4 * Copyright (c) 2003-2008 Fabrice Bellard
5 * Copyright (c) 2009-2015 Red Hat Inc
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
29 #include "qemu/osdep.h"
30 #include "hw/boards.h"
31 #include "hw/xen/xen.h"
33 #include "migration.h"
34 #include "migration/snapshot.h"
35 #include "migration/misc.h"
36 #include "migration/register.h"
37 #include "migration/global_state.h"
39 #include "qemu-file-channel.h"
40 #include "qemu-file.h"
42 #include "postcopy-ram.h"
43 #include "qapi/qmp/qerror.h"
44 #include "qemu/error-report.h"
45 #include "sysemu/cpus.h"
46 #include "exec/memory.h"
47 #include "exec/target_page.h"
48 #include "qmp-commands.h"
51 #include "block/snapshot.h"
52 #include "qemu/cutils.h"
53 #include "io/channel-buffer.h"
54 #include "io/channel-file.h"
57 #define ETH_P_RARP 0x8035
59 #define ARP_HTYPE_ETH 0x0001
60 #define ARP_PTYPE_IP 0x0800
61 #define ARP_OP_REQUEST_REV 0x3
63 const unsigned int postcopy_ram_discard_version = 0;
65 /* Subcommands for QEMU_VM_COMMAND */
67 MIG_CMD_INVALID = 0, /* Must be 0 */
68 MIG_CMD_OPEN_RETURN_PATH, /* Tell the dest to open the Return path */
69 MIG_CMD_PING, /* Request a PONG on the RP */
71 MIG_CMD_POSTCOPY_ADVISE, /* Prior to any page transfers, just
72 warn we might want to do PC */
73 MIG_CMD_POSTCOPY_LISTEN, /* Start listening for incoming
74 pages as it's running. */
75 MIG_CMD_POSTCOPY_RUN, /* Start execution */
77 MIG_CMD_POSTCOPY_RAM_DISCARD, /* A list of pages to discard that
78 were previously sent during
79 precopy but are dirty. */
80 MIG_CMD_PACKAGED, /* Send a wrapped stream within this stream */
84 #define MAX_VM_CMD_PACKAGED_SIZE (1ul << 24)
85 static struct mig_cmd_args {
86 ssize_t len; /* -1 = variable */
89 [MIG_CMD_INVALID] = { .len = -1, .name = "INVALID" },
90 [MIG_CMD_OPEN_RETURN_PATH] = { .len = 0, .name = "OPEN_RETURN_PATH" },
91 [MIG_CMD_PING] = { .len = sizeof(uint32_t), .name = "PING" },
92 [MIG_CMD_POSTCOPY_ADVISE] = { .len = 16, .name = "POSTCOPY_ADVISE" },
93 [MIG_CMD_POSTCOPY_LISTEN] = { .len = 0, .name = "POSTCOPY_LISTEN" },
94 [MIG_CMD_POSTCOPY_RUN] = { .len = 0, .name = "POSTCOPY_RUN" },
95 [MIG_CMD_POSTCOPY_RAM_DISCARD] = {
96 .len = -1, .name = "POSTCOPY_RAM_DISCARD" },
97 [MIG_CMD_PACKAGED] = { .len = 4, .name = "PACKAGED" },
98 [MIG_CMD_MAX] = { .len = -1, .name = "MAX" },
101 static int announce_self_create(uint8_t *buf,
104 /* Ethernet header. */
105 memset(buf, 0xff, 6); /* destination MAC addr */
106 memcpy(buf + 6, mac_addr, 6); /* source MAC addr */
107 *(uint16_t *)(buf + 12) = htons(ETH_P_RARP); /* ethertype */
110 *(uint16_t *)(buf + 14) = htons(ARP_HTYPE_ETH); /* hardware addr space */
111 *(uint16_t *)(buf + 16) = htons(ARP_PTYPE_IP); /* protocol addr space */
112 *(buf + 18) = 6; /* hardware addr length (ethernet) */
113 *(buf + 19) = 4; /* protocol addr length (IPv4) */
114 *(uint16_t *)(buf + 20) = htons(ARP_OP_REQUEST_REV); /* opcode */
115 memcpy(buf + 22, mac_addr, 6); /* source hw addr */
116 memset(buf + 28, 0x00, 4); /* source protocol addr */
117 memcpy(buf + 32, mac_addr, 6); /* target hw addr */
118 memset(buf + 38, 0x00, 4); /* target protocol addr */
120 /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */
121 memset(buf + 42, 0x00, 18);
123 return 60; /* len (FCS will be added by hardware) */
126 static void qemu_announce_self_iter(NICState *nic, void *opaque)
131 trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic->conf->macaddr));
132 len = announce_self_create(buf, nic->conf->macaddr.a);
134 qemu_send_packet_raw(qemu_get_queue(nic), buf, len);
138 static void qemu_announce_self_once(void *opaque)
140 static int count = SELF_ANNOUNCE_ROUNDS;
141 QEMUTimer *timer = *(QEMUTimer **)opaque;
143 qemu_foreach_nic(qemu_announce_self_iter, NULL);
146 /* delay 50ms, 150ms, 250ms, ... */
147 timer_mod(timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) +
148 self_announce_delay(count));
155 void qemu_announce_self(void)
157 static QEMUTimer *timer;
158 timer = timer_new_ms(QEMU_CLOCK_REALTIME, qemu_announce_self_once, &timer);
159 qemu_announce_self_once(&timer);
162 /***********************************************************/
163 /* savevm/loadvm support */
165 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt,
171 qemu_iovec_init_external(&qiov, iov, iovcnt);
172 ret = bdrv_writev_vmstate(opaque, &qiov, pos);
180 static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos,
183 return bdrv_load_vmstate(opaque, buf, pos, size);
186 static int bdrv_fclose(void *opaque)
188 return bdrv_flush(opaque);
191 static const QEMUFileOps bdrv_read_ops = {
192 .get_buffer = block_get_buffer,
196 static const QEMUFileOps bdrv_write_ops = {
197 .writev_buffer = block_writev_buffer,
201 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
204 return qemu_fopen_ops(bs, &bdrv_write_ops);
206 return qemu_fopen_ops(bs, &bdrv_read_ops);
210 /* QEMUFile timer support.
211 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
214 void timer_put(QEMUFile *f, QEMUTimer *ts)
216 uint64_t expire_time;
218 expire_time = timer_expire_time_ns(ts);
219 qemu_put_be64(f, expire_time);
222 void timer_get(QEMUFile *f, QEMUTimer *ts)
224 uint64_t expire_time;
226 expire_time = qemu_get_be64(f);
227 if (expire_time != -1) {
228 timer_mod_ns(ts, expire_time);
235 /* VMState timer support.
236 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
239 static int get_timer(QEMUFile *f, void *pv, size_t size, VMStateField *field)
246 static int put_timer(QEMUFile *f, void *pv, size_t size, VMStateField *field,
255 const VMStateInfo vmstate_info_timer = {
262 typedef struct CompatEntry {
267 typedef struct SaveStateEntry {
268 QTAILQ_ENTRY(SaveStateEntry) entry;
273 /* version id read from the stream */
276 /* section id read from the stream */
279 const VMStateDescription *vmsd;
285 typedef struct SaveState {
286 QTAILQ_HEAD(, SaveStateEntry) handlers;
287 int global_section_id;
290 uint32_t target_page_bits;
293 static SaveState savevm_state = {
294 .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
295 .global_section_id = 0,
298 static void configuration_pre_save(void *opaque)
300 SaveState *state = opaque;
301 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
303 state->len = strlen(current_name);
304 state->name = current_name;
305 state->target_page_bits = qemu_target_page_bits();
308 static int configuration_pre_load(void *opaque)
310 SaveState *state = opaque;
312 /* If there is no target-page-bits subsection it means the source
313 * predates the variable-target-page-bits support and is using the
314 * minimum possible value for this CPU.
316 state->target_page_bits = qemu_target_page_bits_min();
320 static int configuration_post_load(void *opaque, int version_id)
322 SaveState *state = opaque;
323 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
325 if (strncmp(state->name, current_name, state->len) != 0) {
326 error_report("Machine type received is '%.*s' and local is '%s'",
327 (int) state->len, state->name, current_name);
331 if (state->target_page_bits != qemu_target_page_bits()) {
332 error_report("Received TARGET_PAGE_BITS is %d but local is %d",
333 state->target_page_bits, qemu_target_page_bits());
340 /* The target-page-bits subsection is present only if the
341 * target page size is not the same as the default (ie the
342 * minimum page size for a variable-page-size guest CPU).
343 * If it is present then it contains the actual target page
344 * bits for the machine, and migration will fail if the
345 * two ends don't agree about it.
347 static bool vmstate_target_page_bits_needed(void *opaque)
349 return qemu_target_page_bits()
350 > qemu_target_page_bits_min();
353 static const VMStateDescription vmstate_target_page_bits = {
354 .name = "configuration/target-page-bits",
356 .minimum_version_id = 1,
357 .needed = vmstate_target_page_bits_needed,
358 .fields = (VMStateField[]) {
359 VMSTATE_UINT32(target_page_bits, SaveState),
360 VMSTATE_END_OF_LIST()
364 static const VMStateDescription vmstate_configuration = {
365 .name = "configuration",
367 .pre_load = configuration_pre_load,
368 .post_load = configuration_post_load,
369 .pre_save = configuration_pre_save,
370 .fields = (VMStateField[]) {
371 VMSTATE_UINT32(len, SaveState),
372 VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len),
373 VMSTATE_END_OF_LIST()
375 .subsections = (const VMStateDescription*[]) {
376 &vmstate_target_page_bits,
381 static void dump_vmstate_vmsd(FILE *out_file,
382 const VMStateDescription *vmsd, int indent,
385 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
388 fprintf(out_file, "%*s{\n", indent, "");
390 fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
391 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
393 fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
394 field->field_exists ? "true" : "false");
395 fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
396 if (field->vmsd != NULL) {
397 fprintf(out_file, ",\n");
398 dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
400 fprintf(out_file, "\n%*s}", indent - 2, "");
403 static void dump_vmstate_vmss(FILE *out_file,
404 const VMStateDescription **subsection,
407 if (*subsection != NULL) {
408 dump_vmstate_vmsd(out_file, *subsection, indent, true);
412 static void dump_vmstate_vmsd(FILE *out_file,
413 const VMStateDescription *vmsd, int indent,
417 fprintf(out_file, "%*s{\n", indent, "");
419 fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
422 fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
423 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
425 fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
426 vmsd->minimum_version_id);
427 if (vmsd->fields != NULL) {
428 const VMStateField *field = vmsd->fields;
431 fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
433 while (field->name != NULL) {
434 if (field->flags & VMS_MUST_EXIST) {
435 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
440 fprintf(out_file, ",\n");
442 dump_vmstate_vmsf(out_file, field, indent + 2);
446 fprintf(out_file, "\n%*s]", indent, "");
448 if (vmsd->subsections != NULL) {
449 const VMStateDescription **subsection = vmsd->subsections;
452 fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
454 while (*subsection != NULL) {
456 fprintf(out_file, ",\n");
458 dump_vmstate_vmss(out_file, subsection, indent + 2);
462 fprintf(out_file, "\n%*s]", indent, "");
464 fprintf(out_file, "\n%*s}", indent - 2, "");
467 static void dump_machine_type(FILE *out_file)
471 mc = MACHINE_GET_CLASS(current_machine);
473 fprintf(out_file, " \"vmschkmachine\": {\n");
474 fprintf(out_file, " \"Name\": \"%s\"\n", mc->name);
475 fprintf(out_file, " },\n");
478 void dump_vmstate_json_to_file(FILE *out_file)
483 fprintf(out_file, "{\n");
484 dump_machine_type(out_file);
487 list = object_class_get_list(TYPE_DEVICE, true);
488 for (elt = list; elt; elt = elt->next) {
489 DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
499 fprintf(out_file, ",\n");
501 name = object_class_get_name(OBJECT_CLASS(dc));
502 fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
504 fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
505 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
506 dc->vmsd->version_id);
507 fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
508 dc->vmsd->minimum_version_id);
510 dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
512 fprintf(out_file, "\n%*s}", indent - 2, "");
515 fprintf(out_file, "\n}\n");
519 static int calculate_new_instance_id(const char *idstr)
524 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
525 if (strcmp(idstr, se->idstr) == 0
526 && instance_id <= se->instance_id) {
527 instance_id = se->instance_id + 1;
533 static int calculate_compat_instance_id(const char *idstr)
538 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
543 if (strcmp(idstr, se->compat->idstr) == 0
544 && instance_id <= se->compat->instance_id) {
545 instance_id = se->compat->instance_id + 1;
551 static inline MigrationPriority save_state_priority(SaveStateEntry *se)
554 return se->vmsd->priority;
556 return MIG_PRI_DEFAULT;
559 static void savevm_state_handler_insert(SaveStateEntry *nse)
561 MigrationPriority priority = save_state_priority(nse);
564 assert(priority <= MIG_PRI_MAX);
566 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
567 if (save_state_priority(se) < priority) {
573 QTAILQ_INSERT_BEFORE(se, nse, entry);
575 QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry);
579 /* TODO: Individual devices generally have very little idea about the rest
580 of the system, so instance_id should be removed/replaced.
581 Meanwhile pass -1 as instance_id if you do not already have a clearly
582 distinguishing id for all instances of your device class. */
583 int register_savevm_live(DeviceState *dev,
592 se = g_new0(SaveStateEntry, 1);
593 se->version_id = version_id;
594 se->section_id = savevm_state.global_section_id++;
598 /* if this is a live_savem then set is_ram */
599 if (ops->save_live_setup != NULL) {
604 char *id = qdev_get_dev_path(dev);
606 if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
608 error_report("Path too long for VMState (%s)", id);
616 se->compat = g_new0(CompatEntry, 1);
617 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), idstr);
618 se->compat->instance_id = instance_id == -1 ?
619 calculate_compat_instance_id(idstr) : instance_id;
623 pstrcat(se->idstr, sizeof(se->idstr), idstr);
625 if (instance_id == -1) {
626 se->instance_id = calculate_new_instance_id(se->idstr);
628 se->instance_id = instance_id;
630 assert(!se->compat || se->instance_id == 0);
631 savevm_state_handler_insert(se);
635 void unregister_savevm(DeviceState *dev, const char *idstr, void *opaque)
637 SaveStateEntry *se, *new_se;
641 char *path = qdev_get_dev_path(dev);
643 pstrcpy(id, sizeof(id), path);
644 pstrcat(id, sizeof(id), "/");
648 pstrcat(id, sizeof(id), idstr);
650 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
651 if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
652 QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
659 int vmstate_register_with_alias_id(DeviceState *dev, int instance_id,
660 const VMStateDescription *vmsd,
661 void *opaque, int alias_id,
662 int required_for_version,
667 /* If this triggers, alias support can be dropped for the vmsd. */
668 assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
670 se = g_new0(SaveStateEntry, 1);
671 se->version_id = vmsd->version_id;
672 se->section_id = savevm_state.global_section_id++;
675 se->alias_id = alias_id;
678 char *id = qdev_get_dev_path(dev);
680 if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
682 error_setg(errp, "Path too long for VMState (%s)", id);
690 se->compat = g_new0(CompatEntry, 1);
691 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
692 se->compat->instance_id = instance_id == -1 ?
693 calculate_compat_instance_id(vmsd->name) : instance_id;
697 pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
699 if (instance_id == -1) {
700 se->instance_id = calculate_new_instance_id(se->idstr);
702 se->instance_id = instance_id;
704 assert(!se->compat || se->instance_id == 0);
705 savevm_state_handler_insert(se);
709 void vmstate_unregister(DeviceState *dev, const VMStateDescription *vmsd,
712 SaveStateEntry *se, *new_se;
714 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
715 if (se->vmsd == vmsd && se->opaque == opaque) {
716 QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
723 static int vmstate_load(QEMUFile *f, SaveStateEntry *se)
725 trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
726 if (!se->vmsd) { /* Old style */
727 return se->ops->load_state(f, se->opaque, se->load_version_id);
729 return vmstate_load_state(f, se->vmsd, se->opaque, se->load_version_id);
732 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
734 int64_t old_offset, size;
736 old_offset = qemu_ftell_fast(f);
737 se->ops->save_state(f, se->opaque);
738 size = qemu_ftell_fast(f) - old_offset;
741 json_prop_int(vmdesc, "size", size);
742 json_start_array(vmdesc, "fields");
743 json_start_object(vmdesc, NULL);
744 json_prop_str(vmdesc, "name", "data");
745 json_prop_int(vmdesc, "size", size);
746 json_prop_str(vmdesc, "type", "buffer");
747 json_end_object(vmdesc);
748 json_end_array(vmdesc);
752 static void vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
754 trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
756 vmstate_save_old_style(f, se, vmdesc);
759 vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
763 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
765 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
766 uint8_t section_type)
768 qemu_put_byte(f, section_type);
769 qemu_put_be32(f, se->section_id);
771 if (section_type == QEMU_VM_SECTION_FULL ||
772 section_type == QEMU_VM_SECTION_START) {
774 size_t len = strlen(se->idstr);
775 qemu_put_byte(f, len);
776 qemu_put_buffer(f, (uint8_t *)se->idstr, len);
778 qemu_put_be32(f, se->instance_id);
779 qemu_put_be32(f, se->version_id);
784 * Write a footer onto device sections that catches cases misformatted device
787 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
789 if (migrate_get_current()->send_section_footer) {
790 qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
791 qemu_put_be32(f, se->section_id);
796 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
797 * command and associated data.
799 * @f: File to send command on
800 * @command: Command type to send
801 * @len: Length of associated data
802 * @data: Data associated with command.
804 static void qemu_savevm_command_send(QEMUFile *f,
805 enum qemu_vm_cmd command,
809 trace_savevm_command_send(command, len);
810 qemu_put_byte(f, QEMU_VM_COMMAND);
811 qemu_put_be16(f, (uint16_t)command);
812 qemu_put_be16(f, len);
813 qemu_put_buffer(f, data, len);
817 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
821 trace_savevm_send_ping(value);
822 buf = cpu_to_be32(value);
823 qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
826 void qemu_savevm_send_open_return_path(QEMUFile *f)
828 trace_savevm_send_open_return_path();
829 qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
832 /* We have a buffer of data to send; we don't want that all to be loaded
833 * by the command itself, so the command contains just the length of the
834 * extra buffer that we then send straight after it.
835 * TODO: Must be a better way to organise that
841 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len)
845 if (len > MAX_VM_CMD_PACKAGED_SIZE) {
846 error_report("%s: Unreasonably large packaged state: %zu",
851 tmp = cpu_to_be32(len);
853 trace_qemu_savevm_send_packaged();
854 qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
856 qemu_put_buffer(f, buf, len);
861 /* Send prior to any postcopy transfer */
862 void qemu_savevm_send_postcopy_advise(QEMUFile *f)
865 tmp[0] = cpu_to_be64(ram_pagesize_summary());
866 tmp[1] = cpu_to_be64(qemu_target_page_size());
868 trace_qemu_savevm_send_postcopy_advise();
869 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 16, (uint8_t *)tmp);
872 /* Sent prior to starting the destination running in postcopy, discard pages
873 * that have already been sent but redirtied on the source.
874 * CMD_POSTCOPY_RAM_DISCARD consist of:
876 * byte Length of name field (not including 0)
877 * n x byte RAM block name
878 * byte 0 terminator (just for safety)
879 * n x Byte ranges within the named RAMBlock
880 * be64 Start of the range
883 * name: RAMBlock name that these entries are part of
884 * len: Number of page entries
885 * start_list: 'len' addresses
886 * length_list: 'len' addresses
889 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
891 uint64_t *start_list,
892 uint64_t *length_list)
897 size_t name_len = strlen(name);
899 trace_qemu_savevm_send_postcopy_ram_discard(name, len);
900 assert(name_len < 256);
901 buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
902 buf[0] = postcopy_ram_discard_version;
904 memcpy(buf + 2, name, name_len);
905 tmplen = 2 + name_len;
906 buf[tmplen++] = '\0';
908 for (t = 0; t < len; t++) {
909 stq_be_p(buf + tmplen, start_list[t]);
911 stq_be_p(buf + tmplen, length_list[t]);
914 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
918 /* Get the destination into a state where it can receive postcopy data. */
919 void qemu_savevm_send_postcopy_listen(QEMUFile *f)
921 trace_savevm_send_postcopy_listen();
922 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
925 /* Kick the destination into running */
926 void qemu_savevm_send_postcopy_run(QEMUFile *f)
928 trace_savevm_send_postcopy_run();
929 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
932 bool qemu_savevm_state_blocked(Error **errp)
936 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
937 if (se->vmsd && se->vmsd->unmigratable) {
938 error_setg(errp, "State blocked by non-migratable device '%s'",
946 void qemu_savevm_state_header(QEMUFile *f)
948 trace_savevm_state_header();
949 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
950 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
952 if (migrate_get_current()->send_configuration) {
953 qemu_put_byte(f, QEMU_VM_CONFIGURATION);
954 vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
958 void qemu_savevm_state_begin(QEMUFile *f)
963 trace_savevm_state_begin();
964 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
965 if (!se->ops || !se->ops->save_live_setup) {
968 if (se->ops && se->ops->is_active) {
969 if (!se->ops->is_active(se->opaque)) {
973 save_section_header(f, se, QEMU_VM_SECTION_START);
975 ret = se->ops->save_live_setup(f, se->opaque);
976 save_section_footer(f, se);
978 qemu_file_set_error(f, ret);
985 * this function has three return values:
986 * negative: there was one error, and we have -errno.
987 * 0 : We haven't finished, caller have to go again
988 * 1 : We have finished, we can go to complete phase
990 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
995 trace_savevm_state_iterate();
996 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
997 if (!se->ops || !se->ops->save_live_iterate) {
1000 if (se->ops && se->ops->is_active) {
1001 if (!se->ops->is_active(se->opaque)) {
1006 * In the postcopy phase, any device that doesn't know how to
1007 * do postcopy should have saved it's state in the _complete
1008 * call that's already run, it might get confused if we call
1009 * iterate afterwards.
1011 if (postcopy && !se->ops->save_live_complete_postcopy) {
1014 if (qemu_file_rate_limit(f)) {
1017 trace_savevm_section_start(se->idstr, se->section_id);
1019 save_section_header(f, se, QEMU_VM_SECTION_PART);
1021 ret = se->ops->save_live_iterate(f, se->opaque);
1022 trace_savevm_section_end(se->idstr, se->section_id, ret);
1023 save_section_footer(f, se);
1026 qemu_file_set_error(f, ret);
1029 /* Do not proceed to the next vmstate before this one reported
1030 completion of the current stage. This serializes the migration
1031 and reduces the probability that a faster changing state is
1032 synchronized over and over again. */
1039 static bool should_send_vmdesc(void)
1041 MachineState *machine = MACHINE(qdev_get_machine());
1042 bool in_postcopy = migration_in_postcopy();
1043 return !machine->suppress_vmdesc && !in_postcopy;
1047 * Calls the save_live_complete_postcopy methods
1048 * causing the last few pages to be sent immediately and doing any associated
1050 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1051 * all the other devices, but that happens at the point we switch to postcopy.
1053 void qemu_savevm_state_complete_postcopy(QEMUFile *f)
1058 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1059 if (!se->ops || !se->ops->save_live_complete_postcopy) {
1062 if (se->ops && se->ops->is_active) {
1063 if (!se->ops->is_active(se->opaque)) {
1067 trace_savevm_section_start(se->idstr, se->section_id);
1069 qemu_put_byte(f, QEMU_VM_SECTION_END);
1070 qemu_put_be32(f, se->section_id);
1072 ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1073 trace_savevm_section_end(se->idstr, se->section_id, ret);
1074 save_section_footer(f, se);
1076 qemu_file_set_error(f, ret);
1081 qemu_put_byte(f, QEMU_VM_EOF);
1085 int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only,
1086 bool inactivate_disks)
1092 bool in_postcopy = migration_in_postcopy();
1094 trace_savevm_state_complete_precopy();
1096 cpu_synchronize_all_states();
1098 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1100 (in_postcopy && se->ops->save_live_complete_postcopy) ||
1101 (in_postcopy && !iterable_only) ||
1102 !se->ops->save_live_complete_precopy) {
1106 if (se->ops && se->ops->is_active) {
1107 if (!se->ops->is_active(se->opaque)) {
1111 trace_savevm_section_start(se->idstr, se->section_id);
1113 save_section_header(f, se, QEMU_VM_SECTION_END);
1115 ret = se->ops->save_live_complete_precopy(f, se->opaque);
1116 trace_savevm_section_end(se->idstr, se->section_id, ret);
1117 save_section_footer(f, se);
1119 qemu_file_set_error(f, ret);
1124 if (iterable_only) {
1128 vmdesc = qjson_new();
1129 json_prop_int(vmdesc, "page_size", qemu_target_page_size());
1130 json_start_array(vmdesc, "devices");
1131 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1133 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1136 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1137 trace_savevm_section_skip(se->idstr, se->section_id);
1141 trace_savevm_section_start(se->idstr, se->section_id);
1143 json_start_object(vmdesc, NULL);
1144 json_prop_str(vmdesc, "name", se->idstr);
1145 json_prop_int(vmdesc, "instance_id", se->instance_id);
1147 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1148 vmstate_save(f, se, vmdesc);
1149 trace_savevm_section_end(se->idstr, se->section_id, 0);
1150 save_section_footer(f, se);
1152 json_end_object(vmdesc);
1155 if (inactivate_disks) {
1156 /* Inactivate before sending QEMU_VM_EOF so that the
1157 * bdrv_invalidate_cache_all() on the other end won't fail. */
1158 ret = bdrv_inactivate_all();
1160 qemu_file_set_error(f, ret);
1165 /* Postcopy stream will still be going */
1166 qemu_put_byte(f, QEMU_VM_EOF);
1169 json_end_array(vmdesc);
1170 qjson_finish(vmdesc);
1171 vmdesc_len = strlen(qjson_get_str(vmdesc));
1173 if (should_send_vmdesc()) {
1174 qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1175 qemu_put_be32(f, vmdesc_len);
1176 qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len);
1178 qjson_destroy(vmdesc);
1184 /* Give an estimate of the amount left to be transferred,
1185 * the result is split into the amount for units that can and
1186 * for units that can't do postcopy.
1188 void qemu_savevm_state_pending(QEMUFile *f, uint64_t threshold_size,
1189 uint64_t *res_non_postcopiable,
1190 uint64_t *res_postcopiable)
1194 *res_non_postcopiable = 0;
1195 *res_postcopiable = 0;
1198 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1199 if (!se->ops || !se->ops->save_live_pending) {
1202 if (se->ops && se->ops->is_active) {
1203 if (!se->ops->is_active(se->opaque)) {
1207 se->ops->save_live_pending(f, se->opaque, threshold_size,
1208 res_non_postcopiable, res_postcopiable);
1212 void qemu_savevm_state_cleanup(void)
1216 trace_savevm_state_cleanup();
1217 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1218 if (se->ops && se->ops->cleanup) {
1219 se->ops->cleanup(se->opaque);
1224 static int qemu_savevm_state(QEMUFile *f, Error **errp)
1227 MigrationState *ms = migrate_init();
1228 MigrationStatus status;
1229 ms->to_dst_file = f;
1231 if (migration_is_blocked(errp)) {
1236 if (migrate_use_block()) {
1237 error_setg(errp, "Block migration and snapshots are incompatible");
1242 qemu_mutex_unlock_iothread();
1243 qemu_savevm_state_header(f);
1244 qemu_savevm_state_begin(f);
1245 qemu_mutex_lock_iothread();
1247 while (qemu_file_get_error(f) == 0) {
1248 if (qemu_savevm_state_iterate(f, false) > 0) {
1253 ret = qemu_file_get_error(f);
1255 qemu_savevm_state_complete_precopy(f, false, false);
1256 ret = qemu_file_get_error(f);
1258 qemu_savevm_state_cleanup();
1260 error_setg_errno(errp, -ret, "Error while writing VM state");
1265 status = MIGRATION_STATUS_FAILED;
1267 status = MIGRATION_STATUS_COMPLETED;
1269 migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status);
1271 /* f is outer parameter, it should not stay in global migration state after
1272 * this function finished */
1273 ms->to_dst_file = NULL;
1278 static int qemu_save_device_state(QEMUFile *f)
1282 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1283 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1285 cpu_synchronize_all_states();
1287 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1291 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1294 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1298 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1300 vmstate_save(f, se, NULL);
1302 save_section_footer(f, se);
1305 qemu_put_byte(f, QEMU_VM_EOF);
1307 return qemu_file_get_error(f);
1310 static SaveStateEntry *find_se(const char *idstr, int instance_id)
1314 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1315 if (!strcmp(se->idstr, idstr) &&
1316 (instance_id == se->instance_id ||
1317 instance_id == se->alias_id))
1319 /* Migrating from an older version? */
1320 if (strstr(se->idstr, idstr) && se->compat) {
1321 if (!strcmp(se->compat->idstr, idstr) &&
1322 (instance_id == se->compat->instance_id ||
1323 instance_id == se->alias_id))
1330 enum LoadVMExitCodes {
1331 /* Allow a command to quit all layers of nested loadvm loops */
1335 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis);
1337 /* ------ incoming postcopy messages ------ */
1338 /* 'advise' arrives before any transfers just to tell us that a postcopy
1339 * *might* happen - it might be skipped if precopy transferred everything
1342 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis)
1344 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1345 uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps;
1347 trace_loadvm_postcopy_handle_advise();
1348 if (ps != POSTCOPY_INCOMING_NONE) {
1349 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1353 if (!postcopy_ram_supported_by_host()) {
1354 postcopy_state_set(POSTCOPY_INCOMING_NONE);
1358 remote_pagesize_summary = qemu_get_be64(mis->from_src_file);
1359 local_pagesize_summary = ram_pagesize_summary();
1361 if (remote_pagesize_summary != local_pagesize_summary) {
1363 * This detects two potential causes of mismatch:
1364 * a) A mismatch in host page sizes
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.
1370 * b) The use of different huge page sizes on source/destination
1371 * a more fine grain test is performed during RAM block migration
1372 * but this test here causes a nice early clear failure, and
1373 * also fails when passed to an older qemu that doesn't
1376 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1378 remote_pagesize_summary, local_pagesize_summary);
1382 remote_tps = qemu_get_be64(mis->from_src_file);
1383 if (remote_tps != qemu_target_page_size()) {
1385 * Again, some differences could be dealt with, but for now keep it
1388 error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1389 (int)remote_tps, qemu_target_page_size());
1393 if (ram_postcopy_incoming_init(mis)) {
1397 postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1402 /* After postcopy we will be told to throw some pages away since they're
1403 * dirty and will have to be demand fetched. Must happen before CPU is
1405 * There can be 0..many of these messages, each encoding multiple pages.
1407 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1412 PostcopyState ps = postcopy_state_get();
1414 trace_loadvm_postcopy_ram_handle_discard();
1417 case POSTCOPY_INCOMING_ADVISE:
1419 tmp = postcopy_ram_prepare_discard(mis);
1425 case POSTCOPY_INCOMING_DISCARD:
1426 /* Expected state */
1430 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1434 /* We're expecting a
1436 * a RAM ID string (length byte, name, 0 term)
1437 * then at least 1 16 byte chunk
1439 if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1440 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1444 tmp = qemu_get_byte(mis->from_src_file);
1445 if (tmp != postcopy_ram_discard_version) {
1446 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1450 if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1451 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1454 tmp = qemu_get_byte(mis->from_src_file);
1456 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1460 len -= 3 + strlen(ramid);
1462 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1465 trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1467 uint64_t start_addr, block_length;
1468 start_addr = qemu_get_be64(mis->from_src_file);
1469 block_length = qemu_get_be64(mis->from_src_file);
1472 int ret = ram_discard_range(ramid, start_addr, block_length);
1477 trace_loadvm_postcopy_ram_handle_discard_end();
1483 * Triggered by a postcopy_listen command; this thread takes over reading
1484 * the input stream, leaving the main thread free to carry on loading the rest
1485 * of the device state (from RAM).
1486 * (TODO:This could do with being in a postcopy file - but there again it's
1487 * just another input loop, not that postcopy specific)
1489 static void *postcopy_ram_listen_thread(void *opaque)
1491 QEMUFile *f = opaque;
1492 MigrationIncomingState *mis = migration_incoming_get_current();
1495 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
1496 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1497 qemu_sem_post(&mis->listen_thread_sem);
1498 trace_postcopy_ram_listen_thread_start();
1501 * Because we're a thread and not a coroutine we can't yield
1502 * in qemu_file, and thus we must be blocking now.
1504 qemu_file_set_blocking(f, true);
1505 load_res = qemu_loadvm_state_main(f, mis);
1506 /* And non-blocking again so we don't block in any cleanup */
1507 qemu_file_set_blocking(f, false);
1509 trace_postcopy_ram_listen_thread_exit();
1511 error_report("%s: loadvm failed: %d", __func__, load_res);
1512 qemu_file_set_error(f, load_res);
1513 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1514 MIGRATION_STATUS_FAILED);
1517 * This looks good, but it's possible that the device loading in the
1518 * main thread hasn't finished yet, and so we might not be in 'RUN'
1519 * state yet; wait for the end of the main thread.
1521 qemu_event_wait(&mis->main_thread_load_event);
1523 postcopy_ram_incoming_cleanup(mis);
1527 * If something went wrong then we have a bad state so exit;
1528 * depending how far we got it might be possible at this point
1529 * to leave the guest running and fire MCEs for pages that never
1530 * arrived as a desperate recovery step.
1535 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1536 MIGRATION_STATUS_COMPLETED);
1538 * If everything has worked fine, then the main thread has waited
1539 * for us to start, and we're the last use of the mis.
1540 * (If something broke then qemu will have to exit anyway since it's
1541 * got a bad migration state).
1543 migration_incoming_state_destroy();
1549 /* After this message we must be able to immediately receive postcopy data */
1550 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
1552 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
1553 trace_loadvm_postcopy_handle_listen();
1554 if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
1555 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
1558 if (ps == POSTCOPY_INCOMING_ADVISE) {
1560 * A rare case, we entered listen without having to do any discards,
1561 * so do the setup that's normally done at the time of the 1st discard.
1563 postcopy_ram_prepare_discard(mis);
1567 * Sensitise RAM - can now generate requests for blocks that don't exist
1568 * However, at this point the CPU shouldn't be running, and the IO
1569 * shouldn't be doing anything yet so don't actually expect requests
1571 if (postcopy_ram_enable_notify(mis)) {
1575 if (mis->have_listen_thread) {
1576 error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread");
1580 mis->have_listen_thread = true;
1581 /* Start up the listening thread and wait for it to signal ready */
1582 qemu_sem_init(&mis->listen_thread_sem, 0);
1583 qemu_thread_create(&mis->listen_thread, "postcopy/listen",
1584 postcopy_ram_listen_thread, mis->from_src_file,
1585 QEMU_THREAD_DETACHED);
1586 qemu_sem_wait(&mis->listen_thread_sem);
1587 qemu_sem_destroy(&mis->listen_thread_sem);
1597 static void loadvm_postcopy_handle_run_bh(void *opaque)
1599 Error *local_err = NULL;
1600 HandleRunBhData *data = opaque;
1602 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1605 cpu_synchronize_all_post_init();
1607 qemu_announce_self();
1609 /* Make sure all file formats flush their mutable metadata.
1610 * If we get an error here, just don't restart the VM yet. */
1611 bdrv_invalidate_cache_all(&local_err);
1613 error_report_err(local_err);
1618 trace_loadvm_postcopy_handle_run_cpu_sync();
1619 cpu_synchronize_all_post_init();
1621 trace_loadvm_postcopy_handle_run_vmstart();
1624 /* Hold onto your hats, starting the CPU */
1627 /* leave it paused and let management decide when to start the CPU */
1628 runstate_set(RUN_STATE_PAUSED);
1631 qemu_bh_delete(data->bh);
1635 /* After all discards we can start running and asking for pages */
1636 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
1638 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
1639 HandleRunBhData *data;
1641 trace_loadvm_postcopy_handle_run();
1642 if (ps != POSTCOPY_INCOMING_LISTENING) {
1643 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
1647 data = g_new(HandleRunBhData, 1);
1648 data->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, data);
1649 qemu_bh_schedule(data->bh);
1651 /* We need to finish reading the stream from the package
1652 * and also stop reading anything more from the stream that loaded the
1653 * package (since it's now being read by the listener thread).
1654 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
1660 * Immediately following this command is a blob of data containing an embedded
1661 * chunk of migration stream; read it and load it.
1663 * @mis: Incoming state
1664 * @length: Length of packaged data to read
1666 * Returns: Negative values on error
1669 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
1673 QIOChannelBuffer *bioc;
1675 length = qemu_get_be32(mis->from_src_file);
1676 trace_loadvm_handle_cmd_packaged(length);
1678 if (length > MAX_VM_CMD_PACKAGED_SIZE) {
1679 error_report("Unreasonably large packaged state: %zu", length);
1683 bioc = qio_channel_buffer_new(length);
1684 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer");
1685 ret = qemu_get_buffer(mis->from_src_file,
1688 if (ret != length) {
1689 object_unref(OBJECT(bioc));
1690 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
1692 return (ret < 0) ? ret : -EAGAIN;
1694 bioc->usage += length;
1695 trace_loadvm_handle_cmd_packaged_received(ret);
1697 QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
1699 ret = qemu_loadvm_state_main(packf, mis);
1700 trace_loadvm_handle_cmd_packaged_main(ret);
1702 object_unref(OBJECT(bioc));
1708 * Process an incoming 'QEMU_VM_COMMAND'
1709 * 0 just a normal return
1710 * LOADVM_QUIT All good, but exit the loop
1713 static int loadvm_process_command(QEMUFile *f)
1715 MigrationIncomingState *mis = migration_incoming_get_current();
1720 cmd = qemu_get_be16(f);
1721 len = qemu_get_be16(f);
1723 trace_loadvm_process_command(cmd, len);
1724 if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
1725 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
1729 if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
1730 error_report("%s received with bad length - expecting %zu, got %d",
1731 mig_cmd_args[cmd].name,
1732 (size_t)mig_cmd_args[cmd].len, len);
1737 case MIG_CMD_OPEN_RETURN_PATH:
1738 if (mis->to_src_file) {
1739 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
1740 /* Not really a problem, so don't give up */
1743 mis->to_src_file = qemu_file_get_return_path(f);
1744 if (!mis->to_src_file) {
1745 error_report("CMD_OPEN_RETURN_PATH failed");
1751 tmp32 = qemu_get_be32(f);
1752 trace_loadvm_process_command_ping(tmp32);
1753 if (!mis->to_src_file) {
1754 error_report("CMD_PING (0x%x) received with no return path",
1758 migrate_send_rp_pong(mis, tmp32);
1761 case MIG_CMD_PACKAGED:
1762 return loadvm_handle_cmd_packaged(mis);
1764 case MIG_CMD_POSTCOPY_ADVISE:
1765 return loadvm_postcopy_handle_advise(mis);
1767 case MIG_CMD_POSTCOPY_LISTEN:
1768 return loadvm_postcopy_handle_listen(mis);
1770 case MIG_CMD_POSTCOPY_RUN:
1771 return loadvm_postcopy_handle_run(mis);
1773 case MIG_CMD_POSTCOPY_RAM_DISCARD:
1774 return loadvm_postcopy_ram_handle_discard(mis, len);
1781 * Read a footer off the wire and check that it matches the expected section
1783 * Returns: true if the footer was good
1784 * false if there is a problem (and calls error_report to say why)
1786 static bool check_section_footer(QEMUFile *f, SaveStateEntry *se)
1789 uint32_t read_section_id;
1791 if (!migrate_get_current()->send_section_footer) {
1792 /* No footer to check */
1796 read_mark = qemu_get_byte(f);
1798 if (read_mark != QEMU_VM_SECTION_FOOTER) {
1799 error_report("Missing section footer for %s", se->idstr);
1803 read_section_id = qemu_get_be32(f);
1804 if (read_section_id != se->load_section_id) {
1805 error_report("Mismatched section id in footer for %s -"
1806 " read 0x%x expected 0x%x",
1807 se->idstr, read_section_id, se->load_section_id);
1816 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis)
1818 uint32_t instance_id, version_id, section_id;
1823 /* Read section start */
1824 section_id = qemu_get_be32(f);
1825 if (!qemu_get_counted_string(f, idstr)) {
1826 error_report("Unable to read ID string for section %u",
1830 instance_id = qemu_get_be32(f);
1831 version_id = qemu_get_be32(f);
1833 trace_qemu_loadvm_state_section_startfull(section_id, idstr,
1834 instance_id, version_id);
1835 /* Find savevm section */
1836 se = find_se(idstr, instance_id);
1838 error_report("Unknown savevm section or instance '%s' %d",
1839 idstr, instance_id);
1843 /* Validate version */
1844 if (version_id > se->version_id) {
1845 error_report("savevm: unsupported version %d for '%s' v%d",
1846 version_id, idstr, se->version_id);
1849 se->load_version_id = version_id;
1850 se->load_section_id = section_id;
1852 /* Validate if it is a device's state */
1853 if (xen_enabled() && se->is_ram) {
1854 error_report("loadvm: %s RAM loading not allowed on Xen", idstr);
1858 ret = vmstate_load(f, se);
1860 error_report("error while loading state for instance 0x%x of"
1861 " device '%s'", instance_id, idstr);
1864 if (!check_section_footer(f, se)) {
1872 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis)
1874 uint32_t section_id;
1878 section_id = qemu_get_be32(f);
1880 trace_qemu_loadvm_state_section_partend(section_id);
1881 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1882 if (se->load_section_id == section_id) {
1887 error_report("Unknown savevm section %d", section_id);
1891 ret = vmstate_load(f, se);
1893 error_report("error while loading state section id %d(%s)",
1894 section_id, se->idstr);
1897 if (!check_section_footer(f, se)) {
1904 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
1906 uint8_t section_type;
1909 while ((section_type = qemu_get_byte(f)) != QEMU_VM_EOF) {
1911 trace_qemu_loadvm_state_section(section_type);
1912 switch (section_type) {
1913 case QEMU_VM_SECTION_START:
1914 case QEMU_VM_SECTION_FULL:
1915 ret = qemu_loadvm_section_start_full(f, mis);
1920 case QEMU_VM_SECTION_PART:
1921 case QEMU_VM_SECTION_END:
1922 ret = qemu_loadvm_section_part_end(f, mis);
1927 case QEMU_VM_COMMAND:
1928 ret = loadvm_process_command(f);
1929 trace_qemu_loadvm_state_section_command(ret);
1930 if ((ret < 0) || (ret & LOADVM_QUIT)) {
1935 error_report("Unknown savevm section type %d", section_type);
1943 qemu_file_set_error(f, ret);
1948 int qemu_loadvm_state(QEMUFile *f)
1950 MigrationIncomingState *mis = migration_incoming_get_current();
1951 Error *local_err = NULL;
1955 if (qemu_savevm_state_blocked(&local_err)) {
1956 error_report_err(local_err);
1960 v = qemu_get_be32(f);
1961 if (v != QEMU_VM_FILE_MAGIC) {
1962 error_report("Not a migration stream");
1966 v = qemu_get_be32(f);
1967 if (v == QEMU_VM_FILE_VERSION_COMPAT) {
1968 error_report("SaveVM v2 format is obsolete and don't work anymore");
1971 if (v != QEMU_VM_FILE_VERSION) {
1972 error_report("Unsupported migration stream version");
1976 if (migrate_get_current()->send_configuration) {
1977 if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
1978 error_report("Configuration section missing");
1981 ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
1988 cpu_synchronize_all_pre_loadvm();
1990 ret = qemu_loadvm_state_main(f, mis);
1991 qemu_event_set(&mis->main_thread_load_event);
1993 trace_qemu_loadvm_state_post_main(ret);
1995 if (mis->have_listen_thread) {
1996 /* Listen thread still going, can't clean up yet */
2001 ret = qemu_file_get_error(f);
2005 * Try to read in the VMDESC section as well, so that dumping tools that
2006 * intercept our migration stream have the chance to see it.
2009 /* We've got to be careful; if we don't read the data and just shut the fd
2010 * then the sender can error if we close while it's still sending.
2011 * We also mustn't read data that isn't there; some transports (RDMA)
2012 * will stall waiting for that data when the source has already closed.
2014 if (ret == 0 && should_send_vmdesc()) {
2017 uint8_t section_type = qemu_get_byte(f);
2019 if (section_type != QEMU_VM_VMDESCRIPTION) {
2020 error_report("Expected vmdescription section, but got %d",
2023 * It doesn't seem worth failing at this point since
2024 * we apparently have an otherwise valid VM state
2027 buf = g_malloc(0x1000);
2028 size = qemu_get_be32(f);
2031 uint32_t read_chunk = MIN(size, 0x1000);
2032 qemu_get_buffer(f, buf, read_chunk);
2039 cpu_synchronize_all_post_init();
2044 int save_snapshot(const char *name, Error **errp)
2046 BlockDriverState *bs, *bs1;
2047 QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1;
2050 int saved_vm_running;
2051 uint64_t vm_state_size;
2054 AioContext *aio_context;
2056 if (!bdrv_all_can_snapshot(&bs)) {
2057 error_setg(errp, "Device '%s' is writable but does not support "
2058 "snapshots", bdrv_get_device_name(bs));
2062 /* Delete old snapshots of the same name */
2064 ret = bdrv_all_delete_snapshot(name, &bs1, errp);
2066 error_prepend(errp, "Error while deleting snapshot on device "
2067 "'%s': ", bdrv_get_device_name(bs1));
2072 bs = bdrv_all_find_vmstate_bs();
2074 error_setg(errp, "No block device can accept snapshots");
2077 aio_context = bdrv_get_aio_context(bs);
2079 saved_vm_running = runstate_is_running();
2081 ret = global_state_store();
2083 error_setg(errp, "Error saving global state");
2086 vm_stop(RUN_STATE_SAVE_VM);
2088 bdrv_drain_all_begin();
2090 aio_context_acquire(aio_context);
2092 memset(sn, 0, sizeof(*sn));
2094 /* fill auxiliary fields */
2095 qemu_gettimeofday(&tv);
2096 sn->date_sec = tv.tv_sec;
2097 sn->date_nsec = tv.tv_usec * 1000;
2098 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
2101 ret = bdrv_snapshot_find(bs, old_sn, name);
2103 pstrcpy(sn->name, sizeof(sn->name), old_sn->name);
2104 pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str);
2106 pstrcpy(sn->name, sizeof(sn->name), name);
2109 /* cast below needed for OpenBSD where tv_sec is still 'long' */
2110 localtime_r((const time_t *)&tv.tv_sec, &tm);
2111 strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm);
2114 /* save the VM state */
2115 f = qemu_fopen_bdrv(bs, 1);
2117 error_setg(errp, "Could not open VM state file");
2120 ret = qemu_savevm_state(f, errp);
2121 vm_state_size = qemu_ftell(f);
2127 /* The bdrv_all_create_snapshot() call that follows acquires the AioContext
2128 * for itself. BDRV_POLL_WHILE() does not support nested locking because
2129 * it only releases the lock once. Therefore synchronous I/O will deadlock
2130 * unless we release the AioContext before bdrv_all_create_snapshot().
2132 aio_context_release(aio_context);
2135 ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, &bs);
2137 error_setg(errp, "Error while creating snapshot on '%s'",
2138 bdrv_get_device_name(bs));
2146 aio_context_release(aio_context);
2149 bdrv_drain_all_end();
2151 if (saved_vm_running) {
2157 void qmp_xen_save_devices_state(const char *filename, Error **errp)
2160 QIOChannelFile *ioc;
2161 int saved_vm_running;
2164 saved_vm_running = runstate_is_running();
2165 vm_stop(RUN_STATE_SAVE_VM);
2166 global_state_store_running();
2168 ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT, 0660, errp);
2172 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state");
2173 f = qemu_fopen_channel_output(QIO_CHANNEL(ioc));
2174 ret = qemu_save_device_state(f);
2177 error_setg(errp, QERR_IO_ERROR);
2181 if (saved_vm_running) {
2186 void qmp_xen_load_devices_state(const char *filename, Error **errp)
2189 QIOChannelFile *ioc;
2192 /* Guest must be paused before loading the device state; the RAM state
2193 * will already have been loaded by xc
2195 if (runstate_is_running()) {
2196 error_setg(errp, "Cannot update device state while vm is running");
2199 vm_stop(RUN_STATE_RESTORE_VM);
2201 ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp);
2205 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state");
2206 f = qemu_fopen_channel_input(QIO_CHANNEL(ioc));
2208 ret = qemu_loadvm_state(f);
2211 error_setg(errp, QERR_IO_ERROR);
2213 migration_incoming_state_destroy();
2216 int load_snapshot(const char *name, Error **errp)
2218 BlockDriverState *bs, *bs_vm_state;
2219 QEMUSnapshotInfo sn;
2222 AioContext *aio_context;
2223 MigrationIncomingState *mis = migration_incoming_get_current();
2225 if (!bdrv_all_can_snapshot(&bs)) {
2227 "Device '%s' is writable but does not support snapshots",
2228 bdrv_get_device_name(bs));
2231 ret = bdrv_all_find_snapshot(name, &bs);
2234 "Device '%s' does not have the requested snapshot '%s'",
2235 bdrv_get_device_name(bs), name);
2239 bs_vm_state = bdrv_all_find_vmstate_bs();
2241 error_setg(errp, "No block device supports snapshots");
2244 aio_context = bdrv_get_aio_context(bs_vm_state);
2246 /* Don't even try to load empty VM states */
2247 aio_context_acquire(aio_context);
2248 ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
2249 aio_context_release(aio_context);
2252 } else if (sn.vm_state_size == 0) {
2253 error_setg(errp, "This is a disk-only snapshot. Revert to it "
2254 " offline using qemu-img");
2258 /* Flush all IO requests so they don't interfere with the new state. */
2259 bdrv_drain_all_begin();
2261 ret = bdrv_all_goto_snapshot(name, &bs);
2263 error_setg(errp, "Error %d while activating snapshot '%s' on '%s'",
2264 ret, name, bdrv_get_device_name(bs));
2268 /* restore the VM state */
2269 f = qemu_fopen_bdrv(bs_vm_state, 0);
2271 error_setg(errp, "Could not open VM state file");
2276 qemu_system_reset(SHUTDOWN_CAUSE_NONE);
2277 mis->from_src_file = f;
2279 aio_context_acquire(aio_context);
2280 ret = qemu_loadvm_state(f);
2281 migration_incoming_state_destroy();
2282 aio_context_release(aio_context);
2284 bdrv_drain_all_end();
2287 error_setg(errp, "Error %d while loading VM state", ret);
2294 bdrv_drain_all_end();
2298 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
2300 qemu_ram_set_idstr(mr->ram_block,
2301 memory_region_name(mr), dev);
2304 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
2306 qemu_ram_unset_idstr(mr->ram_block);
2309 void vmstate_register_ram_global(MemoryRegion *mr)
2311 vmstate_register_ram(mr, NULL);
2314 bool vmstate_check_only_migratable(const VMStateDescription *vmsd)
2316 /* check needed if --only-migratable is specified */
2317 if (!migrate_get_current()->only_migratable) {
2321 return !(vmsd && vmsd->unmigratable);