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"
33 #include "hw/xen/xen.h"
35 #include "sysemu/sysemu.h"
36 #include "qemu/timer.h"
37 #include "migration/migration.h"
38 #include "migration/snapshot.h"
40 #include "qemu-file-channel.h"
41 #include "qemu-file.h"
43 #include "postcopy-ram.h"
44 #include "qapi/qmp/qerror.h"
45 #include "qemu/error-report.h"
46 #include "qemu/queue.h"
47 #include "sysemu/cpus.h"
48 #include "exec/memory.h"
49 #include "exec/target_page.h"
50 #include "qmp-commands.h"
52 #include "qemu/bitops.h"
54 #include "block/snapshot.h"
55 #include "qemu/cutils.h"
56 #include "io/channel-buffer.h"
57 #include "io/channel-file.h"
60 #define ETH_P_RARP 0x8035
62 #define ARP_HTYPE_ETH 0x0001
63 #define ARP_PTYPE_IP 0x0800
64 #define ARP_OP_REQUEST_REV 0x3
66 const unsigned int postcopy_ram_discard_version = 0;
68 static bool skip_section_footers;
70 /* Subcommands for QEMU_VM_COMMAND */
72 MIG_CMD_INVALID = 0, /* Must be 0 */
73 MIG_CMD_OPEN_RETURN_PATH, /* Tell the dest to open the Return path */
74 MIG_CMD_PING, /* Request a PONG on the RP */
76 MIG_CMD_POSTCOPY_ADVISE, /* Prior to any page transfers, just
77 warn we might want to do PC */
78 MIG_CMD_POSTCOPY_LISTEN, /* Start listening for incoming
79 pages as it's running. */
80 MIG_CMD_POSTCOPY_RUN, /* Start execution */
82 MIG_CMD_POSTCOPY_RAM_DISCARD, /* A list of pages to discard that
83 were previously sent during
84 precopy but are dirty. */
85 MIG_CMD_PACKAGED, /* Send a wrapped stream within this stream */
89 #define MAX_VM_CMD_PACKAGED_SIZE (1ul << 24)
90 static struct mig_cmd_args {
91 ssize_t len; /* -1 = variable */
94 [MIG_CMD_INVALID] = { .len = -1, .name = "INVALID" },
95 [MIG_CMD_OPEN_RETURN_PATH] = { .len = 0, .name = "OPEN_RETURN_PATH" },
96 [MIG_CMD_PING] = { .len = sizeof(uint32_t), .name = "PING" },
97 [MIG_CMD_POSTCOPY_ADVISE] = { .len = 16, .name = "POSTCOPY_ADVISE" },
98 [MIG_CMD_POSTCOPY_LISTEN] = { .len = 0, .name = "POSTCOPY_LISTEN" },
99 [MIG_CMD_POSTCOPY_RUN] = { .len = 0, .name = "POSTCOPY_RUN" },
100 [MIG_CMD_POSTCOPY_RAM_DISCARD] = {
101 .len = -1, .name = "POSTCOPY_RAM_DISCARD" },
102 [MIG_CMD_PACKAGED] = { .len = 4, .name = "PACKAGED" },
103 [MIG_CMD_MAX] = { .len = -1, .name = "MAX" },
106 static int announce_self_create(uint8_t *buf,
109 /* Ethernet header. */
110 memset(buf, 0xff, 6); /* destination MAC addr */
111 memcpy(buf + 6, mac_addr, 6); /* source MAC addr */
112 *(uint16_t *)(buf + 12) = htons(ETH_P_RARP); /* ethertype */
115 *(uint16_t *)(buf + 14) = htons(ARP_HTYPE_ETH); /* hardware addr space */
116 *(uint16_t *)(buf + 16) = htons(ARP_PTYPE_IP); /* protocol addr space */
117 *(buf + 18) = 6; /* hardware addr length (ethernet) */
118 *(buf + 19) = 4; /* protocol addr length (IPv4) */
119 *(uint16_t *)(buf + 20) = htons(ARP_OP_REQUEST_REV); /* opcode */
120 memcpy(buf + 22, mac_addr, 6); /* source hw addr */
121 memset(buf + 28, 0x00, 4); /* source protocol addr */
122 memcpy(buf + 32, mac_addr, 6); /* target hw addr */
123 memset(buf + 38, 0x00, 4); /* target protocol addr */
125 /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */
126 memset(buf + 42, 0x00, 18);
128 return 60; /* len (FCS will be added by hardware) */
131 static void qemu_announce_self_iter(NICState *nic, void *opaque)
136 trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic->conf->macaddr));
137 len = announce_self_create(buf, nic->conf->macaddr.a);
139 qemu_send_packet_raw(qemu_get_queue(nic), buf, len);
143 static void qemu_announce_self_once(void *opaque)
145 static int count = SELF_ANNOUNCE_ROUNDS;
146 QEMUTimer *timer = *(QEMUTimer **)opaque;
148 qemu_foreach_nic(qemu_announce_self_iter, NULL);
151 /* delay 50ms, 150ms, 250ms, ... */
152 timer_mod(timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) +
153 self_announce_delay(count));
160 void qemu_announce_self(void)
162 static QEMUTimer *timer;
163 timer = timer_new_ms(QEMU_CLOCK_REALTIME, qemu_announce_self_once, &timer);
164 qemu_announce_self_once(&timer);
167 /***********************************************************/
168 /* savevm/loadvm support */
170 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt,
176 qemu_iovec_init_external(&qiov, iov, iovcnt);
177 ret = bdrv_writev_vmstate(opaque, &qiov, pos);
185 static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos,
188 return bdrv_load_vmstate(opaque, buf, pos, size);
191 static int bdrv_fclose(void *opaque)
193 return bdrv_flush(opaque);
196 static const QEMUFileOps bdrv_read_ops = {
197 .get_buffer = block_get_buffer,
201 static const QEMUFileOps bdrv_write_ops = {
202 .writev_buffer = block_writev_buffer,
206 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
209 return qemu_fopen_ops(bs, &bdrv_write_ops);
211 return qemu_fopen_ops(bs, &bdrv_read_ops);
215 /* QEMUFile timer support.
216 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
219 void timer_put(QEMUFile *f, QEMUTimer *ts)
221 uint64_t expire_time;
223 expire_time = timer_expire_time_ns(ts);
224 qemu_put_be64(f, expire_time);
227 void timer_get(QEMUFile *f, QEMUTimer *ts)
229 uint64_t expire_time;
231 expire_time = qemu_get_be64(f);
232 if (expire_time != -1) {
233 timer_mod_ns(ts, expire_time);
240 /* VMState timer support.
241 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
244 static int get_timer(QEMUFile *f, void *pv, size_t size, VMStateField *field)
251 static int put_timer(QEMUFile *f, void *pv, size_t size, VMStateField *field,
260 const VMStateInfo vmstate_info_timer = {
267 typedef struct CompatEntry {
272 typedef struct SaveStateEntry {
273 QTAILQ_ENTRY(SaveStateEntry) entry;
278 /* version id read from the stream */
281 /* section id read from the stream */
284 const VMStateDescription *vmsd;
290 typedef struct SaveState {
291 QTAILQ_HEAD(, SaveStateEntry) handlers;
292 int global_section_id;
293 bool skip_configuration;
296 uint32_t target_page_bits;
299 static SaveState savevm_state = {
300 .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
301 .global_section_id = 0,
302 .skip_configuration = false,
305 void savevm_skip_configuration(void)
307 savevm_state.skip_configuration = true;
311 static void configuration_pre_save(void *opaque)
313 SaveState *state = opaque;
314 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
316 state->len = strlen(current_name);
317 state->name = current_name;
318 state->target_page_bits = qemu_target_page_bits();
321 static int configuration_pre_load(void *opaque)
323 SaveState *state = opaque;
325 /* If there is no target-page-bits subsection it means the source
326 * predates the variable-target-page-bits support and is using the
327 * minimum possible value for this CPU.
329 state->target_page_bits = qemu_target_page_bits_min();
333 static int configuration_post_load(void *opaque, int version_id)
335 SaveState *state = opaque;
336 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
338 if (strncmp(state->name, current_name, state->len) != 0) {
339 error_report("Machine type received is '%.*s' and local is '%s'",
340 (int) state->len, state->name, current_name);
344 if (state->target_page_bits != qemu_target_page_bits()) {
345 error_report("Received TARGET_PAGE_BITS is %d but local is %d",
346 state->target_page_bits, qemu_target_page_bits());
353 /* The target-page-bits subsection is present only if the
354 * target page size is not the same as the default (ie the
355 * minimum page size for a variable-page-size guest CPU).
356 * If it is present then it contains the actual target page
357 * bits for the machine, and migration will fail if the
358 * two ends don't agree about it.
360 static bool vmstate_target_page_bits_needed(void *opaque)
362 return qemu_target_page_bits()
363 > qemu_target_page_bits_min();
366 static const VMStateDescription vmstate_target_page_bits = {
367 .name = "configuration/target-page-bits",
369 .minimum_version_id = 1,
370 .needed = vmstate_target_page_bits_needed,
371 .fields = (VMStateField[]) {
372 VMSTATE_UINT32(target_page_bits, SaveState),
373 VMSTATE_END_OF_LIST()
377 static const VMStateDescription vmstate_configuration = {
378 .name = "configuration",
380 .pre_load = configuration_pre_load,
381 .post_load = configuration_post_load,
382 .pre_save = configuration_pre_save,
383 .fields = (VMStateField[]) {
384 VMSTATE_UINT32(len, SaveState),
385 VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len),
386 VMSTATE_END_OF_LIST()
388 .subsections = (const VMStateDescription*[]) {
389 &vmstate_target_page_bits,
394 static void dump_vmstate_vmsd(FILE *out_file,
395 const VMStateDescription *vmsd, int indent,
398 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
401 fprintf(out_file, "%*s{\n", indent, "");
403 fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
404 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
406 fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
407 field->field_exists ? "true" : "false");
408 fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
409 if (field->vmsd != NULL) {
410 fprintf(out_file, ",\n");
411 dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
413 fprintf(out_file, "\n%*s}", indent - 2, "");
416 static void dump_vmstate_vmss(FILE *out_file,
417 const VMStateDescription **subsection,
420 if (*subsection != NULL) {
421 dump_vmstate_vmsd(out_file, *subsection, indent, true);
425 static void dump_vmstate_vmsd(FILE *out_file,
426 const VMStateDescription *vmsd, int indent,
430 fprintf(out_file, "%*s{\n", indent, "");
432 fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
435 fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
436 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
438 fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
439 vmsd->minimum_version_id);
440 if (vmsd->fields != NULL) {
441 const VMStateField *field = vmsd->fields;
444 fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
446 while (field->name != NULL) {
447 if (field->flags & VMS_MUST_EXIST) {
448 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
453 fprintf(out_file, ",\n");
455 dump_vmstate_vmsf(out_file, field, indent + 2);
459 fprintf(out_file, "\n%*s]", indent, "");
461 if (vmsd->subsections != NULL) {
462 const VMStateDescription **subsection = vmsd->subsections;
465 fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
467 while (*subsection != NULL) {
469 fprintf(out_file, ",\n");
471 dump_vmstate_vmss(out_file, subsection, indent + 2);
475 fprintf(out_file, "\n%*s]", indent, "");
477 fprintf(out_file, "\n%*s}", indent - 2, "");
480 static void dump_machine_type(FILE *out_file)
484 mc = MACHINE_GET_CLASS(current_machine);
486 fprintf(out_file, " \"vmschkmachine\": {\n");
487 fprintf(out_file, " \"Name\": \"%s\"\n", mc->name);
488 fprintf(out_file, " },\n");
491 void dump_vmstate_json_to_file(FILE *out_file)
496 fprintf(out_file, "{\n");
497 dump_machine_type(out_file);
500 list = object_class_get_list(TYPE_DEVICE, true);
501 for (elt = list; elt; elt = elt->next) {
502 DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
512 fprintf(out_file, ",\n");
514 name = object_class_get_name(OBJECT_CLASS(dc));
515 fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
517 fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
518 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
519 dc->vmsd->version_id);
520 fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
521 dc->vmsd->minimum_version_id);
523 dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
525 fprintf(out_file, "\n%*s}", indent - 2, "");
528 fprintf(out_file, "\n}\n");
532 static int calculate_new_instance_id(const char *idstr)
537 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
538 if (strcmp(idstr, se->idstr) == 0
539 && instance_id <= se->instance_id) {
540 instance_id = se->instance_id + 1;
546 static int calculate_compat_instance_id(const char *idstr)
551 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
556 if (strcmp(idstr, se->compat->idstr) == 0
557 && instance_id <= se->compat->instance_id) {
558 instance_id = se->compat->instance_id + 1;
564 static inline MigrationPriority save_state_priority(SaveStateEntry *se)
567 return se->vmsd->priority;
569 return MIG_PRI_DEFAULT;
572 static void savevm_state_handler_insert(SaveStateEntry *nse)
574 MigrationPriority priority = save_state_priority(nse);
577 assert(priority <= MIG_PRI_MAX);
579 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
580 if (save_state_priority(se) < priority) {
586 QTAILQ_INSERT_BEFORE(se, nse, entry);
588 QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry);
592 /* TODO: Individual devices generally have very little idea about the rest
593 of the system, so instance_id should be removed/replaced.
594 Meanwhile pass -1 as instance_id if you do not already have a clearly
595 distinguishing id for all instances of your device class. */
596 int register_savevm_live(DeviceState *dev,
605 se = g_new0(SaveStateEntry, 1);
606 se->version_id = version_id;
607 se->section_id = savevm_state.global_section_id++;
611 /* if this is a live_savem then set is_ram */
612 if (ops->save_live_setup != NULL) {
617 char *id = qdev_get_dev_path(dev);
619 if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
621 error_report("Path too long for VMState (%s)", id);
629 se->compat = g_new0(CompatEntry, 1);
630 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), idstr);
631 se->compat->instance_id = instance_id == -1 ?
632 calculate_compat_instance_id(idstr) : instance_id;
636 pstrcat(se->idstr, sizeof(se->idstr), idstr);
638 if (instance_id == -1) {
639 se->instance_id = calculate_new_instance_id(se->idstr);
641 se->instance_id = instance_id;
643 assert(!se->compat || se->instance_id == 0);
644 savevm_state_handler_insert(se);
648 void unregister_savevm(DeviceState *dev, const char *idstr, void *opaque)
650 SaveStateEntry *se, *new_se;
654 char *path = qdev_get_dev_path(dev);
656 pstrcpy(id, sizeof(id), path);
657 pstrcat(id, sizeof(id), "/");
661 pstrcat(id, sizeof(id), idstr);
663 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
664 if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
665 QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
672 int vmstate_register_with_alias_id(DeviceState *dev, int instance_id,
673 const VMStateDescription *vmsd,
674 void *opaque, int alias_id,
675 int required_for_version,
680 /* If this triggers, alias support can be dropped for the vmsd. */
681 assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
683 se = g_new0(SaveStateEntry, 1);
684 se->version_id = vmsd->version_id;
685 se->section_id = savevm_state.global_section_id++;
688 se->alias_id = alias_id;
691 char *id = qdev_get_dev_path(dev);
693 if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >=
695 error_setg(errp, "Path too long for VMState (%s)", id);
703 se->compat = g_new0(CompatEntry, 1);
704 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
705 se->compat->instance_id = instance_id == -1 ?
706 calculate_compat_instance_id(vmsd->name) : instance_id;
710 pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
712 if (instance_id == -1) {
713 se->instance_id = calculate_new_instance_id(se->idstr);
715 se->instance_id = instance_id;
717 assert(!se->compat || se->instance_id == 0);
718 savevm_state_handler_insert(se);
722 void vmstate_unregister(DeviceState *dev, const VMStateDescription *vmsd,
725 SaveStateEntry *se, *new_se;
727 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
728 if (se->vmsd == vmsd && se->opaque == opaque) {
729 QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
736 static int vmstate_load(QEMUFile *f, SaveStateEntry *se)
738 trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
739 if (!se->vmsd) { /* Old style */
740 return se->ops->load_state(f, se->opaque, se->load_version_id);
742 return vmstate_load_state(f, se->vmsd, se->opaque, se->load_version_id);
745 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
747 int64_t old_offset, size;
749 old_offset = qemu_ftell_fast(f);
750 se->ops->save_state(f, se->opaque);
751 size = qemu_ftell_fast(f) - old_offset;
754 json_prop_int(vmdesc, "size", size);
755 json_start_array(vmdesc, "fields");
756 json_start_object(vmdesc, NULL);
757 json_prop_str(vmdesc, "name", "data");
758 json_prop_int(vmdesc, "size", size);
759 json_prop_str(vmdesc, "type", "buffer");
760 json_end_object(vmdesc);
761 json_end_array(vmdesc);
765 static void vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
767 trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
769 vmstate_save_old_style(f, se, vmdesc);
772 vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
775 void savevm_skip_section_footers(void)
777 skip_section_footers = true;
781 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
783 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
784 uint8_t section_type)
786 qemu_put_byte(f, section_type);
787 qemu_put_be32(f, se->section_id);
789 if (section_type == QEMU_VM_SECTION_FULL ||
790 section_type == QEMU_VM_SECTION_START) {
792 size_t len = strlen(se->idstr);
793 qemu_put_byte(f, len);
794 qemu_put_buffer(f, (uint8_t *)se->idstr, len);
796 qemu_put_be32(f, se->instance_id);
797 qemu_put_be32(f, se->version_id);
802 * Write a footer onto device sections that catches cases misformatted device
805 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
807 if (!skip_section_footers) {
808 qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
809 qemu_put_be32(f, se->section_id);
814 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
815 * command and associated data.
817 * @f: File to send command on
818 * @command: Command type to send
819 * @len: Length of associated data
820 * @data: Data associated with command.
822 static void qemu_savevm_command_send(QEMUFile *f,
823 enum qemu_vm_cmd command,
827 trace_savevm_command_send(command, len);
828 qemu_put_byte(f, QEMU_VM_COMMAND);
829 qemu_put_be16(f, (uint16_t)command);
830 qemu_put_be16(f, len);
831 qemu_put_buffer(f, data, len);
835 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
839 trace_savevm_send_ping(value);
840 buf = cpu_to_be32(value);
841 qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
844 void qemu_savevm_send_open_return_path(QEMUFile *f)
846 trace_savevm_send_open_return_path();
847 qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
850 /* We have a buffer of data to send; we don't want that all to be loaded
851 * by the command itself, so the command contains just the length of the
852 * extra buffer that we then send straight after it.
853 * TODO: Must be a better way to organise that
859 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len)
863 if (len > MAX_VM_CMD_PACKAGED_SIZE) {
864 error_report("%s: Unreasonably large packaged state: %zu",
869 tmp = cpu_to_be32(len);
871 trace_qemu_savevm_send_packaged();
872 qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
874 qemu_put_buffer(f, buf, len);
879 /* Send prior to any postcopy transfer */
880 void qemu_savevm_send_postcopy_advise(QEMUFile *f)
883 tmp[0] = cpu_to_be64(ram_pagesize_summary());
884 tmp[1] = cpu_to_be64(qemu_target_page_size());
886 trace_qemu_savevm_send_postcopy_advise();
887 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 16, (uint8_t *)tmp);
890 /* Sent prior to starting the destination running in postcopy, discard pages
891 * that have already been sent but redirtied on the source.
892 * CMD_POSTCOPY_RAM_DISCARD consist of:
894 * byte Length of name field (not including 0)
895 * n x byte RAM block name
896 * byte 0 terminator (just for safety)
897 * n x Byte ranges within the named RAMBlock
898 * be64 Start of the range
901 * name: RAMBlock name that these entries are part of
902 * len: Number of page entries
903 * start_list: 'len' addresses
904 * length_list: 'len' addresses
907 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
909 uint64_t *start_list,
910 uint64_t *length_list)
915 size_t name_len = strlen(name);
917 trace_qemu_savevm_send_postcopy_ram_discard(name, len);
918 assert(name_len < 256);
919 buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
920 buf[0] = postcopy_ram_discard_version;
922 memcpy(buf + 2, name, name_len);
923 tmplen = 2 + name_len;
924 buf[tmplen++] = '\0';
926 for (t = 0; t < len; t++) {
927 stq_be_p(buf + tmplen, start_list[t]);
929 stq_be_p(buf + tmplen, length_list[t]);
932 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
936 /* Get the destination into a state where it can receive postcopy data. */
937 void qemu_savevm_send_postcopy_listen(QEMUFile *f)
939 trace_savevm_send_postcopy_listen();
940 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
943 /* Kick the destination into running */
944 void qemu_savevm_send_postcopy_run(QEMUFile *f)
946 trace_savevm_send_postcopy_run();
947 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
950 bool qemu_savevm_state_blocked(Error **errp)
954 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
955 if (se->vmsd && se->vmsd->unmigratable) {
956 error_setg(errp, "State blocked by non-migratable device '%s'",
964 static bool enforce_config_section(void)
966 MachineState *machine = MACHINE(qdev_get_machine());
967 return machine->enforce_config_section;
970 void qemu_savevm_state_header(QEMUFile *f)
972 trace_savevm_state_header();
973 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
974 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
976 if (!savevm_state.skip_configuration || enforce_config_section()) {
977 qemu_put_byte(f, QEMU_VM_CONFIGURATION);
978 vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
983 void qemu_savevm_state_begin(QEMUFile *f)
988 trace_savevm_state_begin();
989 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
990 if (!se->ops || !se->ops->save_live_setup) {
993 if (se->ops && se->ops->is_active) {
994 if (!se->ops->is_active(se->opaque)) {
998 save_section_header(f, se, QEMU_VM_SECTION_START);
1000 ret = se->ops->save_live_setup(f, se->opaque);
1001 save_section_footer(f, se);
1003 qemu_file_set_error(f, ret);
1010 * this function has three return values:
1011 * negative: there was one error, and we have -errno.
1012 * 0 : We haven't finished, caller have to go again
1013 * 1 : We have finished, we can go to complete phase
1015 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
1020 trace_savevm_state_iterate();
1021 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1022 if (!se->ops || !se->ops->save_live_iterate) {
1025 if (se->ops && se->ops->is_active) {
1026 if (!se->ops->is_active(se->opaque)) {
1031 * In the postcopy phase, any device that doesn't know how to
1032 * do postcopy should have saved it's state in the _complete
1033 * call that's already run, it might get confused if we call
1034 * iterate afterwards.
1036 if (postcopy && !se->ops->save_live_complete_postcopy) {
1039 if (qemu_file_rate_limit(f)) {
1042 trace_savevm_section_start(se->idstr, se->section_id);
1044 save_section_header(f, se, QEMU_VM_SECTION_PART);
1046 ret = se->ops->save_live_iterate(f, se->opaque);
1047 trace_savevm_section_end(se->idstr, se->section_id, ret);
1048 save_section_footer(f, se);
1051 qemu_file_set_error(f, ret);
1054 /* Do not proceed to the next vmstate before this one reported
1055 completion of the current stage. This serializes the migration
1056 and reduces the probability that a faster changing state is
1057 synchronized over and over again. */
1064 static bool should_send_vmdesc(void)
1066 MachineState *machine = MACHINE(qdev_get_machine());
1067 bool in_postcopy = migration_in_postcopy();
1068 return !machine->suppress_vmdesc && !in_postcopy;
1072 * Calls the save_live_complete_postcopy methods
1073 * causing the last few pages to be sent immediately and doing any associated
1075 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1076 * all the other devices, but that happens at the point we switch to postcopy.
1078 void qemu_savevm_state_complete_postcopy(QEMUFile *f)
1083 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1084 if (!se->ops || !se->ops->save_live_complete_postcopy) {
1087 if (se->ops && se->ops->is_active) {
1088 if (!se->ops->is_active(se->opaque)) {
1092 trace_savevm_section_start(se->idstr, se->section_id);
1094 qemu_put_byte(f, QEMU_VM_SECTION_END);
1095 qemu_put_be32(f, se->section_id);
1097 ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1098 trace_savevm_section_end(se->idstr, se->section_id, ret);
1099 save_section_footer(f, se);
1101 qemu_file_set_error(f, ret);
1106 qemu_put_byte(f, QEMU_VM_EOF);
1110 void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only)
1116 bool in_postcopy = migration_in_postcopy();
1118 trace_savevm_state_complete_precopy();
1120 cpu_synchronize_all_states();
1122 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1124 (in_postcopy && se->ops->save_live_complete_postcopy) ||
1125 (in_postcopy && !iterable_only) ||
1126 !se->ops->save_live_complete_precopy) {
1130 if (se->ops && se->ops->is_active) {
1131 if (!se->ops->is_active(se->opaque)) {
1135 trace_savevm_section_start(se->idstr, se->section_id);
1137 save_section_header(f, se, QEMU_VM_SECTION_END);
1139 ret = se->ops->save_live_complete_precopy(f, se->opaque);
1140 trace_savevm_section_end(se->idstr, se->section_id, ret);
1141 save_section_footer(f, se);
1143 qemu_file_set_error(f, ret);
1148 if (iterable_only) {
1152 vmdesc = qjson_new();
1153 json_prop_int(vmdesc, "page_size", qemu_target_page_size());
1154 json_start_array(vmdesc, "devices");
1155 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1157 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1160 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1161 trace_savevm_section_skip(se->idstr, se->section_id);
1165 trace_savevm_section_start(se->idstr, se->section_id);
1167 json_start_object(vmdesc, NULL);
1168 json_prop_str(vmdesc, "name", se->idstr);
1169 json_prop_int(vmdesc, "instance_id", se->instance_id);
1171 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1172 vmstate_save(f, se, vmdesc);
1173 trace_savevm_section_end(se->idstr, se->section_id, 0);
1174 save_section_footer(f, se);
1176 json_end_object(vmdesc);
1180 /* Postcopy stream will still be going */
1181 qemu_put_byte(f, QEMU_VM_EOF);
1184 json_end_array(vmdesc);
1185 qjson_finish(vmdesc);
1186 vmdesc_len = strlen(qjson_get_str(vmdesc));
1188 if (should_send_vmdesc()) {
1189 qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1190 qemu_put_be32(f, vmdesc_len);
1191 qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len);
1193 qjson_destroy(vmdesc);
1198 /* Give an estimate of the amount left to be transferred,
1199 * the result is split into the amount for units that can and
1200 * for units that can't do postcopy.
1202 void qemu_savevm_state_pending(QEMUFile *f, uint64_t threshold_size,
1203 uint64_t *res_non_postcopiable,
1204 uint64_t *res_postcopiable)
1208 *res_non_postcopiable = 0;
1209 *res_postcopiable = 0;
1212 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1213 if (!se->ops || !se->ops->save_live_pending) {
1216 if (se->ops && se->ops->is_active) {
1217 if (!se->ops->is_active(se->opaque)) {
1221 se->ops->save_live_pending(f, se->opaque, threshold_size,
1222 res_non_postcopiable, res_postcopiable);
1226 void qemu_savevm_state_cleanup(void)
1230 trace_savevm_state_cleanup();
1231 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1232 if (se->ops && se->ops->cleanup) {
1233 se->ops->cleanup(se->opaque);
1238 static int qemu_savevm_state(QEMUFile *f, Error **errp)
1241 MigrationState *ms = migrate_init();
1242 MigrationStatus status;
1243 ms->to_dst_file = f;
1245 if (migration_is_blocked(errp)) {
1250 if (migrate_use_block()) {
1251 error_setg(errp, "Block migration and snapshots are incompatible");
1256 qemu_mutex_unlock_iothread();
1257 qemu_savevm_state_header(f);
1258 qemu_savevm_state_begin(f);
1259 qemu_mutex_lock_iothread();
1261 while (qemu_file_get_error(f) == 0) {
1262 if (qemu_savevm_state_iterate(f, false) > 0) {
1267 ret = qemu_file_get_error(f);
1269 qemu_savevm_state_complete_precopy(f, false);
1270 ret = qemu_file_get_error(f);
1272 qemu_savevm_state_cleanup();
1274 error_setg_errno(errp, -ret, "Error while writing VM state");
1279 status = MIGRATION_STATUS_FAILED;
1281 status = MIGRATION_STATUS_COMPLETED;
1283 migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status);
1285 /* f is outer parameter, it should not stay in global migration state after
1286 * this function finished */
1287 ms->to_dst_file = NULL;
1292 static int qemu_save_device_state(QEMUFile *f)
1296 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1297 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1299 cpu_synchronize_all_states();
1301 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1305 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1308 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1312 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1314 vmstate_save(f, se, NULL);
1316 save_section_footer(f, se);
1319 qemu_put_byte(f, QEMU_VM_EOF);
1321 return qemu_file_get_error(f);
1324 static SaveStateEntry *find_se(const char *idstr, int instance_id)
1328 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1329 if (!strcmp(se->idstr, idstr) &&
1330 (instance_id == se->instance_id ||
1331 instance_id == se->alias_id))
1333 /* Migrating from an older version? */
1334 if (strstr(se->idstr, idstr) && se->compat) {
1335 if (!strcmp(se->compat->idstr, idstr) &&
1336 (instance_id == se->compat->instance_id ||
1337 instance_id == se->alias_id))
1344 enum LoadVMExitCodes {
1345 /* Allow a command to quit all layers of nested loadvm loops */
1349 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis);
1351 /* ------ incoming postcopy messages ------ */
1352 /* 'advise' arrives before any transfers just to tell us that a postcopy
1353 * *might* happen - it might be skipped if precopy transferred everything
1356 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis)
1358 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1359 uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps;
1361 trace_loadvm_postcopy_handle_advise();
1362 if (ps != POSTCOPY_INCOMING_NONE) {
1363 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1367 if (!postcopy_ram_supported_by_host()) {
1368 postcopy_state_set(POSTCOPY_INCOMING_NONE);
1372 remote_pagesize_summary = qemu_get_be64(mis->from_src_file);
1373 local_pagesize_summary = ram_pagesize_summary();
1375 if (remote_pagesize_summary != local_pagesize_summary) {
1377 * This detects two potential causes of mismatch:
1378 * a) A mismatch in host page sizes
1379 * Some combinations of mismatch are probably possible but it gets
1380 * a bit more complicated. In particular we need to place whole
1381 * host pages on the dest at once, and we need to ensure that we
1382 * handle dirtying to make sure we never end up sending part of
1383 * a hostpage on it's own.
1384 * b) The use of different huge page sizes on source/destination
1385 * a more fine grain test is performed during RAM block migration
1386 * but this test here causes a nice early clear failure, and
1387 * also fails when passed to an older qemu that doesn't
1390 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1392 remote_pagesize_summary, local_pagesize_summary);
1396 remote_tps = qemu_get_be64(mis->from_src_file);
1397 if (remote_tps != qemu_target_page_size()) {
1399 * Again, some differences could be dealt with, but for now keep it
1402 error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1403 (int)remote_tps, qemu_target_page_size());
1407 if (ram_postcopy_incoming_init(mis)) {
1411 postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1416 /* After postcopy we will be told to throw some pages away since they're
1417 * dirty and will have to be demand fetched. Must happen before CPU is
1419 * There can be 0..many of these messages, each encoding multiple pages.
1421 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1426 PostcopyState ps = postcopy_state_get();
1428 trace_loadvm_postcopy_ram_handle_discard();
1431 case POSTCOPY_INCOMING_ADVISE:
1433 tmp = postcopy_ram_prepare_discard(mis);
1439 case POSTCOPY_INCOMING_DISCARD:
1440 /* Expected state */
1444 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1448 /* We're expecting a
1450 * a RAM ID string (length byte, name, 0 term)
1451 * then at least 1 16 byte chunk
1453 if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1454 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1458 tmp = qemu_get_byte(mis->from_src_file);
1459 if (tmp != postcopy_ram_discard_version) {
1460 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1464 if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1465 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1468 tmp = qemu_get_byte(mis->from_src_file);
1470 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1474 len -= 3 + strlen(ramid);
1476 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1479 trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1481 uint64_t start_addr, block_length;
1482 start_addr = qemu_get_be64(mis->from_src_file);
1483 block_length = qemu_get_be64(mis->from_src_file);
1486 int ret = ram_discard_range(ramid, start_addr, block_length);
1491 trace_loadvm_postcopy_ram_handle_discard_end();
1497 * Triggered by a postcopy_listen command; this thread takes over reading
1498 * the input stream, leaving the main thread free to carry on loading the rest
1499 * of the device state (from RAM).
1500 * (TODO:This could do with being in a postcopy file - but there again it's
1501 * just another input loop, not that postcopy specific)
1503 static void *postcopy_ram_listen_thread(void *opaque)
1505 QEMUFile *f = opaque;
1506 MigrationIncomingState *mis = migration_incoming_get_current();
1509 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
1510 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1511 qemu_sem_post(&mis->listen_thread_sem);
1512 trace_postcopy_ram_listen_thread_start();
1515 * Because we're a thread and not a coroutine we can't yield
1516 * in qemu_file, and thus we must be blocking now.
1518 qemu_file_set_blocking(f, true);
1519 load_res = qemu_loadvm_state_main(f, mis);
1520 /* And non-blocking again so we don't block in any cleanup */
1521 qemu_file_set_blocking(f, false);
1523 trace_postcopy_ram_listen_thread_exit();
1525 error_report("%s: loadvm failed: %d", __func__, load_res);
1526 qemu_file_set_error(f, load_res);
1527 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1528 MIGRATION_STATUS_FAILED);
1531 * This looks good, but it's possible that the device loading in the
1532 * main thread hasn't finished yet, and so we might not be in 'RUN'
1533 * state yet; wait for the end of the main thread.
1535 qemu_event_wait(&mis->main_thread_load_event);
1537 postcopy_ram_incoming_cleanup(mis);
1541 * If something went wrong then we have a bad state so exit;
1542 * depending how far we got it might be possible at this point
1543 * to leave the guest running and fire MCEs for pages that never
1544 * arrived as a desperate recovery step.
1549 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1550 MIGRATION_STATUS_COMPLETED);
1552 * If everything has worked fine, then the main thread has waited
1553 * for us to start, and we're the last use of the mis.
1554 * (If something broke then qemu will have to exit anyway since it's
1555 * got a bad migration state).
1557 migration_incoming_state_destroy();
1563 /* After this message we must be able to immediately receive postcopy data */
1564 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
1566 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
1567 trace_loadvm_postcopy_handle_listen();
1568 if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
1569 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
1572 if (ps == POSTCOPY_INCOMING_ADVISE) {
1574 * A rare case, we entered listen without having to do any discards,
1575 * so do the setup that's normally done at the time of the 1st discard.
1577 postcopy_ram_prepare_discard(mis);
1581 * Sensitise RAM - can now generate requests for blocks that don't exist
1582 * However, at this point the CPU shouldn't be running, and the IO
1583 * shouldn't be doing anything yet so don't actually expect requests
1585 if (postcopy_ram_enable_notify(mis)) {
1589 if (mis->have_listen_thread) {
1590 error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread");
1594 mis->have_listen_thread = true;
1595 /* Start up the listening thread and wait for it to signal ready */
1596 qemu_sem_init(&mis->listen_thread_sem, 0);
1597 qemu_thread_create(&mis->listen_thread, "postcopy/listen",
1598 postcopy_ram_listen_thread, mis->from_src_file,
1599 QEMU_THREAD_DETACHED);
1600 qemu_sem_wait(&mis->listen_thread_sem);
1601 qemu_sem_destroy(&mis->listen_thread_sem);
1611 static void loadvm_postcopy_handle_run_bh(void *opaque)
1613 Error *local_err = NULL;
1614 HandleRunBhData *data = opaque;
1616 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1619 cpu_synchronize_all_post_init();
1621 qemu_announce_self();
1623 /* Make sure all file formats flush their mutable metadata.
1624 * If we get an error here, just don't restart the VM yet. */
1625 bdrv_invalidate_cache_all(&local_err);
1627 error_report_err(local_err);
1632 trace_loadvm_postcopy_handle_run_cpu_sync();
1633 cpu_synchronize_all_post_init();
1635 trace_loadvm_postcopy_handle_run_vmstart();
1638 /* Hold onto your hats, starting the CPU */
1641 /* leave it paused and let management decide when to start the CPU */
1642 runstate_set(RUN_STATE_PAUSED);
1645 qemu_bh_delete(data->bh);
1649 /* After all discards we can start running and asking for pages */
1650 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
1652 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
1653 HandleRunBhData *data;
1655 trace_loadvm_postcopy_handle_run();
1656 if (ps != POSTCOPY_INCOMING_LISTENING) {
1657 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
1661 data = g_new(HandleRunBhData, 1);
1662 data->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, data);
1663 qemu_bh_schedule(data->bh);
1665 /* We need to finish reading the stream from the package
1666 * and also stop reading anything more from the stream that loaded the
1667 * package (since it's now being read by the listener thread).
1668 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
1674 * Immediately following this command is a blob of data containing an embedded
1675 * chunk of migration stream; read it and load it.
1677 * @mis: Incoming state
1678 * @length: Length of packaged data to read
1680 * Returns: Negative values on error
1683 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
1687 QIOChannelBuffer *bioc;
1689 length = qemu_get_be32(mis->from_src_file);
1690 trace_loadvm_handle_cmd_packaged(length);
1692 if (length > MAX_VM_CMD_PACKAGED_SIZE) {
1693 error_report("Unreasonably large packaged state: %zu", length);
1697 bioc = qio_channel_buffer_new(length);
1698 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer");
1699 ret = qemu_get_buffer(mis->from_src_file,
1702 if (ret != length) {
1703 object_unref(OBJECT(bioc));
1704 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
1706 return (ret < 0) ? ret : -EAGAIN;
1708 bioc->usage += length;
1709 trace_loadvm_handle_cmd_packaged_received(ret);
1711 QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
1713 ret = qemu_loadvm_state_main(packf, mis);
1714 trace_loadvm_handle_cmd_packaged_main(ret);
1716 object_unref(OBJECT(bioc));
1722 * Process an incoming 'QEMU_VM_COMMAND'
1723 * 0 just a normal return
1724 * LOADVM_QUIT All good, but exit the loop
1727 static int loadvm_process_command(QEMUFile *f)
1729 MigrationIncomingState *mis = migration_incoming_get_current();
1734 cmd = qemu_get_be16(f);
1735 len = qemu_get_be16(f);
1737 trace_loadvm_process_command(cmd, len);
1738 if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
1739 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
1743 if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
1744 error_report("%s received with bad length - expecting %zu, got %d",
1745 mig_cmd_args[cmd].name,
1746 (size_t)mig_cmd_args[cmd].len, len);
1751 case MIG_CMD_OPEN_RETURN_PATH:
1752 if (mis->to_src_file) {
1753 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
1754 /* Not really a problem, so don't give up */
1757 mis->to_src_file = qemu_file_get_return_path(f);
1758 if (!mis->to_src_file) {
1759 error_report("CMD_OPEN_RETURN_PATH failed");
1765 tmp32 = qemu_get_be32(f);
1766 trace_loadvm_process_command_ping(tmp32);
1767 if (!mis->to_src_file) {
1768 error_report("CMD_PING (0x%x) received with no return path",
1772 migrate_send_rp_pong(mis, tmp32);
1775 case MIG_CMD_PACKAGED:
1776 return loadvm_handle_cmd_packaged(mis);
1778 case MIG_CMD_POSTCOPY_ADVISE:
1779 return loadvm_postcopy_handle_advise(mis);
1781 case MIG_CMD_POSTCOPY_LISTEN:
1782 return loadvm_postcopy_handle_listen(mis);
1784 case MIG_CMD_POSTCOPY_RUN:
1785 return loadvm_postcopy_handle_run(mis);
1787 case MIG_CMD_POSTCOPY_RAM_DISCARD:
1788 return loadvm_postcopy_ram_handle_discard(mis, len);
1795 * Read a footer off the wire and check that it matches the expected section
1797 * Returns: true if the footer was good
1798 * false if there is a problem (and calls error_report to say why)
1800 static bool check_section_footer(QEMUFile *f, SaveStateEntry *se)
1803 uint32_t read_section_id;
1805 if (skip_section_footers) {
1806 /* No footer to check */
1810 read_mark = qemu_get_byte(f);
1812 if (read_mark != QEMU_VM_SECTION_FOOTER) {
1813 error_report("Missing section footer for %s", se->idstr);
1817 read_section_id = qemu_get_be32(f);
1818 if (read_section_id != se->load_section_id) {
1819 error_report("Mismatched section id in footer for %s -"
1820 " read 0x%x expected 0x%x",
1821 se->idstr, read_section_id, se->load_section_id);
1830 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis)
1832 uint32_t instance_id, version_id, section_id;
1837 /* Read section start */
1838 section_id = qemu_get_be32(f);
1839 if (!qemu_get_counted_string(f, idstr)) {
1840 error_report("Unable to read ID string for section %u",
1844 instance_id = qemu_get_be32(f);
1845 version_id = qemu_get_be32(f);
1847 trace_qemu_loadvm_state_section_startfull(section_id, idstr,
1848 instance_id, version_id);
1849 /* Find savevm section */
1850 se = find_se(idstr, instance_id);
1852 error_report("Unknown savevm section or instance '%s' %d",
1853 idstr, instance_id);
1857 /* Validate version */
1858 if (version_id > se->version_id) {
1859 error_report("savevm: unsupported version %d for '%s' v%d",
1860 version_id, idstr, se->version_id);
1863 se->load_version_id = version_id;
1864 se->load_section_id = section_id;
1866 /* Validate if it is a device's state */
1867 if (xen_enabled() && se->is_ram) {
1868 error_report("loadvm: %s RAM loading not allowed on Xen", idstr);
1872 ret = vmstate_load(f, se);
1874 error_report("error while loading state for instance 0x%x of"
1875 " device '%s'", instance_id, idstr);
1878 if (!check_section_footer(f, se)) {
1886 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis)
1888 uint32_t section_id;
1892 section_id = qemu_get_be32(f);
1894 trace_qemu_loadvm_state_section_partend(section_id);
1895 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1896 if (se->load_section_id == section_id) {
1901 error_report("Unknown savevm section %d", section_id);
1905 ret = vmstate_load(f, se);
1907 error_report("error while loading state section id %d(%s)",
1908 section_id, se->idstr);
1911 if (!check_section_footer(f, se)) {
1918 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
1920 uint8_t section_type;
1923 while ((section_type = qemu_get_byte(f)) != QEMU_VM_EOF) {
1925 trace_qemu_loadvm_state_section(section_type);
1926 switch (section_type) {
1927 case QEMU_VM_SECTION_START:
1928 case QEMU_VM_SECTION_FULL:
1929 ret = qemu_loadvm_section_start_full(f, mis);
1934 case QEMU_VM_SECTION_PART:
1935 case QEMU_VM_SECTION_END:
1936 ret = qemu_loadvm_section_part_end(f, mis);
1941 case QEMU_VM_COMMAND:
1942 ret = loadvm_process_command(f);
1943 trace_qemu_loadvm_state_section_command(ret);
1944 if ((ret < 0) || (ret & LOADVM_QUIT)) {
1949 error_report("Unknown savevm section type %d", section_type);
1957 qemu_file_set_error(f, ret);
1962 int qemu_loadvm_state(QEMUFile *f)
1964 MigrationIncomingState *mis = migration_incoming_get_current();
1965 Error *local_err = NULL;
1969 if (qemu_savevm_state_blocked(&local_err)) {
1970 error_report_err(local_err);
1974 v = qemu_get_be32(f);
1975 if (v != QEMU_VM_FILE_MAGIC) {
1976 error_report("Not a migration stream");
1980 v = qemu_get_be32(f);
1981 if (v == QEMU_VM_FILE_VERSION_COMPAT) {
1982 error_report("SaveVM v2 format is obsolete and don't work anymore");
1985 if (v != QEMU_VM_FILE_VERSION) {
1986 error_report("Unsupported migration stream version");
1990 if (!savevm_state.skip_configuration || enforce_config_section()) {
1991 if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
1992 error_report("Configuration section missing");
1995 ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
2002 cpu_synchronize_all_pre_loadvm();
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 */
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()) {
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",
2037 * It doesn't seem worth failing at this point since
2038 * we apparently have an otherwise valid VM state
2041 buf = g_malloc(0x1000);
2042 size = qemu_get_be32(f);
2045 uint32_t read_chunk = MIN(size, 0x1000);
2046 qemu_get_buffer(f, buf, read_chunk);
2053 cpu_synchronize_all_post_init();
2058 int save_snapshot(const char *name, Error **errp)
2060 BlockDriverState *bs, *bs1;
2061 QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1;
2064 int saved_vm_running;
2065 uint64_t vm_state_size;
2068 AioContext *aio_context;
2070 if (!bdrv_all_can_snapshot(&bs)) {
2071 error_setg(errp, "Device '%s' is writable but does not support "
2072 "snapshots", bdrv_get_device_name(bs));
2076 /* Delete old snapshots of the same name */
2078 ret = bdrv_all_delete_snapshot(name, &bs1, errp);
2080 error_prepend(errp, "Error while deleting snapshot on device "
2081 "'%s': ", bdrv_get_device_name(bs1));
2086 bs = bdrv_all_find_vmstate_bs();
2088 error_setg(errp, "No block device can accept snapshots");
2091 aio_context = bdrv_get_aio_context(bs);
2093 saved_vm_running = runstate_is_running();
2095 ret = global_state_store();
2097 error_setg(errp, "Error saving global state");
2100 vm_stop(RUN_STATE_SAVE_VM);
2102 aio_context_acquire(aio_context);
2104 memset(sn, 0, sizeof(*sn));
2106 /* fill auxiliary fields */
2107 qemu_gettimeofday(&tv);
2108 sn->date_sec = tv.tv_sec;
2109 sn->date_nsec = tv.tv_usec * 1000;
2110 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
2113 ret = bdrv_snapshot_find(bs, old_sn, name);
2115 pstrcpy(sn->name, sizeof(sn->name), old_sn->name);
2116 pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str);
2118 pstrcpy(sn->name, sizeof(sn->name), name);
2121 /* cast below needed for OpenBSD where tv_sec is still 'long' */
2122 localtime_r((const time_t *)&tv.tv_sec, &tm);
2123 strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm);
2126 /* save the VM state */
2127 f = qemu_fopen_bdrv(bs, 1);
2129 error_setg(errp, "Could not open VM state file");
2132 ret = qemu_savevm_state(f, errp);
2133 vm_state_size = qemu_ftell(f);
2139 ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, &bs);
2141 error_setg(errp, "Error while creating snapshot on '%s'",
2142 bdrv_get_device_name(bs));
2149 aio_context_release(aio_context);
2150 if (saved_vm_running) {
2156 void qmp_xen_save_devices_state(const char *filename, Error **errp)
2159 QIOChannelFile *ioc;
2160 int saved_vm_running;
2163 saved_vm_running = runstate_is_running();
2164 vm_stop(RUN_STATE_SAVE_VM);
2165 global_state_store_running();
2167 ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT, 0660, errp);
2171 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state");
2172 f = qemu_fopen_channel_output(QIO_CHANNEL(ioc));
2173 ret = qemu_save_device_state(f);
2176 error_setg(errp, QERR_IO_ERROR);
2180 if (saved_vm_running) {
2185 void qmp_xen_load_devices_state(const char *filename, Error **errp)
2188 QIOChannelFile *ioc;
2191 /* Guest must be paused before loading the device state; the RAM state
2192 * will already have been loaded by xc
2194 if (runstate_is_running()) {
2195 error_setg(errp, "Cannot update device state while vm is running");
2198 vm_stop(RUN_STATE_RESTORE_VM);
2200 ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp);
2204 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state");
2205 f = qemu_fopen_channel_input(QIO_CHANNEL(ioc));
2207 ret = qemu_loadvm_state(f);
2210 error_setg(errp, QERR_IO_ERROR);
2212 migration_incoming_state_destroy();
2215 int load_snapshot(const char *name, Error **errp)
2217 BlockDriverState *bs, *bs_vm_state;
2218 QEMUSnapshotInfo sn;
2221 AioContext *aio_context;
2222 MigrationIncomingState *mis = migration_incoming_get_current();
2224 if (!bdrv_all_can_snapshot(&bs)) {
2226 "Device '%s' is writable but does not support snapshots",
2227 bdrv_get_device_name(bs));
2230 ret = bdrv_all_find_snapshot(name, &bs);
2233 "Device '%s' does not have the requested snapshot '%s'",
2234 bdrv_get_device_name(bs), name);
2238 bs_vm_state = bdrv_all_find_vmstate_bs();
2240 error_setg(errp, "No block device supports snapshots");
2243 aio_context = bdrv_get_aio_context(bs_vm_state);
2245 /* Don't even try to load empty VM states */
2246 aio_context_acquire(aio_context);
2247 ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
2248 aio_context_release(aio_context);
2251 } else if (sn.vm_state_size == 0) {
2252 error_setg(errp, "This is a disk-only snapshot. Revert to it "
2253 " offline using qemu-img");
2257 /* Flush all IO requests so they don't interfere with the new state. */
2260 ret = bdrv_all_goto_snapshot(name, &bs);
2262 error_setg(errp, "Error %d while activating snapshot '%s' on '%s'",
2263 ret, name, bdrv_get_device_name(bs));
2267 /* restore the VM state */
2268 f = qemu_fopen_bdrv(bs_vm_state, 0);
2270 error_setg(errp, "Could not open VM state file");
2274 qemu_system_reset(SHUTDOWN_CAUSE_NONE);
2275 mis->from_src_file = f;
2277 aio_context_acquire(aio_context);
2278 ret = qemu_loadvm_state(f);
2279 aio_context_release(aio_context);
2281 migration_incoming_state_destroy();
2283 error_setg(errp, "Error %d while loading VM state", ret);
2290 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
2292 qemu_ram_set_idstr(mr->ram_block,
2293 memory_region_name(mr), dev);
2296 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
2298 qemu_ram_unset_idstr(mr->ram_block);
2301 void vmstate_register_ram_global(MemoryRegion *mr)
2303 vmstate_register_ram(mr, NULL);
2306 bool vmstate_check_only_migratable(const VMStateDescription *vmsd)
2308 /* check needed if --only-migratable is specified */
2309 if (!only_migratable) {
2313 return !(vmsd && vmsd->unmigratable);