
2116	char ramid[256];
2117	PostcopyState ps = postcopy_state_get();
2118
2119	trace_loadvm_postcopy_ram_handle_discard();
2120
2121	switch (ps) {
2122	case POSTCOPY_INCOMING_ADVISE:
2123	/* 1st discard */
2124	tmp = postcopy_ram_prepare_discard(mis);
2125	if (tmp) {
2126	error_setg(errp, "Failed to prepare for RAM discard: %d", tmp);
2127	return tmp;
2128	}
2129	break;
2130
2131	case POSTCOPY_INCOMING_DISCARD:
2132	/* Expected state */
2133	break;
2134
2135	default:
2136	error_setg(errp,
2137	"CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
2138	ps);
2139	return -1;
2140	}
2141	/* We're expecting a
2142	* Version (0)
2143	* a RAM ID string (length byte, name, 0 term)
2144	* then at least 1 16 byte chunk
2145	*/
2146	if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
2147	error_setg(errp, "CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
2148	return -1;
2149	}
2150
2151	tmp = qemu_get_byte(mis->from_src_file);
2152	if (tmp != postcopy_ram_discard_version) {
2153	error_setg(errp, "CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
2154	return -1;
2155	}
2156
2157	if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
2158	error_setg(errp,
2159	"CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
2160	return -1;
2161	}
2162	tmp = qemu_get_byte(mis->from_src_file);
2163	if (tmp != 0) {
2164	error_setg(errp, "CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
2165	return -1;
2166	}
2167
2168	len -= 3 + strlen(ramid);
2169	if (len % 16) {
2170	error_setg(errp, "CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
2171	return -1;
2172	}
2173	trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
2174	while (len) {
2175	uint64_t start_addr, block_length;
2176	start_addr = qemu_get_be64(mis->from_src_file);
2177	block_length = qemu_get_be64(mis->from_src_file);
2178
2179	len -= 16;
2180	int ret = ram_discard_range(ramid, start_addr, block_length);
2181	if (ret) {
2182	error_setg(errp, "Failed to discard RAM range %s: %d", ramid, ret);
2183	return ret;
2184	}
2185	}
2186	trace_loadvm_postcopy_ram_handle_discard_end();
2187
2188	return 0;
2189	}
2190
2191	/* After this message we must be able to immediately receive postcopy data */
2192	static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis,
2193	Error **errp)
2194	{
2195	PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
2196
2197	trace_loadvm_postcopy_handle_listen("enter");
2198
2199	if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
2200	error_setg(errp,
2201	"CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
2202	return -1;
2203	}
2204	if (ps == POSTCOPY_INCOMING_ADVISE) {
2205	/*
2206	* A rare case, we entered listen without having to do any discards,
2207	* so do the setup that's normally done at the time of the 1st discard.
2208	*/
2209	if (migrate_postcopy_ram()) {
2210	postcopy_ram_prepare_discard(mis);
2211	}
2212	}
2213
2214	trace_loadvm_postcopy_handle_listen("after discard");
2215
2216	int rc = postcopy_incoming_setup(mis, errp);
2217
2218	trace_loadvm_postcopy_handle_listen("return");
2219
2220	return rc;
2221	}
2222
2223	static void loadvm_postcopy_handle_run_bh(void *opaque)
2224	{
2225	MigrationIncomingState *mis = opaque;
2226
2227	trace_vmstate_downtime_checkpoint("dst-postcopy-bh-enter");
2228
2229	/* TODO we should move all of this lot into postcopy_ram.c or a shared code
2230	* in migration.c
2231	*/
2232	cpu_synchronize_all_post_init();
2233
2234	trace_vmstate_downtime_checkpoint("dst-postcopy-bh-cpu-synced");
2235
2236	qemu_announce_self(&mis->announce_timer, migrate_announce_params());
2237
2238	trace_vmstate_downtime_checkpoint("dst-postcopy-bh-announced");
2239
2240	dirty_bitmap_mig_before_vm_start();
2241
2242	if (autostart) {
2243	/*
2244	* Make sure all file formats throw away their mutable metadata.
2245	* If we get an error here, just don't restart the VM yet.
2246	*/
2247	bool success = migration_block_activate(NULL);
2248
2249	trace_vmstate_downtime_checkpoint("dst-postcopy-bh-cache-invalidated");
2250
2251	if (success) {
2252	vm_start();
2253	}
2254	} else {
2255	/* leave it paused and let management decide when to start the CPU */
2256	runstate_set(RUN_STATE_PAUSED);
2257	}
2258
2259	trace_vmstate_downtime_checkpoint("dst-postcopy-bh-vm-started");
2260	}
2261
2262	/* After all discards we can start running and asking for pages */
2263	static int loadvm_postcopy_handle_run(MigrationIncomingState mis, Error *errp)
2264	{
2265	PostcopyState ps = postcopy_state_get();
2266
2267	trace_loadvm_postcopy_handle_run();
2268	if (ps != POSTCOPY_INCOMING_LISTENING) {
2269	error_setg(errp, "CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
2270	return -1;
2271	}
2272
2273	/* We might be already in POSTCOPY_ACTIVE if there is no return path */
2274	if (mis->state == MIGRATION_STATUS_POSTCOPY_DEVICE) {
2275	migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_DEVICE,
2276	MIGRATION_STATUS_POSTCOPY_ACTIVE);
2277	}
2278	postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
2279	migration_bh_schedule(loadvm_postcopy_handle_run_bh, mis);
2280
2281	/* We need to finish reading the stream from the package
2282	* and also stop reading anything more from the stream that loaded the
2283	* package (since it's now being read by the listener thread).
2284	* LOADVM_QUIT will quit all the layers of nested loadvm loops.
2285	*/
2286	return LOADVM_QUIT;
2287	}
2288
2289	/* We must be with page_request_mutex held */
2290	static gboolean postcopy_sync_page_req(gpointer key, gpointer value,
2291	gpointer data)
2292	{
2293	MigrationIncomingState *mis = data;
2294	void host_addr = (void ) key;
2295	ram_addr_t rb_offset;
2296	RAMBlock *rb;
2297	int ret;
2298
2299	rb = qemu_ram_block_from_host(host_addr, true, &rb_offset);
2300	if (!rb) {
2301	/*
2302	* This should _never_ happen. However be nice for a migrating VM to
2303	* not crash/assert. Post an error (note: intended to not use *_once
2304	* because we do want to see all the illegal addresses; and this can
2305	* never be triggered by the guest so we're safe) and move on next.
2306	*/
2307	error_report("%s: illegal host addr %p", __func__, host_addr);
2308	/* Try the next entry */
2309	return FALSE;
2310	}
2311
2312	ret = migrate_send_rp_message_req_pages(mis, rb, rb_offset);
2313	if (ret) {
2314	/* Please refer to above comment. */
2315	error_report("%s: send rp message failed for addr %p",
2316	__func__, host_addr);
2317	return FALSE;
2318	}
2319
2320	trace_postcopy_page_req_sync(host_addr);
2321
2322	return FALSE;
2323	}
2324
2325	static void migrate_send_rp_req_pages_pending(MigrationIncomingState *mis)
2326	{
2327	WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) {
2328	g_tree_foreach(mis->page_requested, postcopy_sync_page_req, mis);
2329	}
2330	}
2331
2332	static void loadvm_postcopy_handle_resume(MigrationIncomingState *mis)
2333	{
2334	if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
2335	warn_report("%s: illegal resume received", __func__);
2336	/* Don't fail the load, only for this. */
2337	return;
2338	}
2339
2340	/*
2341	* Reset the last_rb before we resend any page req to source again, since
2342	* the source should have it reset already.
2343	*/
2344	mis->last_rb = NULL;
2345
2346	/*
2347	* This means source VM is ready to resume the postcopy migration.
2348	*/
2349	migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2350	MIGRATION_STATUS_POSTCOPY_ACTIVE);
2351
2352	trace_loadvm_postcopy_handle_resume();
2353
2354	/* Tell source that "we are ready" */
2355	migrate_send_rp_resume_ack(mis, MIGRATION_RESUME_ACK_VALUE);
2356
2357	/*
2358	* After a postcopy recovery, the source should have lost the postcopy
2359	* queue, or potentially the requested pages could have been lost during
2360	* the network down phase. Let's re-sync with the source VM by re-sending
2361	* all the pending pages that we eagerly need, so these threads won't get
2362	* blocked too long due to the recovery.
2363	*
2364	* Without this procedure, the faulted destination VM threads (waiting for
2365	* page requests right before the postcopy is interrupted) can keep hanging
2366	* until the pages are sent by the source during the background copying of
2367	* pages, or another thread faulted on the same address accidentally.
2368	*/
2369	migrate_send_rp_req_pages_pending(mis);
2370
2371	/*
2372	* It's time to switch state and release the fault thread to continue
2373	* service page faults. Note that this should be explicitly after the
2374	* above call to migrate_send_rp_req_pages_pending(). In short:
2375	* migrate_send_rp_message_req_pages() is not thread safe, yet.
2376	*/
2377	qemu_sem_post(&mis->postcopy_pause_sem_fault);
2378
2379	if (migrate_postcopy_preempt()) {
2380	/*
2381	* The preempt channel will be created in async manner, now let's
2382	* wait for it and make sure it's created.
2383	*/
2384	qemu_sem_wait(&mis->postcopy_qemufile_dst_done);
2385	assert(mis->postcopy_qemufile_dst);
2386	/* Kick the fast ram load thread too */
2387	qemu_sem_post(&mis->postcopy_pause_sem_fast_load);
2388	}
2389	}
2390
2391	/**
2392	* Immediately following this command is a blob of data containing an embedded
2393	* chunk of migration stream; read it and load it.
2394	*
2395	* @mis: Incoming state
2396	* @length: Length of packaged data to read
2397	*
2398	* Returns: Negative values on error
2399	*
2400	*/
2401	static int loadvm_handle_cmd_packaged(MigrationIncomingState mis, Error *errp)
2402	{
2403	int ret;
2404	size_t length;
2405	QIOChannelBuffer *bioc;
2406
2407	length = qemu_get_be32(mis->from_src_file);
2408	trace_loadvm_handle_cmd_packaged(length);
2409
2410	if (length > MAX_VM_CMD_PACKAGED_SIZE) {
2411	error_setg(errp, "Unreasonably large packaged state: %zu", length);
2412	return -1;
2413	}
2414
2415	bioc = qio_channel_buffer_new(length);
2416	qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer");
2417	ret = qemu_get_buffer(mis->from_src_file,
2418	bioc->data,
2419	length);
2420	if (ret != length) {
2421	object_unref(OBJECT(bioc));
2422	error_setg(errp, "CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
2423	ret, length);
2424	return (ret < 0) ? ret : -EAGAIN;
2425	}
2426	bioc->usage += length;
2427	trace_loadvm_handle_cmd_packaged_received(ret);
2428
2429	QEMUFile *packf = qemu_file_new_input(QIO_CHANNEL(bioc));
2430
2431	/*
2432	* Before loading the guest states, ensure that the preempt channel has
2433	* been ready to use, as some of the states (e.g. via virtio_load) might
2434	* trigger page faults that will be handled through the preempt channel.
2435	* So yield to the main thread in the case that the channel create event
2436	* hasn't been dispatched.
2437	*
2438	* TODO: if we can move migration loadvm out of main thread, then we
2439	* won't block main thread from polling the accept() fds. We can drop
2440	* this as a whole when that is done.
2441	*/
2442	do {
2443	if (!migrate_postcopy_preempt() \|\| !qemu_in_coroutine() \|\|
2444	mis->postcopy_qemufile_dst) {
2445	break;
2446	}
2447
2448	aio_co_schedule(qemu_get_current_aio_context(), qemu_coroutine_self());
2449	qemu_coroutine_yield();
2450	} while (1);
2451
2452	ret = qemu_loadvm_state_main(packf, mis, errp);
2453	trace_loadvm_handle_cmd_packaged_main(ret);
2454	qemu_fclose(packf);
2455	object_unref(OBJECT(bioc));
2456
2457	return ret;
2458	}
2459
2460	/*
2461	* Handle request that source requests for recved_bitmap on
2462	* destination. Payload format:
2463	*
2464	* len (1 byte) + ramblock_name (<255 bytes)
2465	*/
2466	static int loadvm_handle_recv_bitmap(MigrationIncomingState *mis,
2467	uint16_t len, Error **errp)
2468	{
2469	QEMUFile *file = mis->from_src_file;
2470	RAMBlock *rb;
2471	char block_name[256];
2472	size_t cnt;
2473	int ret;
2474
2475	cnt = qemu_get_counted_string(file, block_name);
2476	if (!cnt) {
2477	error_setg(errp, "failed to read block name");
2478	return -EINVAL;
2479	}
2480
2481	/* Validate before using the data */
2482	ret = qemu_file_get_error(file);
2483	if (ret < 0) {
2484	error_setg(errp, "loadvm failed: stream error: %d", ret);
2485	return ret;
2486	}
2487
2488	if (len != cnt + 1) {
2489	error_setg(errp, "invalid payload length (%d)", len);
2490	return -EINVAL;
2491	}
2492
2493	rb = qemu_ram_block_by_name(block_name);
2494	if (!rb) {
2495	error_setg(errp, "block '%s' not found", block_name);
2496	return -EINVAL;
2497	}
2498
2499	migrate_send_rp_recv_bitmap(mis, block_name);
2500
2501	trace_loadvm_handle_recv_bitmap(block_name);
2502
2503	return 0;
2504	}
2505
2506	static int loadvm_postcopy_handle_switchover_start(Error **errp)
2507	{
2508	SaveStateEntry *se;
2509
2510	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2511	int ret;
2512
2513	if (!se->ops \|\| !se->ops->switchover_start) {
2514	continue;
2515	}
2516
2517	ret = se->ops->switchover_start(se->opaque);
2518	if (ret < 0) {
2519	error_setg(errp, "Switchover start failed: %d", ret);
2520	return ret;
2521	}
2522	}
2523
2524	return 0;
2525	}
2526
2527	/*
2528	* If legacy switchover-ack is enabled but no device uses it, need to send an
2529	* ACK to source that it's OK to switch over.
2530	*/
2531	static int loadvm_switchover_ack_no_users_legacy(MigrationIncomingState *mis,
2532	Error **errp)
2533	{
2534	int ret;
2535
2536	if (!migrate_switchover_ack() \|\| !migrate_switchover_ack_legacy()) {
2537	return 0;
2538	}
2539
2540	if (!mis->switchover_ack_pending_num_legacy) {
2541	ret = migrate_send_rp_switchover_ack(mis);
2542	if (ret) {
2543	error_setg_errno(errp, -ret,
2544	"Could not send switchover ack RP MSG");
2545	return ret;
2546	}
2547	}
2548
2549	return 0;
2550	}
2551
2552	/*
2553	* Process an incoming 'QEMU_VM_COMMAND'
2554	* 0 just a normal return
2555	* LOADVM_QUIT All good, but exit the loop
2556	* <0 Error
2557	*/
2558	static int loadvm_process_command(QEMUFile f, Error *errp)
2559	{
2560	MigrationIncomingState *mis = migration_incoming_get_current();
2561	uint16_t cmd;
2562	uint16_t len;
2563	uint32_t tmp32;
2564	int ret;
2565
2566	cmd = qemu_get_be16(f);
2567	len = qemu_get_be16(f);
2568
2569	/* Check validity before continue processing of cmds */
2570	ret = qemu_file_get_error(f);
2571	if (ret) {
2572	error_setg(errp,
2573	"Failed to load VM process command: stream error: %d",
2574	ret);
2575	return ret;
2576	}
2577
2578	if (cmd >= MIG_CMD_MAX \|\| cmd == MIG_CMD_INVALID) {
2579	error_setg(errp, "MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
2580	return -EINVAL;
2581	}
2582
2583	trace_loadvm_process_command(mig_cmd_args[cmd].name, len);
2584
2585	if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
2586	error_setg(errp, "%s received with bad length - expecting %zu, got %d",
2587	mig_cmd_args[cmd].name,
2588	(size_t)mig_cmd_args[cmd].len, len);
2589	return -ERANGE;
2590	}
2591
2592	switch (cmd) {
2593	case MIG_CMD_OPEN_RETURN_PATH:
2594	if (mis->to_src_file) {
2595	error_report("CMD_OPEN_RETURN_PATH called when RP already open");
2596	/* Not really a problem, so don't give up */
2597	return 0;
2598	}
2599	mis->to_src_file = qemu_file_get_return_path(f);
2600
2601	ret = loadvm_switchover_ack_no_users_legacy(mis, errp);
2602	if (ret) {
2603	return ret;
2604	}
2605	return 0;
2606
2607	case MIG_CMD_PING:
2608	tmp32 = qemu_get_be32(f);
2609	trace_loadvm_process_command_ping(tmp32);
2610	if (!mis->to_src_file) {
2611	error_setg(errp, "CMD_PING (0x%x) received with no return path",
2612	tmp32);
2613	return -1;
2614	}
2615	migrate_send_rp_pong(mis, tmp32);
2616	return 0;
2617
2618	case MIG_CMD_PACKAGED:
2619	return loadvm_handle_cmd_packaged(mis, errp);
2620
2621	case MIG_CMD_POSTCOPY_ADVISE:
2622	return loadvm_postcopy_handle_advise(mis, len, errp);
2623
2624	case MIG_CMD_POSTCOPY_LISTEN:
2625	return loadvm_postcopy_handle_listen(mis, errp);
2626
2627	case MIG_CMD_POSTCOPY_RUN:
2628	return loadvm_postcopy_handle_run(mis, errp);
2629
2630	case MIG_CMD_POSTCOPY_RAM_DISCARD:
2631	return loadvm_postcopy_ram_handle_discard(mis, len, errp);
2632
2633	case MIG_CMD_POSTCOPY_RESUME:
2634	loadvm_postcopy_handle_resume(mis);
2635	return 0;
2636
2637	case MIG_CMD_RECV_BITMAP:
2638	return loadvm_handle_recv_bitmap(mis, len, errp);
2639
2640	case MIG_CMD_SWITCHOVER_START:
2641	return loadvm_postcopy_handle_switchover_start(errp);
2642	}
2643
2644	error_setg(errp, "MIG_CMD 0x%x deprecated (len 0x%x)", cmd, len);
2645	return -EINVAL;
2646	}
2647
2648	/*
2649	* Read a footer off the wire and check that it matches the expected section
2650	*
2651	* Returns: true if the footer was good
2652	* false if there is a problem (and calls error_report to say why)
2653	*/
2654	static bool check_section_footer(QEMUFile f, SaveStateEntry se)
2655	{
2656	int ret;
2657	uint8_t read_mark;
2658	uint32_t read_section_id;
2659
2660	if (!migrate_get_current()->send_section_footer) {
2661	/* No footer to check */
2662	return true;
2663	}
2664
2665	read_mark = qemu_get_byte(f);
2666
2667	ret = qemu_file_get_error(f);
2668	if (ret) {
2669	error_report("%s: Read section footer failed: %d",
2670	__func__, ret);
2671	return false;
2672	}
2673
2674	if (read_mark != QEMU_VM_SECTION_FOOTER) {
2675	error_report("Missing section footer for %s", se->idstr);
2676	return false;
2677	}
2678
2679	read_section_id = qemu_get_be32(f);
2680	if (read_section_id != se->load_section_id) {
2681	error_report("Mismatched section id in footer for %s -"
2682	" read 0x%x expected 0x%x",
2683	se->idstr, read_section_id, se->load_section_id);
2684	return false;
2685	}
2686
2687	/* All good */
2688	return true;
2689	}
2690
2691	static int
2692	qemu_loadvm_section_start_full(QEMUFile f, uint8_t type, Error *errp)
2693	{
2694	ERRP_GUARD();
2695	bool trace_downtime = (type == QEMU_VM_SECTION_FULL);
2696	uint32_t instance_id, version_id, section_id;
2697	int64_t start_ts, end_ts;
2698	SaveStateEntry *se;
2699	char idstr[256];
2700	int ret;
2701
2702	/* Read section start */
2703	section_id = qemu_get_be32(f);
2704	if (!qemu_get_counted_string(f, idstr)) {
2705	error_setg(errp, "Unable to read ID string for section %u",
2706	section_id);
2707	return -EINVAL;
2708	}
2709	instance_id = qemu_get_be32(f);
2710	version_id = qemu_get_be32(f);
2711
2712	ret = qemu_file_get_error(f);
2713	if (ret) {
2714	error_setg(errp, "Failed to read instance/version ID: %d", ret);
2715	return ret;
2716	}
2717
2718	trace_qemu_loadvm_state_section_startfull(section_id, idstr,
2719	instance_id, version_id);
2720	/* Find savevm section */
2721	se = find_se(idstr, instance_id);
2722	if (se == NULL) {
2723	error_setg(errp, "Unknown section or instance '%s' %"PRIu32". "
2724	"Make sure that your current VM setup matches your "
2725	"saved VM setup, including any hotplugged devices",
2726	idstr, instance_id);
2727	return -EINVAL;
2728	}
2729
2730	/* Validate version */
2731	if (version_id > se->version_id) {
2732	error_setg(errp, "unsupported version %d for '%s' v%d",
2733	version_id, idstr, se->version_id);
2734	return -EINVAL;
2735	}
2736	se->load_version_id = version_id;
2737	se->load_section_id = section_id;
2738
2739	/* Validate if it is a device's state */
2740	if (xen_enabled() && qemu_savevm_se_iterable(se)) {
2741	error_setg(errp, "loadvm: %s RAM loading not allowed on Xen", idstr);
2742	return -EINVAL;
2743	}
2744
2745	if (trace_downtime) {
2746	start_ts = qemu_clock_get_us(QEMU_CLOCK_REALTIME);
2747	}
2748
2749	ret = vmstate_load(f, se, errp);
2750	if (ret < 0) {
2751	error_prepend(errp,
2752	"error while loading state for instance 0x%"PRIx32" of"
2753	" device '%s': ", instance_id, idstr);
2754	return ret;
2755	}
2756
2757	if (trace_downtime) {
2758	end_ts = qemu_clock_get_us(QEMU_CLOCK_REALTIME);
2759	trace_vmstate_downtime_load("non-iterable", se->idstr,
2760	se->instance_id, end_ts - start_ts);
2761	}
2762
2763	if (!check_section_footer(f, se)) {
2764	error_setg(errp, "Section footer error, section_id: %d",
2765	section_id);
2766	return -EINVAL;
2767	}
2768
2769	return 0;
2770	}
2771
2772	static int
2773	qemu_loadvm_section_part_end(QEMUFile f, uint8_t type, Error *errp)
2774	{
2775	bool trace_downtime = (type == QEMU_VM_SECTION_END);
2776	int64_t start_ts, end_ts;
2777	uint32_t section_id;
2778	SaveStateEntry *se;
2779	int ret;
2780
2781	section_id = qemu_get_be32(f);
2782
2783	ret = qemu_file_get_error(f);
2784	if (ret) {
2785	error_setg(errp, "Failed to read section ID: %d", ret);
2786	return ret;
2787	}
2788
2789	trace_qemu_loadvm_state_section_partend(section_id);
2790	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2791	if (se->load_section_id == section_id) {
2792	break;
2793	}
2794	}
2795	if (se == NULL) {
2796	error_setg(errp, "Unknown section %d", section_id);
2797	return -EINVAL;
2798	}
2799
2800	if (trace_downtime) {
2801	start_ts = qemu_clock_get_us(QEMU_CLOCK_REALTIME);
2802	}
2803
2804	ret = vmstate_load(f, se, errp);
2805	if (ret < 0) {
2806	return ret;
2807	}
2808
2809	if (trace_downtime) {
2810	end_ts = qemu_clock_get_us(QEMU_CLOCK_REALTIME);
2811	trace_vmstate_downtime_load("iterable", se->idstr,
2812	se->instance_id, end_ts - start_ts);
2813	}
2814
2815	if (!check_section_footer(f, se)) {
2816	error_setg(errp, "Section footer error, section_id: %d",
2817	section_id);
2818	return -EINVAL;
2819	}
2820
2821	return 0;
2822	}
2823
2824	static int qemu_loadvm_state_header(QEMUFile f, Error *errp)
2825	{
2826	unsigned int v;
2827
2828	v = qemu_get_be32(f);
2829	if (v != QEMU_VM_FILE_MAGIC) {
2830	error_setg(errp, "Not a migration stream, magic: %x != %x",
2831	v, QEMU_VM_FILE_MAGIC);
2832	return -EINVAL;
2833	}
2834
2835	v = qemu_get_be32(f);
2836	if (v == QEMU_VM_FILE_VERSION_COMPAT) {
2837	error_setg(errp,
2838	"SaveVM v2 format is obsolete and no longer supported");
2839
2840	return -ENOTSUP;
2841	}
2842	if (v != QEMU_VM_FILE_VERSION) {
2843	error_setg(errp, "Unsupported migration stream version, "
2844	"file version %x != %x",
2845	v, QEMU_VM_FILE_VERSION);
2846	return -ENOTSUP;
2847	}
2848
2849	if (migrate_get_current()->send_configuration) {
2850	v = qemu_get_byte(f);
2851	if (v != QEMU_VM_CONFIGURATION) {
2852	error_setg(errp, "Configuration section missing, %x != %x",
2853	v, QEMU_VM_CONFIGURATION);
2854	return -EINVAL;
2855	}
2856
2857	if (!vmstate_load_vmsd(f, &vmstate_configuration, &savevm_state, 0,
2858	errp)) {
2859	return -EINVAL;
2860	}
2861	}
2862	return 0;
2863	}
2864
2865	static int qemu_loadvm_state_setup(QEMUFile f, Error *errp)
2866	{
2867	ERRP_GUARD();
2868	SaveStateEntry *se;
2869	int ret;
2870
2871	trace_loadvm_state_setup();
2872	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2873	if (!se->ops \|\| !se->ops->load_setup) {
2874	continue;
2875	}
2876	if (!qemu_savevm_state_active(se)) {
2877	continue;
2878	}
2879	ret = se->ops->load_setup(f, se->opaque, errp);
2880	if (ret < 0) {
2881	error_prepend(errp, "Load state of device %s failed: ",
2882	se->idstr);
2883	qemu_file_set_error(f, ret);
2884	return ret;
2885	}
2886	}
2887	return 0;
2888	}
2889
2890	struct LoadThreadData {
2891	MigrationLoadThread function;
2892	void *opaque;
2893	};
2894
2895	static int qemu_loadvm_load_thread(void *thread_opaque)
2896	{
2897	struct LoadThreadData *data = thread_opaque;
2898	MigrationIncomingState *mis = migration_incoming_get_current();
2899	Error *local_err = NULL;
2900
2901	if (!data->function(data->opaque, &mis->load_threads_abort, &local_err)) {
2902	/*
2903	* Can't set load_threads_abort here since processing of main migration
2904	* channel data could still be happening, resulting in launching of new
2905	* load threads.
2906	*/
2907
2908	assert(local_err);
2909
2910	/*
2911	* In case of multiple load threads failing which thread error
2912	* return we end setting is purely arbitrary.
2913	*/
2914	migrate_error_propagate(migrate_get_current(), local_err);
2915	}
2916
2917	return 0;
2918	}
2919
2920	void qemu_loadvm_start_load_thread(MigrationLoadThread function,
2921	void *opaque)
2922	{
2923	MigrationIncomingState *mis = migration_incoming_get_current();
2924	struct LoadThreadData *data;
2925
2926	/* We only set it from this thread so it's okay to read it directly */
2927	assert(!mis->load_threads_abort);
2928
2929	data = g_new(struct LoadThreadData, 1);
2930	data->function = function;
2931	data->opaque = opaque;
2932
2933	thread_pool_submit_immediate(mis->load_threads, qemu_loadvm_load_thread,
2934	data, g_free);
2935	}
2936
2937	void qemu_loadvm_state_cleanup(MigrationIncomingState *mis)
2938	{
2939	SaveStateEntry *se;
2940
2941	trace_loadvm_state_cleanup();
2942
2943	QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
2944	if (se->ops && se->ops->load_cleanup) {
2945	se->ops->load_cleanup(se->opaque);
2946	}
2947	}
2948
2949	qemu_loadvm_thread_pool_destroy(mis);
2950	}
2951
2952	/* Return true if we should continue the migration, or false. */
2953	static bool postcopy_pause_incoming(MigrationIncomingState *mis)
2954	{
2955	int i;
2956
2957	trace_postcopy_pause_incoming();
2958
2959	assert(migrate_postcopy_ram());
2960
2961	assert(mis->from_src_file);
2962
2963	/*
2964	* Unregister yank with either from/to src would work, since ioc behind it
2965	* is the same
2966	*/
2967	migration_ioc_unregister_yank_from_file(mis->from_src_file);
2968
2969	qemu_file_shutdown(mis->from_src_file);
2970	qemu_fclose(mis->from_src_file);
2971	mis->from_src_file = NULL;
2972
2973	assert(mis->to_src_file);
2974	qemu_file_shutdown(mis->to_src_file);
2975	qemu_mutex_lock(&mis->rp_mutex);
2976	qemu_fclose(mis->to_src_file);
2977	mis->to_src_file = NULL;
2978	qemu_mutex_unlock(&mis->rp_mutex);
2979
2980	/*
2981	* NOTE: this must happen before reset the PostcopyTmpPages below,
2982	* otherwise it's racy to reset those fields when the fast load thread
2983	* can be accessing it in parallel.
2984	*/
2985	if (mis->postcopy_qemufile_dst) {
2986	qemu_file_shutdown(mis->postcopy_qemufile_dst);
2987	/* Take the mutex to make sure the fast ram load thread halted */
2988	qemu_mutex_lock(&mis->postcopy_prio_thread_mutex);
2989	migration_ioc_unregister_yank_from_file(mis->postcopy_qemufile_dst);
2990	qemu_fclose(mis->postcopy_qemufile_dst);
2991	mis->postcopy_qemufile_dst = NULL;
2992	qemu_mutex_unlock(&mis->postcopy_prio_thread_mutex);
2993	}
2994
2995	/* Current state can be either ACTIVE or RECOVER */
2996	migrate_set_state(&mis->state, mis->state,
2997	MIGRATION_STATUS_POSTCOPY_PAUSED);
2998
2999	/* Notify the fault thread for the invalidated file handle */
3000	postcopy_fault_thread_notify(mis);
3001
3002	/*
3003	* If network is interrupted, any temp page we received will be useless
3004	* because we didn't mark them as "received" in receivedmap. After a
3005	* proper recovery later (which will sync src dirty bitmap with receivedmap
3006	* on dest) these cached small pages will be resent again.
3007	*/
3008	for (i = 0; i < mis->postcopy_channels; i++) {
3009	postcopy_temp_page_reset(&mis->postcopy_tmp_pages[i]);
3010	}
3011
3012	error_report("Detected IO failure for postcopy. "
3013	"Migration paused.");
3014
3015	do {
3016	qemu_sem_wait(&mis->postcopy_pause_sem_dst);
3017	} while (postcopy_is_paused(mis->state));
3018
3019	trace_postcopy_pause_incoming_continued();
3020
3021	return true;
3022	}
3023
3024	/*
3025	* Starts the VM and launches the eager thread for fast snapshot load
3026	*/
3027	void qemu_loadvm_run_fast_snapshot_load(QEMUFile *f,
3028	MigrationIncomingState *mis)
3029	{
3030	postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
3031
3032	migration_bh_schedule(loadvm_postcopy_handle_run_bh, mis);
3033
3034	migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_DEVICE,
3035	MIGRATION_STATUS_POSTCOPY_ACTIVE);
3036
3037	postcopy_ram_eager_load_setup(mis);
3038	}
3039
3040	int qemu_loadvm_state_main(QEMUFile f, MigrationIncomingState mis,
3041	Error **errp)
3042	{
3043	ERRP_GUARD();
3044	uint8_t section_type;
3045	int ret = 0;
3046
3047	retry:
3048	while (true) {
3049	section_type = qemu_get_byte(f);
3050
3051	ret = qemu_file_get_error_obj_any(f, mis->postcopy_qemufile_dst, errp);
3052	if (ret) {
3053	error_prepend(errp,
3054	"Failed to load section ID: stream error: %d: ",
3055	ret);
3056	break;
3057	}
3058
3059	trace_qemu_loadvm_state_section(section_type);
3060	switch (section_type) {
3061	case QEMU_VM_SECTION_START:
3062	case QEMU_VM_SECTION_FULL:
3063	ret = qemu_loadvm_section_start_full(f, section_type, errp);
3064	if (ret < 0) {
3065	goto out;
3066	}
3067	break;
3068	case QEMU_VM_SECTION_PART:
3069	case QEMU_VM_SECTION_END:
3070	ret = qemu_loadvm_section_part_end(f, section_type, errp);
3071	if (ret < 0) {
3072	goto out;
3073	}
3074	break;
3075	case QEMU_VM_COMMAND:
3076	ret = loadvm_process_command(f, errp);
3077	trace_qemu_loadvm_state_section_command(ret);
3078	if ((ret < 0) \|\| (ret == LOADVM_QUIT)) {
3079	goto out;
3080	}
3081	break;
3082	case QEMU_VM_EOF:
3083	/* This is the end of migration */
3084	goto out;
3085	default:
3086	error_setg(errp, "Unknown section type %d", section_type);
3087	ret = -EINVAL;
3088	goto out;
3089	}
3090	}
3091
3092	out:
3093	if (ret < 0) {
3094	qemu_file_set_error(f, ret);
3095
3096	/* Cancel bitmaps incoming regardless of recovery */
3097	dirty_bitmap_mig_cancel_incoming();
3098
3099	/*
3100	* If we are during an active postcopy, then we pause instead
3101	* of bail out to at least keep the VM's dirty data. Note
3102	* that POSTCOPY_INCOMING_LISTENING stage is still not enough,
3103	* during which we're still receiving device states and we
3104	* still haven't yet started the VM on destination.
3105	*
3106	* Only RAM postcopy supports recovery. Still, if RAM postcopy is
3107	* enabled, canceled bitmaps postcopy will not affect RAM postcopy
3108	* recovering.
3109	*/
3110	if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING &&
3111	migrate_postcopy_ram() && postcopy_pause_incoming(mis)) {
3112	/* Reset f to point to the newly created channel */
3113	f = mis->from_src_file;
3114	error_free_or_abort(errp);
3115	goto retry;
3116	}
3117	}
3118	return ret;
3119	}
3120
3121	int qemu_loadvm_state(QEMUFile f, Error *errp)
3122	{
3123	MigrationState *s = migrate_get_current();
3124	MigrationIncomingState *mis = migration_incoming_get_current();
3125	int ret;
3126
3127	if (qemu_savevm_state_blocked(errp)) {
3128	return -EINVAL;
3129	}
3130
3131	qemu_loadvm_thread_pool_create(mis);
3132
3133	ret = qemu_loadvm_state_header(f, errp);
3134	if (ret) {
3135	return ret;
3136	}
3137
3138	if (qemu_loadvm_state_setup(f, errp) != 0) {
3139	return -EINVAL;
3140	}
3141
3142	cpu_synchronize_all_pre_loadvm();
3143
3144	ret = qemu_loadvm_state_main(f, mis, errp);
3145	qemu_event_set(&mis->main_thread_load_event);
3146
3147	trace_qemu_loadvm_state_post_main(ret);
3148
3149	if (mis->have_listen_thread) {
3150	/*
3151	* Postcopy listen thread still going, don't synchronize the
3152	* cpus yet.
3153	*/
3154	return ret;
3155	}
3156
3157	/* When reaching here, it must be precopy */
3158	if (ret == 0) {
3159	if (migrate_has_error(migrate_get_current()) \|\|
3160	!qemu_loadvm_thread_pool_wait(s, mis)) {
3161	ret = -EINVAL;
3162	error_setg(errp,
3163	"Error while loading vmstate");
3164	} else {
3165	ret = qemu_file_get_error(f);
3166	if (ret < 0) {
3167	error_setg(errp,
3168	"Error while loading vmstate: stream error: %d",
3169	ret);
3170	}
3171	}
3172	}
3173	/*
3174	* Set this flag unconditionally so we'll catch further attempts to
3175	* start additional threads via an appropriate assert()
3176	*/
3177	qatomic_set(&mis->load_threads_abort, true);
3178
3179	/*
3180	* Try to read in the VMDESC section as well, so that dumping tools that
3181	* intercept our migration stream have the chance to see it.
3182	*/
3183
3184	/* We've got to be careful; if we don't read the data and just shut the fd
3185	* then the sender can error if we close while it's still sending.
3186	* We also mustn't read data that isn't there; some transports (RDMA)
3187	* will stall waiting for that data when the source has already closed.
3188	*/
3189	if (ret == 0 && should_send_vmdesc()) {
3190	uint8_t *buf;
3191	uint32_t size;
3192	uint8_t section_type = qemu_get_byte(f);
3193
3194	if (section_type != QEMU_VM_VMDESCRIPTION) {
3195	error_report("Expected vmdescription section, but got %d",
3196	section_type);
3197	/*
3198	* It doesn't seem worth failing at this point since
3199	* we apparently have an otherwise valid VM state
3200	*/
3201	} else {
3202	buf = g_malloc(0x1000);
3203	size = qemu_get_be32(f);
3204
3205	while (size > 0) {
3206	uint32_t read_chunk = MIN(size, 0x1000);
3207	qemu_get_buffer(f, buf, read_chunk);
3208	size -= read_chunk;
3209	}
3210	g_free(buf);
3211	}
3212	}
3213
3214	cpu_synchronize_all_post_init();
3215
3216	return ret;
3217	}
3218
3219	int qemu_load_device_state(QEMUFile f, Error *errp)
3220	{
3221	MigrationIncomingState *mis = migration_incoming_get_current();
3222	int ret;
3223
3224	/* Load QEMU_VM_SECTION_FULL section */
3225	ret = qemu_loadvm_state_main(f, mis, errp);
3226	if (ret < 0) {
3227	return ret;
3228	}
3229
3230	cpu_synchronize_all_post_init();
3231	return 0;
3232	}
3233
3234	static int qemu_loadvm_approve_switchover_legacy(const char *approver)
3235	{
3236	MigrationIncomingState *mis = migration_incoming_get_current();
3237
3238	if (!mis->switchover_ack_pending_num_legacy) {
3239	return -EINVAL;
3240	}
3241
3242	mis->switchover_ack_pending_num_legacy--;
3243	trace_loadvm_approve_switchover_legacy(
3244	approver, mis->switchover_ack_pending_num_legacy);
3245
3246	if (mis->switchover_ack_pending_num_legacy) {
3247	return 0;
3248	}
3249
3250	return migrate_send_rp_switchover_ack(mis);
3251	}
3252
3253	int qemu_loadvm_approve_switchover(const char *approver)
3254	{
3255	MigrationIncomingState *mis = migration_incoming_get_current();
3256
3257	if (!migrate_switchover_ack()) {
3258	return 0;
3259	}
3260
3261	if (migrate_switchover_ack_legacy()) {
3262	return qemu_loadvm_approve_switchover_legacy(approver);
3263	}
3264
3265	trace_loadvm_approve_switchover(approver);
3266
3267	return migrate_send_rp_switchover_ack(mis);
3268	}
3269
3270	bool qemu_loadvm_load_state_buffer(const char *idstr, uint32_t instance_id,
3271	char buf, size_t len, Error *errp)
3272	{
3273	SaveStateEntry *se;
3274
3275	se = find_se(idstr, instance_id);
3276	if (!se) {
3277	error_setg(errp,
3278	"Unknown idstr %s or instance id %u for load state buffer",
3279	idstr, instance_id);
3280	return false;
3281	}
3282
3283	if (!se->ops \|\| !se->ops->load_state_buffer) {
3284	error_setg(errp,
3285	"idstr %s / instance %u has no load state buffer operation",
3286	idstr, instance_id);
3287	return false;
3288	}
3289
3290	return se->ops->load_state_buffer(se->opaque, buf, len, errp);
3291	}
3292
3293	bool save_snapshot(const char name, bool overwrite, const char vmstate,
3294	bool has_devices, strList devices, Error *errp)
3295	{
3296	BlockDriverState *bs;
3297	QEMUSnapshotInfo sn1, *sn = &sn1;
3298	int ret = -1, ret2;
3299	QEMUFile *f;
3300	RunState saved_state = runstate_get();
3301	uint64_t vm_state_size;
3302	g_autoptr(GDateTime) now = g_date_time_new_now_local();
3303
3304	GLOBAL_STATE_CODE();
3305
3306	if (!migrate_can_snapshot(errp)) {
3307	return false;
3308	}
3309
3310	if (migration_is_blocked(errp)) {
3311	return false;
3312	}
3313
3314	if (!replay_can_snapshot()) {
3315	error_setg(errp, "Record/replay does not allow making snapshot "
3316	"right now. Try once more later.");
3317	return false;
3318	}
3319
3320	if (!bdrv_all_can_snapshot(has_devices, devices, errp)) {
3321	return false;
3322	}
3323
3324	/* Delete old snapshots of the same name */
3325	if (name) {
3326	if (overwrite) {
3327	if (bdrv_all_delete_snapshot(name, has_devices,
3328	devices, errp) < 0) {
3329	return false;
3330	}
3331	} else {
3332	ret2 = bdrv_all_has_snapshot(name, has_devices, devices, errp);
3333	if (ret2 < 0) {
3334	return false;
3335	}
3336	if (ret2 == 1) {
3337	error_setg(errp,
3338	"Snapshot '%s' already exists in one or more devices",
3339	name);
3340	return false;
3341	}
3342	}
3343	}
3344
3345	bs = bdrv_all_find_vmstate_bs(vmstate, has_devices, devices, errp);
3346	if (bs == NULL) {
3347	return false;
3348	}
3349
3350	global_state_store();
3351	vm_stop(RUN_STATE_SAVE_VM);
3352
3353	bdrv_drain_all_begin();
3354
3355	memset(sn, 0, sizeof(*sn));
3356
3357	/* fill auxiliary fields */
3358	sn->date_sec = g_date_time_to_unix(now);
3359	sn->date_nsec = g_date_time_get_microsecond(now) * 1000;
3360	sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
3361	if (replay_mode != REPLAY_MODE_NONE) {
3362	sn->icount = replay_get_current_icount();
3363	} else {
3364	sn->icount = -1ULL;
3365	}
3366
3367	if (name) {
3368	pstrcpy(sn->name, sizeof(sn->name), name);
3369	} else {
3370	g_autofree char *autoname = g_date_time_format(now, "vm-%Y%m%d%H%M%S");
3371	pstrcpy(sn->name, sizeof(sn->name), autoname);
3372	}
3373
3374	/* save the VM state */
3375	f = qemu_fopen_bdrv(bs, 1);
3376	if (!f) {
3377	error_setg(errp, "Could not open VM state file");
3378	goto the_end;
3379	}
3380	ret = qemu_savevm_state(f, errp);
3381	vm_state_size = qemu_file_transferred(f);
3382	ret2 = qemu_fclose(f);
3383	if (ret < 0) {
3384	goto the_end;
3385	}
3386	if (ret2 < 0) {
3387	ret = ret2;
3388	goto the_end;
3389	}
3390
3391	ret = bdrv_all_create_snapshot(sn, bs, vm_state_size,
3392	has_devices, devices, errp);
3393	if (ret < 0) {
3394	bdrv_all_delete_snapshot(sn->name, has_devices, devices, NULL);
3395	goto the_end;
3396	}
3397
3398	ret = 0;
3399
3400	the_end:
3401	bdrv_drain_all_end();
3402
3403	vm_resume(saved_state);
3404	return ret == 0;
3405	}
3406
3407	void qmp_xen_save_devices_state(const char *filename, bool has_live, bool live,
3408	Error **errp)
3409	{
3410	QEMUFile *f;
3411	QIOChannelFile *ioc;
3412	int saved_vm_running;
3413	int ret;
3414
3415	if (!has_live) {
3416	/* live default to true so old version of Xen tool stack can have a
3417	* successful live migration */
3418	live = true;
3419	}
3420
3421	saved_vm_running = runstate_is_running();
3422	vm_stop(RUN_STATE_SAVE_VM);
3423	global_state_store_running();
3424
3425	ioc = qio_channel_file_new_path(filename, O_WRONLY \| O_CREAT \| O_TRUNC,
3426	0660, errp);
3427	if (!ioc) {
3428	goto the_end;
3429	}
3430	qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state");
3431	f = qemu_file_new_output(QIO_CHANNEL(ioc));
3432	object_unref(OBJECT(ioc));
3433	qemu_savevm_send_header(f);
3434	ret = qemu_save_device_state(f, errp);
3435	if (ret < 0 \|\| qemu_fclose(f) < 0) {
3436	if (*errp == NULL) {
3437	error_setg(errp, "saving Xen device state failed");
3438	}
3439	} else {
3440	/* libxl calls the QMP command "stop" before calling
3441	* "xen-save-devices-state" and in case of migration failure, libxl
3442	* would call "cont".
3443	* So call bdrv_inactivate_all (release locks) here to let the other
3444	* side of the migration take control of the images.
3445	*/
3446	if (live && !saved_vm_running) {
3447	migration_block_inactivate();
3448	}
3449	}
3450
3451	the_end:
3452	if (saved_vm_running) {
3453	vm_start();
3454	}
3455	}
3456
3457	void qmp_xen_load_devices_state(const char filename, Error *errp)
3458	{
3459	ERRP_GUARD();
3460	QEMUFile *f;
3461	QIOChannelFile *ioc;
3462	int ret;
3463
3464	/* Guest must be paused before loading the device state; the RAM state
3465	* will already have been loaded by xc
3466	*/
3467	if (runstate_is_running()) {
3468	error_setg(errp, "Cannot update device state while vm is running");
3469	return;
3470	}
3471	vm_stop(RUN_STATE_RESTORE_VM);
3472
3473	ioc = qio_channel_file_new_path(filename, O_RDONLY \| O_BINARY, 0, errp);
3474	if (!ioc) {
3475	return;
3476	}
3477	qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state");
3478	f = qemu_file_new_input(QIO_CHANNEL(ioc));
3479	object_unref(OBJECT(ioc));
3480
3481	ret = qemu_loadvm_state(f, errp);
3482	qemu_fclose(f);
3483	if (ret < 0) {
3484	error_prepend(errp, "loading Xen device state failed: ");
3485	}
3486	migration_incoming_state_destroy();
3487	}
3488
3489	bool load_snapshot(const char name, const char vmstate,
3490	bool has_devices, strList devices, Error *errp)
3491	{
3492	BlockDriverState *bs_vm_state;
3493	QEMUSnapshotInfo sn;
3494	QEMUFile *f;
3495	int ret;
3496	MigrationIncomingState *mis = migration_incoming_get_current();
3497
3498	if (!migrate_can_snapshot(errp)) {
3499	return false;
3500	}
3501
3502	if (!bdrv_all_can_snapshot(has_devices, devices, errp)) {
3503	return false;
3504	}
3505	ret = bdrv_all_has_snapshot(name, has_devices, devices, errp);
3506	if (ret < 0) {
3507	return false;
3508	}
3509	if (ret == 0) {
3510	error_setg(errp, "Snapshot '%s' does not exist in one or more devices",
3511	name);
3512	return false;
3513	}
3514
3515	bs_vm_state = bdrv_all_find_vmstate_bs(vmstate, has_devices, devices, errp);
3516	if (!bs_vm_state) {
3517	return false;
3518	}
3519
3520	/* Don't even try to load empty VM states */
3521	ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
3522	if (ret < 0) {
3523	error_setg(errp, "Snapshot can not be found");
3524	return false;
3525	} else if (sn.vm_state_size == 0) {
3526	error_setg(errp, "This is a disk-only snapshot. Revert to it "
3527	" offline using qemu-img");
3528	return false;
3529	}
3530
3531	/*
3532	* Flush the record/replay queue. Now the VM state is going
3533	* to change. Therefore we don't need to preserve its consistency
3534	*/
3535	replay_flush_events();
3536
3537	/* Flush all IO requests so they don't interfere with the new state. */
3538	bdrv_drain_all_begin();
3539
3540	ret = bdrv_all_goto_snapshot(name, has_devices, devices, errp);
3541	if (ret < 0) {
3542	goto err_drain;
3543	}
3544
3545	/* restore the VM state */
3546	f = qemu_fopen_bdrv(bs_vm_state, 0);
3547	if (!f) {
3548	error_setg(errp, "Could not open VM state file");
3549	goto err_drain;
3550	}
3551
3552	qemu_system_reset(SHUTDOWN_CAUSE_SNAPSHOT_LOAD);
3553	mis->from_src_file = f;
3554
3555	if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) {
3556	ret = -EINVAL;
3557	goto err_drain;
3558	}
3559	ret = qemu_loadvm_state(f, errp);
3560	migration_incoming_state_destroy();
3561
3562	bdrv_drain_all_end();
3563
3564	if (ret < 0) {
3565	return false;
3566	}
3567
3568	return true;
3569
3570	err_drain:
3571	bdrv_drain_all_end();
3572	return false;
3573	}
3574
3575	void load_snapshot_resume(RunState state)
3576	{
3577	vm_resume(state);
3578	if (state == RUN_STATE_RUNNING && runstate_get() == RUN_STATE_SUSPENDED) {
3579	qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, &error_abort);
3580	}
3581	}
3582
3583	bool delete_snapshot(const char *name, bool has_devices,
3584	strList devices, Error *errp)
3585	{
3586	if (!bdrv_all_can_snapshot(has_devices, devices, errp)) {
3587	return false;
3588	}
3589
3590	if (bdrv_all_delete_snapshot(name, has_devices, devices, errp) < 0) {
3591	return false;
3592	}
3593
3594	return true;
3595	}
3596
3597	void vmstate_register_ram(MemoryRegion mr, DeviceState dev)
3598	{
3599	qemu_ram_set_idstr(mr->ram_block,
3600	memory_region_name(mr), dev);
3601	qemu_ram_set_migratable(mr->ram_block);
3602	ram_block_add_cpr_blocker(mr->ram_block, &error_fatal);
3603	}
3604
3605	void vmstate_unregister_ram(MemoryRegion mr, DeviceState dev)
3606	{
3607	qemu_ram_unset_idstr(mr->ram_block);
3608	qemu_ram_unset_migratable(mr->ram_block);
3609	ram_block_del_cpr_blocker(mr->ram_block);
3610	}
3611
3612	void vmstate_register_ram_global(MemoryRegion *mr)
3613	{
3614	vmstate_register_ram(mr, NULL);
3615	}
3616
3617	bool vmstate_check_only_migratable(const VMStateDescription *vmsd)
3618	{
3619	/* check needed if --only-migratable is specified */
3620	if (!only_migratable) {
3621	return true;
3622	}
3623
3624	return !(vmsd && vmsd->unmigratable);
3625	}
3626
3627	typedef struct SnapshotJob {
3628	Job common;
3629	char *tag;
3630	char *vmstate;
3631	strList *devices;
3632	Coroutine *co;
3633	Error **errp;
3634	bool ret;
3635	} SnapshotJob;
3636
3637	static void qmp_snapshot_job_free(SnapshotJob *s)
3638	{
3639	g_free(s->tag);
3640	g_free(s->vmstate);
3641	qapi_free_strList(s->devices);
3642	}
3643
3644
3645	static void snapshot_load_job_bh(void *opaque)
3646	{
3647	Job *job = opaque;
3648	SnapshotJob *s = container_of(job, SnapshotJob, common);
3649	RunState orig_state = runstate_get();
3650
3651	job_progress_set_remaining(&s->common, 1);
3652
3653	vm_stop(RUN_STATE_RESTORE_VM);
3654
3655	s->ret = load_snapshot(s->tag, s->vmstate, true, s->devices, s->errp);
3656	if (s->ret) {
3657	load_snapshot_resume(orig_state);
3658	}
3659
3660	job_progress_update(&s->common, 1);
3661
3662	qmp_snapshot_job_free(s);
3663	aio_co_wake(s->co);
3664	}
3665
3666	static void snapshot_save_job_bh(void *opaque)
3667	{
3668	Job *job = opaque;
3669	SnapshotJob *s = container_of(job, SnapshotJob, common);
3670
3671	job_progress_set_remaining(&s->common, 1);
3672	s->ret = save_snapshot(s->tag, false, s->vmstate,
3673	true, s->devices, s->errp);
3674	job_progress_update(&s->common, 1);
3675
3676	qmp_snapshot_job_free(s);
3677	aio_co_wake(s->co);
3678	}
3679
3680	static void snapshot_delete_job_bh(void *opaque)
3681	{
3682	Job *job = opaque;
3683	SnapshotJob *s = container_of(job, SnapshotJob, common);
3684
3685	job_progress_set_remaining(&s->common, 1);
3686	s->ret = delete_snapshot(s->tag, true, s->devices, s->errp);
3687	job_progress_update(&s->common, 1);
3688
3689	qmp_snapshot_job_free(s);
3690	aio_co_wake(s->co);
3691	}
3692
3693	static int coroutine_fn snapshot_save_job_run(Job job, Error *errp)
3694	{
3695	SnapshotJob *s = container_of(job, SnapshotJob, common);
3696	s->errp = errp;
3697	s->co = qemu_coroutine_self();
3698	aio_bh_schedule_oneshot(qemu_get_aio_context(),
3699	snapshot_save_job_bh, job);
3700	qemu_coroutine_yield();
3701	return s->ret ? 0 : -1;
3702	}
3703
3704	static int coroutine_fn snapshot_load_job_run(Job job, Error *errp)
3705	{
3706	SnapshotJob *s = container_of(job, SnapshotJob, common);
3707	s->errp = errp;
3708	s->co = qemu_coroutine_self();
3709	aio_bh_schedule_oneshot(qemu_get_aio_context(),
3710	snapshot_load_job_bh, job);
3711	qemu_coroutine_yield();
3712	return s->ret ? 0 : -1;
3713	}
3714
3715	static int coroutine_fn snapshot_delete_job_run(Job job, Error *errp)
3716	{
3717	SnapshotJob *s = container_of(job, SnapshotJob, common);
3718	s->errp = errp;
3719	s->co = qemu_coroutine_self();
3720	aio_bh_schedule_oneshot(qemu_get_aio_context(),
3721	snapshot_delete_job_bh, job);
3722	qemu_coroutine_yield();
3723	return s->ret ? 0 : -1;
3724	}
3725
3726
3727	static const JobDriver snapshot_load_job_driver = {
3728	.instance_size = sizeof(SnapshotJob),
3729	.job_type = JOB_TYPE_SNAPSHOT_LOAD,
3730	.run = snapshot_load_job_run,
3731	};
3732
3733	static const JobDriver snapshot_save_job_driver = {
3734	.instance_size = sizeof(SnapshotJob),
3735	.job_type = JOB_TYPE_SNAPSHOT_SAVE,
3736	.run = snapshot_save_job_run,
3737	};
3738
3739	static const JobDriver snapshot_delete_job_driver = {
3740	.instance_size = sizeof(SnapshotJob),
3741	.job_type = JOB_TYPE_SNAPSHOT_DELETE,
3742	.run = snapshot_delete_job_run,
3743	};
3744
3745
3746	void qmp_snapshot_save(const char *job_id,
3747	const char *tag,
3748	const char *vmstate,
3749	strList *devices,
3750	Error **errp)
3751	{
3752	SnapshotJob *s;
3753
3754	s = job_create(job_id, &snapshot_save_job_driver, NULL,
3755	qemu_get_aio_context(), JOB_MANUAL_DISMISS,
3756	NULL, NULL, errp);
3757	if (!s) {
3758	return;
3759	}
3760
3761	s->tag = g_strdup(tag);
3762	s->vmstate = g_strdup(vmstate);
3763	s->devices = QAPI_CLONE(strList, devices);
3764
3765	job_start(&s->common);
3766	}
3767
3768	void qmp_snapshot_load(const char *job_id,
3769	const char *tag,
3770	const char *vmstate,
3771	strList *devices,
3772	Error **errp)
3773	{
3774	SnapshotJob *s;
3775
3776	s = job_create(job_id, &snapshot_load_job_driver, NULL,
3777	qemu_get_aio_context(), JOB_MANUAL_DISMISS,
3778	NULL, NULL, errp);
3779	if (!s) {
3780	return;
3781	}
3782
3783	s->tag = g_strdup(tag);
3784	s->vmstate = g_strdup(vmstate);
3785	s->devices = QAPI_CLONE(strList, devices);
3786
3787	job_start(&s->common);
3788	}
3789
3790	void qmp_snapshot_delete(const char *job_id,
3791	const char *tag,
3792	strList *devices,
3793	Error **errp)
3794	{
3795	SnapshotJob *s;
3796
3797	s = job_create(job_id, &snapshot_delete_job_driver, NULL,
3798	qemu_get_aio_context(), JOB_MANUAL_DISMISS,
3799	NULL, NULL, errp);
3800	if (!s) {
3801	return;
3802	}
3803
3804	s->tag = g_strdup(tag);
3805	s->devices = QAPI_CLONE(strList, devices);
3806
3807	job_start(&s->common);
3808	}