target/s390x: Fix typo
[qemu/ar7.git] / migration / savevm.c
blob5ecd264134108bf72c6bddbddd12f2bdf4087538
1 /*
2 * QEMU System Emulator
4 * Copyright (c) 2003-2008 Fabrice Bellard
5 * Copyright (c) 2009-2015 Red Hat Inc
7 * Authors:
8 * Juan Quintela <quintela@redhat.com>
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:
17 * The above copyright notice and this permission notice shall be included in
18 * all copies or substantial portions of the Software.
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.
29 #include "qemu/osdep.h"
30 #include "cpu.h"
31 #include "hw/boards.h"
32 #include "hw/hw.h"
33 #include "hw/qdev.h"
34 #include "hw/xen/xen.h"
35 #include "net/net.h"
36 #include "monitor/monitor.h"
37 #include "sysemu/sysemu.h"
38 #include "qemu/timer.h"
39 #include "audio/audio.h"
40 #include "migration/migration.h"
41 #include "migration/postcopy-ram.h"
42 #include "qapi/qmp/qerror.h"
43 #include "qemu/error-report.h"
44 #include "qemu/sockets.h"
45 #include "qemu/queue.h"
46 #include "sysemu/cpus.h"
47 #include "exec/memory.h"
48 #include "qmp-commands.h"
49 #include "trace.h"
50 #include "qemu/bitops.h"
51 #include "qemu/iov.h"
52 #include "block/snapshot.h"
53 #include "block/qapi.h"
54 #include "qemu/cutils.h"
55 #include "io/channel-buffer.h"
56 #include "io/channel-file.h"
58 #ifndef ETH_P_RARP
59 #define ETH_P_RARP 0x8035
60 #endif
61 #define ARP_HTYPE_ETH 0x0001
62 #define ARP_PTYPE_IP 0x0800
63 #define ARP_OP_REQUEST_REV 0x3
65 const unsigned int postcopy_ram_discard_version = 0;
67 static bool skip_section_footers;
69 static struct mig_cmd_args {
70 ssize_t len; /* -1 = variable */
71 const char *name;
72 } mig_cmd_args[] = {
73 [MIG_CMD_INVALID] = { .len = -1, .name = "INVALID" },
74 [MIG_CMD_OPEN_RETURN_PATH] = { .len = 0, .name = "OPEN_RETURN_PATH" },
75 [MIG_CMD_PING] = { .len = sizeof(uint32_t), .name = "PING" },
76 [MIG_CMD_POSTCOPY_ADVISE] = { .len = 16, .name = "POSTCOPY_ADVISE" },
77 [MIG_CMD_POSTCOPY_LISTEN] = { .len = 0, .name = "POSTCOPY_LISTEN" },
78 [MIG_CMD_POSTCOPY_RUN] = { .len = 0, .name = "POSTCOPY_RUN" },
79 [MIG_CMD_POSTCOPY_RAM_DISCARD] = {
80 .len = -1, .name = "POSTCOPY_RAM_DISCARD" },
81 [MIG_CMD_PACKAGED] = { .len = 4, .name = "PACKAGED" },
82 [MIG_CMD_MAX] = { .len = -1, .name = "MAX" },
85 static int announce_self_create(uint8_t *buf,
86 uint8_t *mac_addr)
88 /* Ethernet header. */
89 memset(buf, 0xff, 6); /* destination MAC addr */
90 memcpy(buf + 6, mac_addr, 6); /* source MAC addr */
91 *(uint16_t *)(buf + 12) = htons(ETH_P_RARP); /* ethertype */
93 /* RARP header. */
94 *(uint16_t *)(buf + 14) = htons(ARP_HTYPE_ETH); /* hardware addr space */
95 *(uint16_t *)(buf + 16) = htons(ARP_PTYPE_IP); /* protocol addr space */
96 *(buf + 18) = 6; /* hardware addr length (ethernet) */
97 *(buf + 19) = 4; /* protocol addr length (IPv4) */
98 *(uint16_t *)(buf + 20) = htons(ARP_OP_REQUEST_REV); /* opcode */
99 memcpy(buf + 22, mac_addr, 6); /* source hw addr */
100 memset(buf + 28, 0x00, 4); /* source protocol addr */
101 memcpy(buf + 32, mac_addr, 6); /* target hw addr */
102 memset(buf + 38, 0x00, 4); /* target protocol addr */
104 /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */
105 memset(buf + 42, 0x00, 18);
107 return 60; /* len (FCS will be added by hardware) */
110 static void qemu_announce_self_iter(NICState *nic, void *opaque)
112 uint8_t buf[60];
113 int len;
115 trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic->conf->macaddr));
116 len = announce_self_create(buf, nic->conf->macaddr.a);
118 qemu_send_packet_raw(qemu_get_queue(nic), buf, len);
122 static void qemu_announce_self_once(void *opaque)
124 static int count = SELF_ANNOUNCE_ROUNDS;
125 QEMUTimer *timer = *(QEMUTimer **)opaque;
127 qemu_foreach_nic(qemu_announce_self_iter, NULL);
129 if (--count) {
130 /* delay 50ms, 150ms, 250ms, ... */
131 timer_mod(timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) +
132 self_announce_delay(count));
133 } else {
134 timer_del(timer);
135 timer_free(timer);
139 void qemu_announce_self(void)
141 static QEMUTimer *timer;
142 timer = timer_new_ms(QEMU_CLOCK_REALTIME, qemu_announce_self_once, &timer);
143 qemu_announce_self_once(&timer);
146 /***********************************************************/
147 /* savevm/loadvm support */
149 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt,
150 int64_t pos)
152 int ret;
153 QEMUIOVector qiov;
155 qemu_iovec_init_external(&qiov, iov, iovcnt);
156 ret = bdrv_writev_vmstate(opaque, &qiov, pos);
157 if (ret < 0) {
158 return ret;
161 return qiov.size;
164 static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos,
165 size_t size)
167 return bdrv_load_vmstate(opaque, buf, pos, size);
170 static int bdrv_fclose(void *opaque)
172 return bdrv_flush(opaque);
175 static const QEMUFileOps bdrv_read_ops = {
176 .get_buffer = block_get_buffer,
177 .close = bdrv_fclose
180 static const QEMUFileOps bdrv_write_ops = {
181 .writev_buffer = block_writev_buffer,
182 .close = bdrv_fclose
185 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
187 if (is_writable) {
188 return qemu_fopen_ops(bs, &bdrv_write_ops);
190 return qemu_fopen_ops(bs, &bdrv_read_ops);
194 /* QEMUFile timer support.
195 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
198 void timer_put(QEMUFile *f, QEMUTimer *ts)
200 uint64_t expire_time;
202 expire_time = timer_expire_time_ns(ts);
203 qemu_put_be64(f, expire_time);
206 void timer_get(QEMUFile *f, QEMUTimer *ts)
208 uint64_t expire_time;
210 expire_time = qemu_get_be64(f);
211 if (expire_time != -1) {
212 timer_mod_ns(ts, expire_time);
213 } else {
214 timer_del(ts);
219 /* VMState timer support.
220 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
223 static int get_timer(QEMUFile *f, void *pv, size_t size, VMStateField *field)
225 QEMUTimer *v = pv;
226 timer_get(f, v);
227 return 0;
230 static int put_timer(QEMUFile *f, void *pv, size_t size, VMStateField *field,
231 QJSON *vmdesc)
233 QEMUTimer *v = pv;
234 timer_put(f, v);
236 return 0;
239 const VMStateInfo vmstate_info_timer = {
240 .name = "timer",
241 .get = get_timer,
242 .put = put_timer,
246 typedef struct CompatEntry {
247 char idstr[256];
248 int instance_id;
249 } CompatEntry;
251 typedef struct SaveStateEntry {
252 QTAILQ_ENTRY(SaveStateEntry) entry;
253 char idstr[256];
254 int instance_id;
255 int alias_id;
256 int version_id;
257 int section_id;
258 SaveVMHandlers *ops;
259 const VMStateDescription *vmsd;
260 void *opaque;
261 CompatEntry *compat;
262 int is_ram;
263 } SaveStateEntry;
265 typedef struct SaveState {
266 QTAILQ_HEAD(, SaveStateEntry) handlers;
267 int global_section_id;
268 bool skip_configuration;
269 uint32_t len;
270 const char *name;
271 uint32_t target_page_bits;
272 } SaveState;
274 static SaveState savevm_state = {
275 .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
276 .global_section_id = 0,
277 .skip_configuration = false,
280 void savevm_skip_configuration(void)
282 savevm_state.skip_configuration = true;
286 static void configuration_pre_save(void *opaque)
288 SaveState *state = opaque;
289 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
291 state->len = strlen(current_name);
292 state->name = current_name;
293 state->target_page_bits = TARGET_PAGE_BITS;
296 static int configuration_pre_load(void *opaque)
298 SaveState *state = opaque;
300 /* If there is no target-page-bits subsection it means the source
301 * predates the variable-target-page-bits support and is using the
302 * minimum possible value for this CPU.
304 state->target_page_bits = TARGET_PAGE_BITS_MIN;
305 return 0;
308 static int configuration_post_load(void *opaque, int version_id)
310 SaveState *state = opaque;
311 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
313 if (strncmp(state->name, current_name, state->len) != 0) {
314 error_report("Machine type received is '%.*s' and local is '%s'",
315 (int) state->len, state->name, current_name);
316 return -EINVAL;
319 if (state->target_page_bits != TARGET_PAGE_BITS) {
320 error_report("Received TARGET_PAGE_BITS is %d but local is %d",
321 state->target_page_bits, TARGET_PAGE_BITS);
322 return -EINVAL;
325 return 0;
328 /* The target-page-bits subsection is present only if the
329 * target page size is not the same as the default (ie the
330 * minimum page size for a variable-page-size guest CPU).
331 * If it is present then it contains the actual target page
332 * bits for the machine, and migration will fail if the
333 * two ends don't agree about it.
335 static bool vmstate_target_page_bits_needed(void *opaque)
337 return TARGET_PAGE_BITS > TARGET_PAGE_BITS_MIN;
340 static const VMStateDescription vmstate_target_page_bits = {
341 .name = "configuration/target-page-bits",
342 .version_id = 1,
343 .minimum_version_id = 1,
344 .needed = vmstate_target_page_bits_needed,
345 .fields = (VMStateField[]) {
346 VMSTATE_UINT32(target_page_bits, SaveState),
347 VMSTATE_END_OF_LIST()
351 static const VMStateDescription vmstate_configuration = {
352 .name = "configuration",
353 .version_id = 1,
354 .pre_load = configuration_pre_load,
355 .post_load = configuration_post_load,
356 .pre_save = configuration_pre_save,
357 .fields = (VMStateField[]) {
358 VMSTATE_UINT32(len, SaveState),
359 VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len),
360 VMSTATE_END_OF_LIST()
362 .subsections = (const VMStateDescription*[]) {
363 &vmstate_target_page_bits,
364 NULL
368 static void dump_vmstate_vmsd(FILE *out_file,
369 const VMStateDescription *vmsd, int indent,
370 bool is_subsection);
372 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
373 int indent)
375 fprintf(out_file, "%*s{\n", indent, "");
376 indent += 2;
377 fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
378 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
379 field->version_id);
380 fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
381 field->field_exists ? "true" : "false");
382 fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
383 if (field->vmsd != NULL) {
384 fprintf(out_file, ",\n");
385 dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
387 fprintf(out_file, "\n%*s}", indent - 2, "");
390 static void dump_vmstate_vmss(FILE *out_file,
391 const VMStateDescription **subsection,
392 int indent)
394 if (*subsection != NULL) {
395 dump_vmstate_vmsd(out_file, *subsection, indent, true);
399 static void dump_vmstate_vmsd(FILE *out_file,
400 const VMStateDescription *vmsd, int indent,
401 bool is_subsection)
403 if (is_subsection) {
404 fprintf(out_file, "%*s{\n", indent, "");
405 } else {
406 fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
408 indent += 2;
409 fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
410 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
411 vmsd->version_id);
412 fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
413 vmsd->minimum_version_id);
414 if (vmsd->fields != NULL) {
415 const VMStateField *field = vmsd->fields;
416 bool first;
418 fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
419 first = true;
420 while (field->name != NULL) {
421 if (field->flags & VMS_MUST_EXIST) {
422 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
423 field++;
424 continue;
426 if (!first) {
427 fprintf(out_file, ",\n");
429 dump_vmstate_vmsf(out_file, field, indent + 2);
430 field++;
431 first = false;
433 fprintf(out_file, "\n%*s]", indent, "");
435 if (vmsd->subsections != NULL) {
436 const VMStateDescription **subsection = vmsd->subsections;
437 bool first;
439 fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
440 first = true;
441 while (*subsection != NULL) {
442 if (!first) {
443 fprintf(out_file, ",\n");
445 dump_vmstate_vmss(out_file, subsection, indent + 2);
446 subsection++;
447 first = false;
449 fprintf(out_file, "\n%*s]", indent, "");
451 fprintf(out_file, "\n%*s}", indent - 2, "");
454 static void dump_machine_type(FILE *out_file)
456 MachineClass *mc;
458 mc = MACHINE_GET_CLASS(current_machine);
460 fprintf(out_file, " \"vmschkmachine\": {\n");
461 fprintf(out_file, " \"Name\": \"%s\"\n", mc->name);
462 fprintf(out_file, " },\n");
465 void dump_vmstate_json_to_file(FILE *out_file)
467 GSList *list, *elt;
468 bool first;
470 fprintf(out_file, "{\n");
471 dump_machine_type(out_file);
473 first = true;
474 list = object_class_get_list(TYPE_DEVICE, true);
475 for (elt = list; elt; elt = elt->next) {
476 DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
477 TYPE_DEVICE);
478 const char *name;
479 int indent = 2;
481 if (!dc->vmsd) {
482 continue;
485 if (!first) {
486 fprintf(out_file, ",\n");
488 name = object_class_get_name(OBJECT_CLASS(dc));
489 fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
490 indent += 2;
491 fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
492 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
493 dc->vmsd->version_id);
494 fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
495 dc->vmsd->minimum_version_id);
497 dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
499 fprintf(out_file, "\n%*s}", indent - 2, "");
500 first = false;
502 fprintf(out_file, "\n}\n");
503 fclose(out_file);
506 static int calculate_new_instance_id(const char *idstr)
508 SaveStateEntry *se;
509 int instance_id = 0;
511 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
512 if (strcmp(idstr, se->idstr) == 0
513 && instance_id <= se->instance_id) {
514 instance_id = se->instance_id + 1;
517 return instance_id;
520 static int calculate_compat_instance_id(const char *idstr)
522 SaveStateEntry *se;
523 int instance_id = 0;
525 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
526 if (!se->compat) {
527 continue;
530 if (strcmp(idstr, se->compat->idstr) == 0
531 && instance_id <= se->compat->instance_id) {
532 instance_id = se->compat->instance_id + 1;
535 return instance_id;
538 static inline MigrationPriority save_state_priority(SaveStateEntry *se)
540 if (se->vmsd) {
541 return se->vmsd->priority;
543 return MIG_PRI_DEFAULT;
546 static void savevm_state_handler_insert(SaveStateEntry *nse)
548 MigrationPriority priority = save_state_priority(nse);
549 SaveStateEntry *se;
551 assert(priority <= MIG_PRI_MAX);
553 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
554 if (save_state_priority(se) < priority) {
555 break;
559 if (se) {
560 QTAILQ_INSERT_BEFORE(se, nse, entry);
561 } else {
562 QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry);
566 /* TODO: Individual devices generally have very little idea about the rest
567 of the system, so instance_id should be removed/replaced.
568 Meanwhile pass -1 as instance_id if you do not already have a clearly
569 distinguishing id for all instances of your device class. */
570 int register_savevm_live(DeviceState *dev,
571 const char *idstr,
572 int instance_id,
573 int version_id,
574 SaveVMHandlers *ops,
575 void *opaque)
577 SaveStateEntry *se;
579 se = g_new0(SaveStateEntry, 1);
580 se->version_id = version_id;
581 se->section_id = savevm_state.global_section_id++;
582 se->ops = ops;
583 se->opaque = opaque;
584 se->vmsd = NULL;
585 /* if this is a live_savem then set is_ram */
586 if (ops->save_live_setup != NULL) {
587 se->is_ram = 1;
590 if (dev) {
591 char *id = qdev_get_dev_path(dev);
592 if (id) {
593 if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
594 sizeof(se->idstr)) {
595 error_report("Path too long for VMState (%s)", id);
596 g_free(id);
597 g_free(se);
599 return -1;
601 g_free(id);
603 se->compat = g_new0(CompatEntry, 1);
604 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), idstr);
605 se->compat->instance_id = instance_id == -1 ?
606 calculate_compat_instance_id(idstr) : instance_id;
607 instance_id = -1;
610 pstrcat(se->idstr, sizeof(se->idstr), idstr);
612 if (instance_id == -1) {
613 se->instance_id = calculate_new_instance_id(se->idstr);
614 } else {
615 se->instance_id = instance_id;
617 assert(!se->compat || se->instance_id == 0);
618 savevm_state_handler_insert(se);
619 return 0;
622 int register_savevm(DeviceState *dev,
623 const char *idstr,
624 int instance_id,
625 int version_id,
626 SaveStateHandler *save_state,
627 LoadStateHandler *load_state,
628 void *opaque)
630 SaveVMHandlers *ops = g_new0(SaveVMHandlers, 1);
631 ops->save_state = save_state;
632 ops->load_state = load_state;
633 return register_savevm_live(dev, idstr, instance_id, version_id,
634 ops, opaque);
637 void unregister_savevm(DeviceState *dev, const char *idstr, void *opaque)
639 SaveStateEntry *se, *new_se;
640 char id[256] = "";
642 if (dev) {
643 char *path = qdev_get_dev_path(dev);
644 if (path) {
645 pstrcpy(id, sizeof(id), path);
646 pstrcat(id, sizeof(id), "/");
647 g_free(path);
650 pstrcat(id, sizeof(id), idstr);
652 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
653 if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
654 QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
655 g_free(se->compat);
656 g_free(se->ops);
657 g_free(se);
662 int vmstate_register_with_alias_id(DeviceState *dev, int instance_id,
663 const VMStateDescription *vmsd,
664 void *opaque, int alias_id,
665 int required_for_version,
666 Error **errp)
668 SaveStateEntry *se;
670 /* If this triggers, alias support can be dropped for the vmsd. */
671 assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
673 se = g_new0(SaveStateEntry, 1);
674 se->version_id = vmsd->version_id;
675 se->section_id = savevm_state.global_section_id++;
676 se->opaque = opaque;
677 se->vmsd = vmsd;
678 se->alias_id = alias_id;
680 if (dev) {
681 char *id = qdev_get_dev_path(dev);
682 if (id) {
683 if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
684 sizeof(se->idstr)) {
685 error_setg(errp, "Path too long for VMState (%s)", id);
686 g_free(id);
687 g_free(se);
689 return -1;
692 se->compat = g_new0(CompatEntry, 1);
693 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
694 se->compat->instance_id = instance_id == -1 ?
695 calculate_compat_instance_id(vmsd->name) : instance_id;
696 instance_id = -1;
699 pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
701 if (instance_id == -1) {
702 se->instance_id = calculate_new_instance_id(se->idstr);
703 } else {
704 se->instance_id = instance_id;
706 assert(!se->compat || se->instance_id == 0);
707 savevm_state_handler_insert(se);
708 return 0;
711 void vmstate_unregister(DeviceState *dev, const VMStateDescription *vmsd,
712 void *opaque)
714 SaveStateEntry *se, *new_se;
716 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
717 if (se->vmsd == vmsd && se->opaque == opaque) {
718 QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
719 g_free(se->compat);
720 g_free(se);
725 static int vmstate_load(QEMUFile *f, SaveStateEntry *se, int version_id)
727 trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
728 if (!se->vmsd) { /* Old style */
729 return se->ops->load_state(f, se->opaque, version_id);
731 return vmstate_load_state(f, se->vmsd, se->opaque, version_id);
734 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
736 int64_t old_offset, size;
738 old_offset = qemu_ftell_fast(f);
739 se->ops->save_state(f, se->opaque);
740 size = qemu_ftell_fast(f) - old_offset;
742 if (vmdesc) {
743 json_prop_int(vmdesc, "size", size);
744 json_start_array(vmdesc, "fields");
745 json_start_object(vmdesc, NULL);
746 json_prop_str(vmdesc, "name", "data");
747 json_prop_int(vmdesc, "size", size);
748 json_prop_str(vmdesc, "type", "buffer");
749 json_end_object(vmdesc);
750 json_end_array(vmdesc);
754 static void vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
756 trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
757 if (!se->vmsd) {
758 vmstate_save_old_style(f, se, vmdesc);
759 return;
761 vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
764 void savevm_skip_section_footers(void)
766 skip_section_footers = true;
770 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
772 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
773 uint8_t section_type)
775 qemu_put_byte(f, section_type);
776 qemu_put_be32(f, se->section_id);
778 if (section_type == QEMU_VM_SECTION_FULL ||
779 section_type == QEMU_VM_SECTION_START) {
780 /* ID string */
781 size_t len = strlen(se->idstr);
782 qemu_put_byte(f, len);
783 qemu_put_buffer(f, (uint8_t *)se->idstr, len);
785 qemu_put_be32(f, se->instance_id);
786 qemu_put_be32(f, se->version_id);
791 * Write a footer onto device sections that catches cases misformatted device
792 * sections.
794 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
796 if (!skip_section_footers) {
797 qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
798 qemu_put_be32(f, se->section_id);
803 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
804 * command and associated data.
806 * @f: File to send command on
807 * @command: Command type to send
808 * @len: Length of associated data
809 * @data: Data associated with command.
811 void qemu_savevm_command_send(QEMUFile *f,
812 enum qemu_vm_cmd command,
813 uint16_t len,
814 uint8_t *data)
816 trace_savevm_command_send(command, len);
817 qemu_put_byte(f, QEMU_VM_COMMAND);
818 qemu_put_be16(f, (uint16_t)command);
819 qemu_put_be16(f, len);
820 qemu_put_buffer(f, data, len);
821 qemu_fflush(f);
824 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
826 uint32_t buf;
828 trace_savevm_send_ping(value);
829 buf = cpu_to_be32(value);
830 qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
833 void qemu_savevm_send_open_return_path(QEMUFile *f)
835 trace_savevm_send_open_return_path();
836 qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
839 /* We have a buffer of data to send; we don't want that all to be loaded
840 * by the command itself, so the command contains just the length of the
841 * extra buffer that we then send straight after it.
842 * TODO: Must be a better way to organise that
844 * Returns:
845 * 0 on success
846 * -ve on error
848 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len)
850 uint32_t tmp;
852 if (len > MAX_VM_CMD_PACKAGED_SIZE) {
853 error_report("%s: Unreasonably large packaged state: %zu",
854 __func__, len);
855 return -1;
858 tmp = cpu_to_be32(len);
860 trace_qemu_savevm_send_packaged();
861 qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
863 qemu_put_buffer(f, buf, len);
865 return 0;
868 /* Send prior to any postcopy transfer */
869 void qemu_savevm_send_postcopy_advise(QEMUFile *f)
871 uint64_t tmp[2];
872 tmp[0] = cpu_to_be64(getpagesize());
873 tmp[1] = cpu_to_be64(1ul << qemu_target_page_bits());
875 trace_qemu_savevm_send_postcopy_advise();
876 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 16, (uint8_t *)tmp);
879 /* Sent prior to starting the destination running in postcopy, discard pages
880 * that have already been sent but redirtied on the source.
881 * CMD_POSTCOPY_RAM_DISCARD consist of:
882 * byte version (0)
883 * byte Length of name field (not including 0)
884 * n x byte RAM block name
885 * byte 0 terminator (just for safety)
886 * n x Byte ranges within the named RAMBlock
887 * be64 Start of the range
888 * be64 Length
890 * name: RAMBlock name that these entries are part of
891 * len: Number of page entries
892 * start_list: 'len' addresses
893 * length_list: 'len' addresses
896 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
897 uint16_t len,
898 uint64_t *start_list,
899 uint64_t *length_list)
901 uint8_t *buf;
902 uint16_t tmplen;
903 uint16_t t;
904 size_t name_len = strlen(name);
906 trace_qemu_savevm_send_postcopy_ram_discard(name, len);
907 assert(name_len < 256);
908 buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
909 buf[0] = postcopy_ram_discard_version;
910 buf[1] = name_len;
911 memcpy(buf + 2, name, name_len);
912 tmplen = 2 + name_len;
913 buf[tmplen++] = '\0';
915 for (t = 0; t < len; t++) {
916 stq_be_p(buf + tmplen, start_list[t]);
917 tmplen += 8;
918 stq_be_p(buf + tmplen, length_list[t]);
919 tmplen += 8;
921 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
922 g_free(buf);
925 /* Get the destination into a state where it can receive postcopy data. */
926 void qemu_savevm_send_postcopy_listen(QEMUFile *f)
928 trace_savevm_send_postcopy_listen();
929 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
932 /* Kick the destination into running */
933 void qemu_savevm_send_postcopy_run(QEMUFile *f)
935 trace_savevm_send_postcopy_run();
936 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
939 bool qemu_savevm_state_blocked(Error **errp)
941 SaveStateEntry *se;
943 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
944 if (se->vmsd && se->vmsd->unmigratable) {
945 error_setg(errp, "State blocked by non-migratable device '%s'",
946 se->idstr);
947 return true;
950 return false;
953 static bool enforce_config_section(void)
955 MachineState *machine = MACHINE(qdev_get_machine());
956 return machine->enforce_config_section;
959 void qemu_savevm_state_header(QEMUFile *f)
961 trace_savevm_state_header();
962 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
963 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
965 if (!savevm_state.skip_configuration || enforce_config_section()) {
966 qemu_put_byte(f, QEMU_VM_CONFIGURATION);
967 vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
972 void qemu_savevm_state_begin(QEMUFile *f,
973 const MigrationParams *params)
975 SaveStateEntry *se;
976 int ret;
978 trace_savevm_state_begin();
979 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
980 if (!se->ops || !se->ops->set_params) {
981 continue;
983 se->ops->set_params(params, se->opaque);
986 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
987 if (!se->ops || !se->ops->save_live_setup) {
988 continue;
990 if (se->ops && se->ops->is_active) {
991 if (!se->ops->is_active(se->opaque)) {
992 continue;
995 save_section_header(f, se, QEMU_VM_SECTION_START);
997 ret = se->ops->save_live_setup(f, se->opaque);
998 save_section_footer(f, se);
999 if (ret < 0) {
1000 qemu_file_set_error(f, ret);
1001 break;
1007 * this function has three return values:
1008 * negative: there was one error, and we have -errno.
1009 * 0 : We haven't finished, caller have to go again
1010 * 1 : We have finished, we can go to complete phase
1012 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
1014 SaveStateEntry *se;
1015 int ret = 1;
1017 trace_savevm_state_iterate();
1018 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1019 if (!se->ops || !se->ops->save_live_iterate) {
1020 continue;
1022 if (se->ops && se->ops->is_active) {
1023 if (!se->ops->is_active(se->opaque)) {
1024 continue;
1028 * In the postcopy phase, any device that doesn't know how to
1029 * do postcopy should have saved it's state in the _complete
1030 * call that's already run, it might get confused if we call
1031 * iterate afterwards.
1033 if (postcopy && !se->ops->save_live_complete_postcopy) {
1034 continue;
1036 if (qemu_file_rate_limit(f)) {
1037 return 0;
1039 trace_savevm_section_start(se->idstr, se->section_id);
1041 save_section_header(f, se, QEMU_VM_SECTION_PART);
1043 ret = se->ops->save_live_iterate(f, se->opaque);
1044 trace_savevm_section_end(se->idstr, se->section_id, ret);
1045 save_section_footer(f, se);
1047 if (ret < 0) {
1048 qemu_file_set_error(f, ret);
1050 if (ret <= 0) {
1051 /* Do not proceed to the next vmstate before this one reported
1052 completion of the current stage. This serializes the migration
1053 and reduces the probability that a faster changing state is
1054 synchronized over and over again. */
1055 break;
1058 return ret;
1061 static bool should_send_vmdesc(void)
1063 MachineState *machine = MACHINE(qdev_get_machine());
1064 bool in_postcopy = migration_in_postcopy(migrate_get_current());
1065 return !machine->suppress_vmdesc && !in_postcopy;
1069 * Calls the save_live_complete_postcopy methods
1070 * causing the last few pages to be sent immediately and doing any associated
1071 * cleanup.
1072 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1073 * all the other devices, but that happens at the point we switch to postcopy.
1075 void qemu_savevm_state_complete_postcopy(QEMUFile *f)
1077 SaveStateEntry *se;
1078 int ret;
1080 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1081 if (!se->ops || !se->ops->save_live_complete_postcopy) {
1082 continue;
1084 if (se->ops && se->ops->is_active) {
1085 if (!se->ops->is_active(se->opaque)) {
1086 continue;
1089 trace_savevm_section_start(se->idstr, se->section_id);
1090 /* Section type */
1091 qemu_put_byte(f, QEMU_VM_SECTION_END);
1092 qemu_put_be32(f, se->section_id);
1094 ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1095 trace_savevm_section_end(se->idstr, se->section_id, ret);
1096 save_section_footer(f, se);
1097 if (ret < 0) {
1098 qemu_file_set_error(f, ret);
1099 return;
1103 qemu_put_byte(f, QEMU_VM_EOF);
1104 qemu_fflush(f);
1107 void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only)
1109 QJSON *vmdesc;
1110 int vmdesc_len;
1111 SaveStateEntry *se;
1112 int ret;
1113 bool in_postcopy = migration_in_postcopy(migrate_get_current());
1115 trace_savevm_state_complete_precopy();
1117 cpu_synchronize_all_states();
1119 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1120 if (!se->ops ||
1121 (in_postcopy && se->ops->save_live_complete_postcopy) ||
1122 (in_postcopy && !iterable_only) ||
1123 !se->ops->save_live_complete_precopy) {
1124 continue;
1127 if (se->ops && se->ops->is_active) {
1128 if (!se->ops->is_active(se->opaque)) {
1129 continue;
1132 trace_savevm_section_start(se->idstr, se->section_id);
1134 save_section_header(f, se, QEMU_VM_SECTION_END);
1136 ret = se->ops->save_live_complete_precopy(f, se->opaque);
1137 trace_savevm_section_end(se->idstr, se->section_id, ret);
1138 save_section_footer(f, se);
1139 if (ret < 0) {
1140 qemu_file_set_error(f, ret);
1141 return;
1145 if (iterable_only) {
1146 return;
1149 vmdesc = qjson_new();
1150 json_prop_int(vmdesc, "page_size", TARGET_PAGE_SIZE);
1151 json_start_array(vmdesc, "devices");
1152 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1154 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1155 continue;
1157 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1158 trace_savevm_section_skip(se->idstr, se->section_id);
1159 continue;
1162 trace_savevm_section_start(se->idstr, se->section_id);
1164 json_start_object(vmdesc, NULL);
1165 json_prop_str(vmdesc, "name", se->idstr);
1166 json_prop_int(vmdesc, "instance_id", se->instance_id);
1168 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1169 vmstate_save(f, se, vmdesc);
1170 trace_savevm_section_end(se->idstr, se->section_id, 0);
1171 save_section_footer(f, se);
1173 json_end_object(vmdesc);
1176 if (!in_postcopy) {
1177 /* Postcopy stream will still be going */
1178 qemu_put_byte(f, QEMU_VM_EOF);
1181 json_end_array(vmdesc);
1182 qjson_finish(vmdesc);
1183 vmdesc_len = strlen(qjson_get_str(vmdesc));
1185 if (should_send_vmdesc()) {
1186 qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1187 qemu_put_be32(f, vmdesc_len);
1188 qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len);
1190 qjson_destroy(vmdesc);
1192 qemu_fflush(f);
1195 /* Give an estimate of the amount left to be transferred,
1196 * the result is split into the amount for units that can and
1197 * for units that can't do postcopy.
1199 void qemu_savevm_state_pending(QEMUFile *f, uint64_t max_size,
1200 uint64_t *res_non_postcopiable,
1201 uint64_t *res_postcopiable)
1203 SaveStateEntry *se;
1205 *res_non_postcopiable = 0;
1206 *res_postcopiable = 0;
1209 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1210 if (!se->ops || !se->ops->save_live_pending) {
1211 continue;
1213 if (se->ops && se->ops->is_active) {
1214 if (!se->ops->is_active(se->opaque)) {
1215 continue;
1218 se->ops->save_live_pending(f, se->opaque, max_size,
1219 res_non_postcopiable, res_postcopiable);
1223 void qemu_savevm_state_cleanup(void)
1225 SaveStateEntry *se;
1227 trace_savevm_state_cleanup();
1228 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1229 if (se->ops && se->ops->cleanup) {
1230 se->ops->cleanup(se->opaque);
1235 static int qemu_savevm_state(QEMUFile *f, Error **errp)
1237 int ret;
1238 MigrationParams params = {
1239 .blk = 0,
1240 .shared = 0
1242 MigrationState *ms = migrate_init(&params);
1243 MigrationStatus status;
1244 ms->to_dst_file = f;
1246 if (migration_is_blocked(errp)) {
1247 ret = -EINVAL;
1248 goto done;
1251 qemu_mutex_unlock_iothread();
1252 qemu_savevm_state_header(f);
1253 qemu_savevm_state_begin(f, &params);
1254 qemu_mutex_lock_iothread();
1256 while (qemu_file_get_error(f) == 0) {
1257 if (qemu_savevm_state_iterate(f, false) > 0) {
1258 break;
1262 ret = qemu_file_get_error(f);
1263 if (ret == 0) {
1264 qemu_savevm_state_complete_precopy(f, false);
1265 ret = qemu_file_get_error(f);
1267 qemu_savevm_state_cleanup();
1268 if (ret != 0) {
1269 error_setg_errno(errp, -ret, "Error while writing VM state");
1272 done:
1273 if (ret != 0) {
1274 status = MIGRATION_STATUS_FAILED;
1275 } else {
1276 status = MIGRATION_STATUS_COMPLETED;
1278 migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status);
1279 return ret;
1282 static int qemu_save_device_state(QEMUFile *f)
1284 SaveStateEntry *se;
1286 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1287 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1289 cpu_synchronize_all_states();
1291 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1292 if (se->is_ram) {
1293 continue;
1295 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1296 continue;
1298 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1299 continue;
1302 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1304 vmstate_save(f, se, NULL);
1306 save_section_footer(f, se);
1309 qemu_put_byte(f, QEMU_VM_EOF);
1311 return qemu_file_get_error(f);
1314 static SaveStateEntry *find_se(const char *idstr, int instance_id)
1316 SaveStateEntry *se;
1318 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1319 if (!strcmp(se->idstr, idstr) &&
1320 (instance_id == se->instance_id ||
1321 instance_id == se->alias_id))
1322 return se;
1323 /* Migrating from an older version? */
1324 if (strstr(se->idstr, idstr) && se->compat) {
1325 if (!strcmp(se->compat->idstr, idstr) &&
1326 (instance_id == se->compat->instance_id ||
1327 instance_id == se->alias_id))
1328 return se;
1331 return NULL;
1334 enum LoadVMExitCodes {
1335 /* Allow a command to quit all layers of nested loadvm loops */
1336 LOADVM_QUIT = 1,
1339 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis);
1341 /* ------ incoming postcopy messages ------ */
1342 /* 'advise' arrives before any transfers just to tell us that a postcopy
1343 * *might* happen - it might be skipped if precopy transferred everything
1344 * quickly.
1346 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis)
1348 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1349 uint64_t remote_hps, remote_tps;
1351 trace_loadvm_postcopy_handle_advise();
1352 if (ps != POSTCOPY_INCOMING_NONE) {
1353 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1354 return -1;
1357 if (!postcopy_ram_supported_by_host()) {
1358 postcopy_state_set(POSTCOPY_INCOMING_NONE);
1359 return -1;
1362 remote_hps = qemu_get_be64(mis->from_src_file);
1363 if (remote_hps != getpagesize()) {
1365 * Some combinations of mismatch are probably possible but it gets
1366 * a bit more complicated. In particular we need to place whole
1367 * host pages on the dest at once, and we need to ensure that we
1368 * handle dirtying to make sure we never end up sending part of
1369 * a hostpage on it's own.
1371 error_report("Postcopy needs matching host page sizes (s=%d d=%d)",
1372 (int)remote_hps, getpagesize());
1373 return -1;
1376 remote_tps = qemu_get_be64(mis->from_src_file);
1377 if (remote_tps != (1ul << qemu_target_page_bits())) {
1379 * Again, some differences could be dealt with, but for now keep it
1380 * simple.
1382 error_report("Postcopy needs matching target page sizes (s=%d d=%d)",
1383 (int)remote_tps, 1 << qemu_target_page_bits());
1384 return -1;
1387 if (ram_postcopy_incoming_init(mis)) {
1388 return -1;
1391 postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1393 return 0;
1396 /* After postcopy we will be told to throw some pages away since they're
1397 * dirty and will have to be demand fetched. Must happen before CPU is
1398 * started.
1399 * There can be 0..many of these messages, each encoding multiple pages.
1401 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1402 uint16_t len)
1404 int tmp;
1405 char ramid[256];
1406 PostcopyState ps = postcopy_state_get();
1408 trace_loadvm_postcopy_ram_handle_discard();
1410 switch (ps) {
1411 case POSTCOPY_INCOMING_ADVISE:
1412 /* 1st discard */
1413 tmp = postcopy_ram_prepare_discard(mis);
1414 if (tmp) {
1415 return tmp;
1417 break;
1419 case POSTCOPY_INCOMING_DISCARD:
1420 /* Expected state */
1421 break;
1423 default:
1424 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1425 ps);
1426 return -1;
1428 /* We're expecting a
1429 * Version (0)
1430 * a RAM ID string (length byte, name, 0 term)
1431 * then at least 1 16 byte chunk
1433 if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1434 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1435 return -1;
1438 tmp = qemu_get_byte(mis->from_src_file);
1439 if (tmp != postcopy_ram_discard_version) {
1440 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1441 return -1;
1444 if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1445 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1446 return -1;
1448 tmp = qemu_get_byte(mis->from_src_file);
1449 if (tmp != 0) {
1450 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1451 return -1;
1454 len -= 3 + strlen(ramid);
1455 if (len % 16) {
1456 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1457 return -1;
1459 trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1460 while (len) {
1461 uint64_t start_addr, block_length;
1462 start_addr = qemu_get_be64(mis->from_src_file);
1463 block_length = qemu_get_be64(mis->from_src_file);
1465 len -= 16;
1466 int ret = ram_discard_range(mis, ramid, start_addr,
1467 block_length);
1468 if (ret) {
1469 return ret;
1472 trace_loadvm_postcopy_ram_handle_discard_end();
1474 return 0;
1478 * Triggered by a postcopy_listen command; this thread takes over reading
1479 * the input stream, leaving the main thread free to carry on loading the rest
1480 * of the device state (from RAM).
1481 * (TODO:This could do with being in a postcopy file - but there again it's
1482 * just another input loop, not that postcopy specific)
1484 static void *postcopy_ram_listen_thread(void *opaque)
1486 QEMUFile *f = opaque;
1487 MigrationIncomingState *mis = migration_incoming_get_current();
1488 int load_res;
1490 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
1491 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1492 qemu_sem_post(&mis->listen_thread_sem);
1493 trace_postcopy_ram_listen_thread_start();
1496 * Because we're a thread and not a coroutine we can't yield
1497 * in qemu_file, and thus we must be blocking now.
1499 qemu_file_set_blocking(f, true);
1500 load_res = qemu_loadvm_state_main(f, mis);
1501 /* And non-blocking again so we don't block in any cleanup */
1502 qemu_file_set_blocking(f, false);
1504 trace_postcopy_ram_listen_thread_exit();
1505 if (load_res < 0) {
1506 error_report("%s: loadvm failed: %d", __func__, load_res);
1507 qemu_file_set_error(f, load_res);
1508 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1509 MIGRATION_STATUS_FAILED);
1510 } else {
1512 * This looks good, but it's possible that the device loading in the
1513 * main thread hasn't finished yet, and so we might not be in 'RUN'
1514 * state yet; wait for the end of the main thread.
1516 qemu_event_wait(&mis->main_thread_load_event);
1518 postcopy_ram_incoming_cleanup(mis);
1520 if (load_res < 0) {
1522 * If something went wrong then we have a bad state so exit;
1523 * depending how far we got it might be possible at this point
1524 * to leave the guest running and fire MCEs for pages that never
1525 * arrived as a desperate recovery step.
1527 exit(EXIT_FAILURE);
1530 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1531 MIGRATION_STATUS_COMPLETED);
1533 * If everything has worked fine, then the main thread has waited
1534 * for us to start, and we're the last use of the mis.
1535 * (If something broke then qemu will have to exit anyway since it's
1536 * got a bad migration state).
1538 migration_incoming_state_destroy();
1541 return NULL;
1544 /* After this message we must be able to immediately receive postcopy data */
1545 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
1547 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
1548 trace_loadvm_postcopy_handle_listen();
1549 if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
1550 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
1551 return -1;
1553 if (ps == POSTCOPY_INCOMING_ADVISE) {
1555 * A rare case, we entered listen without having to do any discards,
1556 * so do the setup that's normally done at the time of the 1st discard.
1558 postcopy_ram_prepare_discard(mis);
1562 * Sensitise RAM - can now generate requests for blocks that don't exist
1563 * However, at this point the CPU shouldn't be running, and the IO
1564 * shouldn't be doing anything yet so don't actually expect requests
1566 if (postcopy_ram_enable_notify(mis)) {
1567 return -1;
1570 if (mis->have_listen_thread) {
1571 error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread");
1572 return -1;
1575 mis->have_listen_thread = true;
1576 /* Start up the listening thread and wait for it to signal ready */
1577 qemu_sem_init(&mis->listen_thread_sem, 0);
1578 qemu_thread_create(&mis->listen_thread, "postcopy/listen",
1579 postcopy_ram_listen_thread, mis->from_src_file,
1580 QEMU_THREAD_DETACHED);
1581 qemu_sem_wait(&mis->listen_thread_sem);
1582 qemu_sem_destroy(&mis->listen_thread_sem);
1584 return 0;
1588 typedef struct {
1589 QEMUBH *bh;
1590 } HandleRunBhData;
1592 static void loadvm_postcopy_handle_run_bh(void *opaque)
1594 Error *local_err = NULL;
1595 HandleRunBhData *data = opaque;
1597 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1598 * in migration.c
1600 cpu_synchronize_all_post_init();
1602 qemu_announce_self();
1604 /* Make sure all file formats flush their mutable metadata */
1605 bdrv_invalidate_cache_all(&local_err);
1606 if (local_err) {
1607 error_report_err(local_err);
1610 trace_loadvm_postcopy_handle_run_cpu_sync();
1611 cpu_synchronize_all_post_init();
1613 trace_loadvm_postcopy_handle_run_vmstart();
1615 if (autostart) {
1616 /* Hold onto your hats, starting the CPU */
1617 vm_start();
1618 } else {
1619 /* leave it paused and let management decide when to start the CPU */
1620 runstate_set(RUN_STATE_PAUSED);
1623 qemu_bh_delete(data->bh);
1624 g_free(data);
1627 /* After all discards we can start running and asking for pages */
1628 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
1630 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
1631 HandleRunBhData *data;
1633 trace_loadvm_postcopy_handle_run();
1634 if (ps != POSTCOPY_INCOMING_LISTENING) {
1635 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
1636 return -1;
1639 data = g_new(HandleRunBhData, 1);
1640 data->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, data);
1641 qemu_bh_schedule(data->bh);
1643 /* We need to finish reading the stream from the package
1644 * and also stop reading anything more from the stream that loaded the
1645 * package (since it's now being read by the listener thread).
1646 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
1648 return LOADVM_QUIT;
1652 * Immediately following this command is a blob of data containing an embedded
1653 * chunk of migration stream; read it and load it.
1655 * @mis: Incoming state
1656 * @length: Length of packaged data to read
1658 * Returns: Negative values on error
1661 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
1663 int ret;
1664 size_t length;
1665 QIOChannelBuffer *bioc;
1667 length = qemu_get_be32(mis->from_src_file);
1668 trace_loadvm_handle_cmd_packaged(length);
1670 if (length > MAX_VM_CMD_PACKAGED_SIZE) {
1671 error_report("Unreasonably large packaged state: %zu", length);
1672 return -1;
1675 bioc = qio_channel_buffer_new(length);
1676 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer");
1677 ret = qemu_get_buffer(mis->from_src_file,
1678 bioc->data,
1679 length);
1680 if (ret != length) {
1681 object_unref(OBJECT(bioc));
1682 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
1683 ret, length);
1684 return (ret < 0) ? ret : -EAGAIN;
1686 bioc->usage += length;
1687 trace_loadvm_handle_cmd_packaged_received(ret);
1689 QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
1691 ret = qemu_loadvm_state_main(packf, mis);
1692 trace_loadvm_handle_cmd_packaged_main(ret);
1693 qemu_fclose(packf);
1694 object_unref(OBJECT(bioc));
1696 return ret;
1700 * Process an incoming 'QEMU_VM_COMMAND'
1701 * 0 just a normal return
1702 * LOADVM_QUIT All good, but exit the loop
1703 * <0 Error
1705 static int loadvm_process_command(QEMUFile *f)
1707 MigrationIncomingState *mis = migration_incoming_get_current();
1708 uint16_t cmd;
1709 uint16_t len;
1710 uint32_t tmp32;
1712 cmd = qemu_get_be16(f);
1713 len = qemu_get_be16(f);
1715 trace_loadvm_process_command(cmd, len);
1716 if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
1717 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
1718 return -EINVAL;
1721 if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
1722 error_report("%s received with bad length - expecting %zu, got %d",
1723 mig_cmd_args[cmd].name,
1724 (size_t)mig_cmd_args[cmd].len, len);
1725 return -ERANGE;
1728 switch (cmd) {
1729 case MIG_CMD_OPEN_RETURN_PATH:
1730 if (mis->to_src_file) {
1731 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
1732 /* Not really a problem, so don't give up */
1733 return 0;
1735 mis->to_src_file = qemu_file_get_return_path(f);
1736 if (!mis->to_src_file) {
1737 error_report("CMD_OPEN_RETURN_PATH failed");
1738 return -1;
1740 break;
1742 case MIG_CMD_PING:
1743 tmp32 = qemu_get_be32(f);
1744 trace_loadvm_process_command_ping(tmp32);
1745 if (!mis->to_src_file) {
1746 error_report("CMD_PING (0x%x) received with no return path",
1747 tmp32);
1748 return -1;
1750 migrate_send_rp_pong(mis, tmp32);
1751 break;
1753 case MIG_CMD_PACKAGED:
1754 return loadvm_handle_cmd_packaged(mis);
1756 case MIG_CMD_POSTCOPY_ADVISE:
1757 return loadvm_postcopy_handle_advise(mis);
1759 case MIG_CMD_POSTCOPY_LISTEN:
1760 return loadvm_postcopy_handle_listen(mis);
1762 case MIG_CMD_POSTCOPY_RUN:
1763 return loadvm_postcopy_handle_run(mis);
1765 case MIG_CMD_POSTCOPY_RAM_DISCARD:
1766 return loadvm_postcopy_ram_handle_discard(mis, len);
1769 return 0;
1772 struct LoadStateEntry {
1773 QLIST_ENTRY(LoadStateEntry) entry;
1774 SaveStateEntry *se;
1775 int section_id;
1776 int version_id;
1780 * Read a footer off the wire and check that it matches the expected section
1782 * Returns: true if the footer was good
1783 * false if there is a problem (and calls error_report to say why)
1785 static bool check_section_footer(QEMUFile *f, LoadStateEntry *le)
1787 uint8_t read_mark;
1788 uint32_t read_section_id;
1790 if (skip_section_footers) {
1791 /* No footer to check */
1792 return true;
1795 read_mark = qemu_get_byte(f);
1797 if (read_mark != QEMU_VM_SECTION_FOOTER) {
1798 error_report("Missing section footer for %s", le->se->idstr);
1799 return false;
1802 read_section_id = qemu_get_be32(f);
1803 if (read_section_id != le->section_id) {
1804 error_report("Mismatched section id in footer for %s -"
1805 " read 0x%x expected 0x%x",
1806 le->se->idstr, read_section_id, le->section_id);
1807 return false;
1810 /* All good */
1811 return true;
1814 void loadvm_free_handlers(MigrationIncomingState *mis)
1816 LoadStateEntry *le, *new_le;
1818 QLIST_FOREACH_SAFE(le, &mis->loadvm_handlers, entry, new_le) {
1819 QLIST_REMOVE(le, entry);
1820 g_free(le);
1824 static int
1825 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis)
1827 uint32_t instance_id, version_id, section_id;
1828 SaveStateEntry *se;
1829 LoadStateEntry *le;
1830 char idstr[256];
1831 int ret;
1833 /* Read section start */
1834 section_id = qemu_get_be32(f);
1835 if (!qemu_get_counted_string(f, idstr)) {
1836 error_report("Unable to read ID string for section %u",
1837 section_id);
1838 return -EINVAL;
1840 instance_id = qemu_get_be32(f);
1841 version_id = qemu_get_be32(f);
1843 trace_qemu_loadvm_state_section_startfull(section_id, idstr,
1844 instance_id, version_id);
1845 /* Find savevm section */
1846 se = find_se(idstr, instance_id);
1847 if (se == NULL) {
1848 error_report("Unknown savevm section or instance '%s' %d",
1849 idstr, instance_id);
1850 return -EINVAL;
1853 /* Validate version */
1854 if (version_id > se->version_id) {
1855 error_report("savevm: unsupported version %d for '%s' v%d",
1856 version_id, idstr, se->version_id);
1857 return -EINVAL;
1860 /* Validate if it is a device's state */
1861 if (xen_enabled() && se->is_ram) {
1862 error_report("loadvm: %s RAM loading not allowed on Xen", idstr);
1863 return -EINVAL;
1866 /* Add entry */
1867 le = g_malloc0(sizeof(*le));
1869 le->se = se;
1870 le->section_id = section_id;
1871 le->version_id = version_id;
1872 QLIST_INSERT_HEAD(&mis->loadvm_handlers, le, entry);
1874 ret = vmstate_load(f, le->se, le->version_id);
1875 if (ret < 0) {
1876 error_report("error while loading state for instance 0x%x of"
1877 " device '%s'", instance_id, idstr);
1878 return ret;
1880 if (!check_section_footer(f, le)) {
1881 return -EINVAL;
1884 return 0;
1887 static int
1888 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis)
1890 uint32_t section_id;
1891 LoadStateEntry *le;
1892 int ret;
1894 section_id = qemu_get_be32(f);
1896 trace_qemu_loadvm_state_section_partend(section_id);
1897 QLIST_FOREACH(le, &mis->loadvm_handlers, entry) {
1898 if (le->section_id == section_id) {
1899 break;
1902 if (le == NULL) {
1903 error_report("Unknown savevm section %d", section_id);
1904 return -EINVAL;
1907 ret = vmstate_load(f, le->se, le->version_id);
1908 if (ret < 0) {
1909 error_report("error while loading state section id %d(%s)",
1910 section_id, le->se->idstr);
1911 return ret;
1913 if (!check_section_footer(f, le)) {
1914 return -EINVAL;
1917 return 0;
1920 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
1922 uint8_t section_type;
1923 int ret = 0;
1925 while ((section_type = qemu_get_byte(f)) != QEMU_VM_EOF) {
1926 ret = 0;
1927 trace_qemu_loadvm_state_section(section_type);
1928 switch (section_type) {
1929 case QEMU_VM_SECTION_START:
1930 case QEMU_VM_SECTION_FULL:
1931 ret = qemu_loadvm_section_start_full(f, mis);
1932 if (ret < 0) {
1933 goto out;
1935 break;
1936 case QEMU_VM_SECTION_PART:
1937 case QEMU_VM_SECTION_END:
1938 ret = qemu_loadvm_section_part_end(f, mis);
1939 if (ret < 0) {
1940 goto out;
1942 break;
1943 case QEMU_VM_COMMAND:
1944 ret = loadvm_process_command(f);
1945 trace_qemu_loadvm_state_section_command(ret);
1946 if ((ret < 0) || (ret & LOADVM_QUIT)) {
1947 goto out;
1949 break;
1950 default:
1951 error_report("Unknown savevm section type %d", section_type);
1952 ret = -EINVAL;
1953 goto out;
1957 out:
1958 if (ret < 0) {
1959 qemu_file_set_error(f, ret);
1961 return ret;
1964 int qemu_loadvm_state(QEMUFile *f)
1966 MigrationIncomingState *mis = migration_incoming_get_current();
1967 Error *local_err = NULL;
1968 unsigned int v;
1969 int ret;
1971 if (qemu_savevm_state_blocked(&local_err)) {
1972 error_report_err(local_err);
1973 return -EINVAL;
1976 v = qemu_get_be32(f);
1977 if (v != QEMU_VM_FILE_MAGIC) {
1978 error_report("Not a migration stream");
1979 return -EINVAL;
1982 v = qemu_get_be32(f);
1983 if (v == QEMU_VM_FILE_VERSION_COMPAT) {
1984 error_report("SaveVM v2 format is obsolete and don't work anymore");
1985 return -ENOTSUP;
1987 if (v != QEMU_VM_FILE_VERSION) {
1988 error_report("Unsupported migration stream version");
1989 return -ENOTSUP;
1992 if (!savevm_state.skip_configuration || enforce_config_section()) {
1993 if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
1994 error_report("Configuration section missing");
1995 return -EINVAL;
1997 ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
1999 if (ret) {
2000 return ret;
2004 ret = qemu_loadvm_state_main(f, mis);
2005 qemu_event_set(&mis->main_thread_load_event);
2007 trace_qemu_loadvm_state_post_main(ret);
2009 if (mis->have_listen_thread) {
2010 /* Listen thread still going, can't clean up yet */
2011 return ret;
2014 if (ret == 0) {
2015 ret = qemu_file_get_error(f);
2019 * Try to read in the VMDESC section as well, so that dumping tools that
2020 * intercept our migration stream have the chance to see it.
2023 /* We've got to be careful; if we don't read the data and just shut the fd
2024 * then the sender can error if we close while it's still sending.
2025 * We also mustn't read data that isn't there; some transports (RDMA)
2026 * will stall waiting for that data when the source has already closed.
2028 if (ret == 0 && should_send_vmdesc()) {
2029 uint8_t *buf;
2030 uint32_t size;
2031 uint8_t section_type = qemu_get_byte(f);
2033 if (section_type != QEMU_VM_VMDESCRIPTION) {
2034 error_report("Expected vmdescription section, but got %d",
2035 section_type);
2037 * It doesn't seem worth failing at this point since
2038 * we apparently have an otherwise valid VM state
2040 } else {
2041 buf = g_malloc(0x1000);
2042 size = qemu_get_be32(f);
2044 while (size > 0) {
2045 uint32_t read_chunk = MIN(size, 0x1000);
2046 qemu_get_buffer(f, buf, read_chunk);
2047 size -= read_chunk;
2049 g_free(buf);
2053 cpu_synchronize_all_post_init();
2055 return ret;
2058 int save_vmstate(Monitor *mon, const char *name)
2060 BlockDriverState *bs, *bs1;
2061 QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1;
2062 int ret = -1;
2063 QEMUFile *f;
2064 int saved_vm_running;
2065 uint64_t vm_state_size;
2066 qemu_timeval tv;
2067 struct tm tm;
2068 Error *local_err = NULL;
2069 AioContext *aio_context;
2071 if (!bdrv_all_can_snapshot(&bs)) {
2072 monitor_printf(mon, "Device '%s' is writable but does not "
2073 "support snapshots.\n", bdrv_get_device_name(bs));
2074 return ret;
2077 /* Delete old snapshots of the same name */
2078 if (name) {
2079 ret = bdrv_all_delete_snapshot(name, &bs1, &local_err);
2080 if (ret < 0) {
2081 error_reportf_err(local_err,
2082 "Error while deleting snapshot on device '%s': ",
2083 bdrv_get_device_name(bs1));
2084 return ret;
2088 bs = bdrv_all_find_vmstate_bs();
2089 if (bs == NULL) {
2090 monitor_printf(mon, "No block device can accept snapshots\n");
2091 return ret;
2093 aio_context = bdrv_get_aio_context(bs);
2095 saved_vm_running = runstate_is_running();
2097 ret = global_state_store();
2098 if (ret) {
2099 monitor_printf(mon, "Error saving global state\n");
2100 return ret;
2102 vm_stop(RUN_STATE_SAVE_VM);
2104 aio_context_acquire(aio_context);
2106 memset(sn, 0, sizeof(*sn));
2108 /* fill auxiliary fields */
2109 qemu_gettimeofday(&tv);
2110 sn->date_sec = tv.tv_sec;
2111 sn->date_nsec = tv.tv_usec * 1000;
2112 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
2114 if (name) {
2115 ret = bdrv_snapshot_find(bs, old_sn, name);
2116 if (ret >= 0) {
2117 pstrcpy(sn->name, sizeof(sn->name), old_sn->name);
2118 pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str);
2119 } else {
2120 pstrcpy(sn->name, sizeof(sn->name), name);
2122 } else {
2123 /* cast below needed for OpenBSD where tv_sec is still 'long' */
2124 localtime_r((const time_t *)&tv.tv_sec, &tm);
2125 strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm);
2128 /* save the VM state */
2129 f = qemu_fopen_bdrv(bs, 1);
2130 if (!f) {
2131 monitor_printf(mon, "Could not open VM state file\n");
2132 goto the_end;
2134 ret = qemu_savevm_state(f, &local_err);
2135 vm_state_size = qemu_ftell(f);
2136 qemu_fclose(f);
2137 if (ret < 0) {
2138 error_report_err(local_err);
2139 goto the_end;
2142 ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, &bs);
2143 if (ret < 0) {
2144 monitor_printf(mon, "Error while creating snapshot on '%s'\n",
2145 bdrv_get_device_name(bs));
2146 goto the_end;
2149 ret = 0;
2151 the_end:
2152 aio_context_release(aio_context);
2153 if (saved_vm_running) {
2154 vm_start();
2156 return ret;
2159 void hmp_savevm(Monitor *mon, const QDict *qdict)
2161 save_vmstate(mon, qdict_get_try_str(qdict, "name"));
2164 void qmp_xen_save_devices_state(const char *filename, Error **errp)
2166 QEMUFile *f;
2167 QIOChannelFile *ioc;
2168 int saved_vm_running;
2169 int ret;
2171 saved_vm_running = runstate_is_running();
2172 vm_stop(RUN_STATE_SAVE_VM);
2173 global_state_store_running();
2175 ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT, 0660, errp);
2176 if (!ioc) {
2177 goto the_end;
2179 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state");
2180 f = qemu_fopen_channel_output(QIO_CHANNEL(ioc));
2181 ret = qemu_save_device_state(f);
2182 qemu_fclose(f);
2183 if (ret < 0) {
2184 error_setg(errp, QERR_IO_ERROR);
2187 the_end:
2188 if (saved_vm_running) {
2189 vm_start();
2193 void qmp_xen_load_devices_state(const char *filename, Error **errp)
2195 QEMUFile *f;
2196 QIOChannelFile *ioc;
2197 int ret;
2199 /* Guest must be paused before loading the device state; the RAM state
2200 * will already have been loaded by xc
2202 if (runstate_is_running()) {
2203 error_setg(errp, "Cannot update device state while vm is running");
2204 return;
2206 vm_stop(RUN_STATE_RESTORE_VM);
2208 ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp);
2209 if (!ioc) {
2210 return;
2212 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state");
2213 f = qemu_fopen_channel_input(QIO_CHANNEL(ioc));
2215 ret = qemu_loadvm_state(f);
2216 qemu_fclose(f);
2217 if (ret < 0) {
2218 error_setg(errp, QERR_IO_ERROR);
2220 migration_incoming_state_destroy();
2223 int load_vmstate(const char *name)
2225 BlockDriverState *bs, *bs_vm_state;
2226 QEMUSnapshotInfo sn;
2227 QEMUFile *f;
2228 int ret;
2229 AioContext *aio_context;
2230 MigrationIncomingState *mis = migration_incoming_get_current();
2232 if (!bdrv_all_can_snapshot(&bs)) {
2233 error_report("Device '%s' is writable but does not support snapshots.",
2234 bdrv_get_device_name(bs));
2235 return -ENOTSUP;
2237 ret = bdrv_all_find_snapshot(name, &bs);
2238 if (ret < 0) {
2239 error_report("Device '%s' does not have the requested snapshot '%s'",
2240 bdrv_get_device_name(bs), name);
2241 return ret;
2244 bs_vm_state = bdrv_all_find_vmstate_bs();
2245 if (!bs_vm_state) {
2246 error_report("No block device supports snapshots");
2247 return -ENOTSUP;
2249 aio_context = bdrv_get_aio_context(bs_vm_state);
2251 /* Don't even try to load empty VM states */
2252 aio_context_acquire(aio_context);
2253 ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
2254 aio_context_release(aio_context);
2255 if (ret < 0) {
2256 return ret;
2257 } else if (sn.vm_state_size == 0) {
2258 error_report("This is a disk-only snapshot. Revert to it offline "
2259 "using qemu-img.");
2260 return -EINVAL;
2263 /* Flush all IO requests so they don't interfere with the new state. */
2264 bdrv_drain_all();
2266 ret = bdrv_all_goto_snapshot(name, &bs);
2267 if (ret < 0) {
2268 error_report("Error %d while activating snapshot '%s' on '%s'",
2269 ret, name, bdrv_get_device_name(bs));
2270 return ret;
2273 /* restore the VM state */
2274 f = qemu_fopen_bdrv(bs_vm_state, 0);
2275 if (!f) {
2276 error_report("Could not open VM state file");
2277 return -EINVAL;
2280 qemu_system_reset(VMRESET_SILENT);
2281 mis->from_src_file = f;
2283 aio_context_acquire(aio_context);
2284 ret = qemu_loadvm_state(f);
2285 qemu_fclose(f);
2286 aio_context_release(aio_context);
2288 migration_incoming_state_destroy();
2289 if (ret < 0) {
2290 error_report("Error %d while loading VM state", ret);
2291 return ret;
2294 return 0;
2297 void hmp_delvm(Monitor *mon, const QDict *qdict)
2299 BlockDriverState *bs;
2300 Error *err;
2301 const char *name = qdict_get_str(qdict, "name");
2303 if (bdrv_all_delete_snapshot(name, &bs, &err) < 0) {
2304 error_reportf_err(err,
2305 "Error while deleting snapshot on device '%s': ",
2306 bdrv_get_device_name(bs));
2310 void hmp_info_snapshots(Monitor *mon, const QDict *qdict)
2312 BlockDriverState *bs, *bs1;
2313 BdrvNextIterator it1;
2314 QEMUSnapshotInfo *sn_tab, *sn;
2315 bool no_snapshot = true;
2316 int nb_sns, i;
2317 int total;
2318 int *global_snapshots;
2319 AioContext *aio_context;
2321 typedef struct SnapshotEntry {
2322 QEMUSnapshotInfo sn;
2323 QTAILQ_ENTRY(SnapshotEntry) next;
2324 } SnapshotEntry;
2326 typedef struct ImageEntry {
2327 const char *imagename;
2328 QTAILQ_ENTRY(ImageEntry) next;
2329 QTAILQ_HEAD(, SnapshotEntry) snapshots;
2330 } ImageEntry;
2332 QTAILQ_HEAD(, ImageEntry) image_list =
2333 QTAILQ_HEAD_INITIALIZER(image_list);
2335 ImageEntry *image_entry, *next_ie;
2336 SnapshotEntry *snapshot_entry;
2338 bs = bdrv_all_find_vmstate_bs();
2339 if (!bs) {
2340 monitor_printf(mon, "No available block device supports snapshots\n");
2341 return;
2343 aio_context = bdrv_get_aio_context(bs);
2345 aio_context_acquire(aio_context);
2346 nb_sns = bdrv_snapshot_list(bs, &sn_tab);
2347 aio_context_release(aio_context);
2349 if (nb_sns < 0) {
2350 monitor_printf(mon, "bdrv_snapshot_list: error %d\n", nb_sns);
2351 return;
2354 for (bs1 = bdrv_first(&it1); bs1; bs1 = bdrv_next(&it1)) {
2355 int bs1_nb_sns = 0;
2356 ImageEntry *ie;
2357 SnapshotEntry *se;
2358 AioContext *ctx = bdrv_get_aio_context(bs1);
2360 aio_context_acquire(ctx);
2361 if (bdrv_can_snapshot(bs1)) {
2362 sn = NULL;
2363 bs1_nb_sns = bdrv_snapshot_list(bs1, &sn);
2364 if (bs1_nb_sns > 0) {
2365 no_snapshot = false;
2366 ie = g_new0(ImageEntry, 1);
2367 ie->imagename = bdrv_get_device_name(bs1);
2368 QTAILQ_INIT(&ie->snapshots);
2369 QTAILQ_INSERT_TAIL(&image_list, ie, next);
2370 for (i = 0; i < bs1_nb_sns; i++) {
2371 se = g_new0(SnapshotEntry, 1);
2372 se->sn = sn[i];
2373 QTAILQ_INSERT_TAIL(&ie->snapshots, se, next);
2376 g_free(sn);
2378 aio_context_release(ctx);
2381 if (no_snapshot) {
2382 monitor_printf(mon, "There is no snapshot available.\n");
2383 return;
2386 global_snapshots = g_new0(int, nb_sns);
2387 total = 0;
2388 for (i = 0; i < nb_sns; i++) {
2389 SnapshotEntry *next_sn;
2390 if (bdrv_all_find_snapshot(sn_tab[i].name, &bs1) == 0) {
2391 global_snapshots[total] = i;
2392 total++;
2393 QTAILQ_FOREACH(image_entry, &image_list, next) {
2394 QTAILQ_FOREACH_SAFE(snapshot_entry, &image_entry->snapshots,
2395 next, next_sn) {
2396 if (!strcmp(sn_tab[i].name, snapshot_entry->sn.name)) {
2397 QTAILQ_REMOVE(&image_entry->snapshots, snapshot_entry,
2398 next);
2399 g_free(snapshot_entry);
2406 monitor_printf(mon, "List of snapshots present on all disks:\n");
2408 if (total > 0) {
2409 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, NULL);
2410 monitor_printf(mon, "\n");
2411 for (i = 0; i < total; i++) {
2412 sn = &sn_tab[global_snapshots[i]];
2413 /* The ID is not guaranteed to be the same on all images, so
2414 * overwrite it.
2416 pstrcpy(sn->id_str, sizeof(sn->id_str), "--");
2417 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, sn);
2418 monitor_printf(mon, "\n");
2420 } else {
2421 monitor_printf(mon, "None\n");
2424 QTAILQ_FOREACH(image_entry, &image_list, next) {
2425 if (QTAILQ_EMPTY(&image_entry->snapshots)) {
2426 continue;
2428 monitor_printf(mon,
2429 "\nList of partial (non-loadable) snapshots on '%s':\n",
2430 image_entry->imagename);
2431 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, NULL);
2432 monitor_printf(mon, "\n");
2433 QTAILQ_FOREACH(snapshot_entry, &image_entry->snapshots, next) {
2434 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon,
2435 &snapshot_entry->sn);
2436 monitor_printf(mon, "\n");
2440 QTAILQ_FOREACH_SAFE(image_entry, &image_list, next, next_ie) {
2441 SnapshotEntry *next_sn;
2442 QTAILQ_FOREACH_SAFE(snapshot_entry, &image_entry->snapshots, next,
2443 next_sn) {
2444 g_free(snapshot_entry);
2446 g_free(image_entry);
2448 g_free(sn_tab);
2449 g_free(global_snapshots);
2453 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
2455 qemu_ram_set_idstr(mr->ram_block,
2456 memory_region_name(mr), dev);
2459 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
2461 qemu_ram_unset_idstr(mr->ram_block);
2464 void vmstate_register_ram_global(MemoryRegion *mr)
2466 vmstate_register_ram(mr, NULL);