hmp: Add read-only-mode option to change command
[qemu.git] / migration / savevm.c
blobbe52314a12dc7244b9b295d5fb5ada928900f2c9
1 /*
2 * QEMU System Emulator
4 * Copyright (c) 2003-2008 Fabrice Bellard
5 * Copyright (c) 2009-2015 Red Hat Inc
7 * Authors:
8 * Juan Quintela <quintela@redhat.com>
10 * Permission is hereby granted, free of charge, to any person obtaining a copy
11 * of this software and associated documentation files (the "Software"), to deal
12 * in the Software without restriction, including without limitation the rights
13 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
14 * copies of the Software, and to permit persons to whom the Software is
15 * furnished to do so, subject to the following conditions:
17 * The above copyright notice and this permission notice shall be included in
18 * all copies or substantial portions of the Software.
20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
21 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
23 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
24 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
25 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 * THE SOFTWARE.
29 #include "config-host.h"
30 #include "qemu-common.h"
31 #include "hw/boards.h"
32 #include "hw/hw.h"
33 #include "hw/qdev.h"
34 #include "net/net.h"
35 #include "monitor/monitor.h"
36 #include "sysemu/sysemu.h"
37 #include "qemu/timer.h"
38 #include "audio/audio.h"
39 #include "migration/migration.h"
40 #include "migration/postcopy-ram.h"
41 #include "qapi/qmp/qerror.h"
42 #include "qemu/error-report.h"
43 #include "qemu/sockets.h"
44 #include "qemu/queue.h"
45 #include "sysemu/cpus.h"
46 #include "exec/memory.h"
47 #include "qmp-commands.h"
48 #include "trace.h"
49 #include "qemu/bitops.h"
50 #include "qemu/iov.h"
51 #include "block/snapshot.h"
52 #include "block/qapi.h"
55 #ifndef ETH_P_RARP
56 #define ETH_P_RARP 0x8035
57 #endif
58 #define ARP_HTYPE_ETH 0x0001
59 #define ARP_PTYPE_IP 0x0800
60 #define ARP_OP_REQUEST_REV 0x3
62 const unsigned int postcopy_ram_discard_version = 0;
64 static bool skip_section_footers;
66 static struct mig_cmd_args {
67 ssize_t len; /* -1 = variable */
68 const char *name;
69 } mig_cmd_args[] = {
70 [MIG_CMD_INVALID] = { .len = -1, .name = "INVALID" },
71 [MIG_CMD_OPEN_RETURN_PATH] = { .len = 0, .name = "OPEN_RETURN_PATH" },
72 [MIG_CMD_PING] = { .len = sizeof(uint32_t), .name = "PING" },
73 [MIG_CMD_POSTCOPY_ADVISE] = { .len = 16, .name = "POSTCOPY_ADVISE" },
74 [MIG_CMD_POSTCOPY_LISTEN] = { .len = 0, .name = "POSTCOPY_LISTEN" },
75 [MIG_CMD_POSTCOPY_RUN] = { .len = 0, .name = "POSTCOPY_RUN" },
76 [MIG_CMD_POSTCOPY_RAM_DISCARD] = {
77 .len = -1, .name = "POSTCOPY_RAM_DISCARD" },
78 [MIG_CMD_PACKAGED] = { .len = 4, .name = "PACKAGED" },
79 [MIG_CMD_MAX] = { .len = -1, .name = "MAX" },
82 static int announce_self_create(uint8_t *buf,
83 uint8_t *mac_addr)
85 /* Ethernet header. */
86 memset(buf, 0xff, 6); /* destination MAC addr */
87 memcpy(buf + 6, mac_addr, 6); /* source MAC addr */
88 *(uint16_t *)(buf + 12) = htons(ETH_P_RARP); /* ethertype */
90 /* RARP header. */
91 *(uint16_t *)(buf + 14) = htons(ARP_HTYPE_ETH); /* hardware addr space */
92 *(uint16_t *)(buf + 16) = htons(ARP_PTYPE_IP); /* protocol addr space */
93 *(buf + 18) = 6; /* hardware addr length (ethernet) */
94 *(buf + 19) = 4; /* protocol addr length (IPv4) */
95 *(uint16_t *)(buf + 20) = htons(ARP_OP_REQUEST_REV); /* opcode */
96 memcpy(buf + 22, mac_addr, 6); /* source hw addr */
97 memset(buf + 28, 0x00, 4); /* source protocol addr */
98 memcpy(buf + 32, mac_addr, 6); /* target hw addr */
99 memset(buf + 38, 0x00, 4); /* target protocol addr */
101 /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */
102 memset(buf + 42, 0x00, 18);
104 return 60; /* len (FCS will be added by hardware) */
107 static void qemu_announce_self_iter(NICState *nic, void *opaque)
109 uint8_t buf[60];
110 int len;
112 trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic->conf->macaddr));
113 len = announce_self_create(buf, nic->conf->macaddr.a);
115 qemu_send_packet_raw(qemu_get_queue(nic), buf, len);
119 static void qemu_announce_self_once(void *opaque)
121 static int count = SELF_ANNOUNCE_ROUNDS;
122 QEMUTimer *timer = *(QEMUTimer **)opaque;
124 qemu_foreach_nic(qemu_announce_self_iter, NULL);
126 if (--count) {
127 /* delay 50ms, 150ms, 250ms, ... */
128 timer_mod(timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) +
129 self_announce_delay(count));
130 } else {
131 timer_del(timer);
132 timer_free(timer);
136 void qemu_announce_self(void)
138 static QEMUTimer *timer;
139 timer = timer_new_ms(QEMU_CLOCK_REALTIME, qemu_announce_self_once, &timer);
140 qemu_announce_self_once(&timer);
143 /***********************************************************/
144 /* savevm/loadvm support */
146 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt,
147 int64_t pos)
149 int ret;
150 QEMUIOVector qiov;
152 qemu_iovec_init_external(&qiov, iov, iovcnt);
153 ret = bdrv_writev_vmstate(opaque, &qiov, pos);
154 if (ret < 0) {
155 return ret;
158 return qiov.size;
161 static ssize_t block_put_buffer(void *opaque, const uint8_t *buf,
162 int64_t pos, size_t size)
164 bdrv_save_vmstate(opaque, buf, pos, size);
165 return size;
168 static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos,
169 size_t size)
171 return bdrv_load_vmstate(opaque, buf, pos, size);
174 static int bdrv_fclose(void *opaque)
176 return bdrv_flush(opaque);
179 static const QEMUFileOps bdrv_read_ops = {
180 .get_buffer = block_get_buffer,
181 .close = bdrv_fclose
184 static const QEMUFileOps bdrv_write_ops = {
185 .put_buffer = block_put_buffer,
186 .writev_buffer = block_writev_buffer,
187 .close = bdrv_fclose
190 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
192 if (is_writable) {
193 return qemu_fopen_ops(bs, &bdrv_write_ops);
195 return qemu_fopen_ops(bs, &bdrv_read_ops);
199 /* QEMUFile timer support.
200 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
203 void timer_put(QEMUFile *f, QEMUTimer *ts)
205 uint64_t expire_time;
207 expire_time = timer_expire_time_ns(ts);
208 qemu_put_be64(f, expire_time);
211 void timer_get(QEMUFile *f, QEMUTimer *ts)
213 uint64_t expire_time;
215 expire_time = qemu_get_be64(f);
216 if (expire_time != -1) {
217 timer_mod_ns(ts, expire_time);
218 } else {
219 timer_del(ts);
224 /* VMState timer support.
225 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
228 static int get_timer(QEMUFile *f, void *pv, size_t size)
230 QEMUTimer *v = pv;
231 timer_get(f, v);
232 return 0;
235 static void put_timer(QEMUFile *f, void *pv, size_t size)
237 QEMUTimer *v = pv;
238 timer_put(f, v);
241 const VMStateInfo vmstate_info_timer = {
242 .name = "timer",
243 .get = get_timer,
244 .put = put_timer,
248 typedef struct CompatEntry {
249 char idstr[256];
250 int instance_id;
251 } CompatEntry;
253 typedef struct SaveStateEntry {
254 QTAILQ_ENTRY(SaveStateEntry) entry;
255 char idstr[256];
256 int instance_id;
257 int alias_id;
258 int version_id;
259 int section_id;
260 SaveVMHandlers *ops;
261 const VMStateDescription *vmsd;
262 void *opaque;
263 CompatEntry *compat;
264 int is_ram;
265 } SaveStateEntry;
267 typedef struct SaveState {
268 QTAILQ_HEAD(, SaveStateEntry) handlers;
269 int global_section_id;
270 bool skip_configuration;
271 uint32_t len;
272 const char *name;
273 } SaveState;
275 static SaveState savevm_state = {
276 .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
277 .global_section_id = 0,
278 .skip_configuration = false,
281 void savevm_skip_configuration(void)
283 savevm_state.skip_configuration = true;
287 static void configuration_pre_save(void *opaque)
289 SaveState *state = opaque;
290 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
292 state->len = strlen(current_name);
293 state->name = current_name;
296 static int configuration_post_load(void *opaque, int version_id)
298 SaveState *state = opaque;
299 const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
301 if (strncmp(state->name, current_name, state->len) != 0) {
302 error_report("Machine type received is '%s' and local is '%s'",
303 state->name, current_name);
304 return -EINVAL;
306 return 0;
309 static const VMStateDescription vmstate_configuration = {
310 .name = "configuration",
311 .version_id = 1,
312 .post_load = configuration_post_load,
313 .pre_save = configuration_pre_save,
314 .fields = (VMStateField[]) {
315 VMSTATE_UINT32(len, SaveState),
316 VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, 0, len),
317 VMSTATE_END_OF_LIST()
321 static void dump_vmstate_vmsd(FILE *out_file,
322 const VMStateDescription *vmsd, int indent,
323 bool is_subsection);
325 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
326 int indent)
328 fprintf(out_file, "%*s{\n", indent, "");
329 indent += 2;
330 fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
331 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
332 field->version_id);
333 fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
334 field->field_exists ? "true" : "false");
335 fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
336 if (field->vmsd != NULL) {
337 fprintf(out_file, ",\n");
338 dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
340 fprintf(out_file, "\n%*s}", indent - 2, "");
343 static void dump_vmstate_vmss(FILE *out_file,
344 const VMStateDescription **subsection,
345 int indent)
347 if (*subsection != NULL) {
348 dump_vmstate_vmsd(out_file, *subsection, indent, true);
352 static void dump_vmstate_vmsd(FILE *out_file,
353 const VMStateDescription *vmsd, int indent,
354 bool is_subsection)
356 if (is_subsection) {
357 fprintf(out_file, "%*s{\n", indent, "");
358 } else {
359 fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
361 indent += 2;
362 fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
363 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
364 vmsd->version_id);
365 fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
366 vmsd->minimum_version_id);
367 if (vmsd->fields != NULL) {
368 const VMStateField *field = vmsd->fields;
369 bool first;
371 fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
372 first = true;
373 while (field->name != NULL) {
374 if (field->flags & VMS_MUST_EXIST) {
375 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
376 field++;
377 continue;
379 if (!first) {
380 fprintf(out_file, ",\n");
382 dump_vmstate_vmsf(out_file, field, indent + 2);
383 field++;
384 first = false;
386 fprintf(out_file, "\n%*s]", indent, "");
388 if (vmsd->subsections != NULL) {
389 const VMStateDescription **subsection = vmsd->subsections;
390 bool first;
392 fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
393 first = true;
394 while (*subsection != NULL) {
395 if (!first) {
396 fprintf(out_file, ",\n");
398 dump_vmstate_vmss(out_file, subsection, indent + 2);
399 subsection++;
400 first = false;
402 fprintf(out_file, "\n%*s]", indent, "");
404 fprintf(out_file, "\n%*s}", indent - 2, "");
407 static void dump_machine_type(FILE *out_file)
409 MachineClass *mc;
411 mc = MACHINE_GET_CLASS(current_machine);
413 fprintf(out_file, " \"vmschkmachine\": {\n");
414 fprintf(out_file, " \"Name\": \"%s\"\n", mc->name);
415 fprintf(out_file, " },\n");
418 void dump_vmstate_json_to_file(FILE *out_file)
420 GSList *list, *elt;
421 bool first;
423 fprintf(out_file, "{\n");
424 dump_machine_type(out_file);
426 first = true;
427 list = object_class_get_list(TYPE_DEVICE, true);
428 for (elt = list; elt; elt = elt->next) {
429 DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
430 TYPE_DEVICE);
431 const char *name;
432 int indent = 2;
434 if (!dc->vmsd) {
435 continue;
438 if (!first) {
439 fprintf(out_file, ",\n");
441 name = object_class_get_name(OBJECT_CLASS(dc));
442 fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
443 indent += 2;
444 fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
445 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
446 dc->vmsd->version_id);
447 fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
448 dc->vmsd->minimum_version_id);
450 dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
452 fprintf(out_file, "\n%*s}", indent - 2, "");
453 first = false;
455 fprintf(out_file, "\n}\n");
456 fclose(out_file);
459 static int calculate_new_instance_id(const char *idstr)
461 SaveStateEntry *se;
462 int instance_id = 0;
464 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
465 if (strcmp(idstr, se->idstr) == 0
466 && instance_id <= se->instance_id) {
467 instance_id = se->instance_id + 1;
470 return instance_id;
473 static int calculate_compat_instance_id(const char *idstr)
475 SaveStateEntry *se;
476 int instance_id = 0;
478 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
479 if (!se->compat) {
480 continue;
483 if (strcmp(idstr, se->compat->idstr) == 0
484 && instance_id <= se->compat->instance_id) {
485 instance_id = se->compat->instance_id + 1;
488 return instance_id;
491 /* TODO: Individual devices generally have very little idea about the rest
492 of the system, so instance_id should be removed/replaced.
493 Meanwhile pass -1 as instance_id if you do not already have a clearly
494 distinguishing id for all instances of your device class. */
495 int register_savevm_live(DeviceState *dev,
496 const char *idstr,
497 int instance_id,
498 int version_id,
499 SaveVMHandlers *ops,
500 void *opaque)
502 SaveStateEntry *se;
504 se = g_new0(SaveStateEntry, 1);
505 se->version_id = version_id;
506 se->section_id = savevm_state.global_section_id++;
507 se->ops = ops;
508 se->opaque = opaque;
509 se->vmsd = NULL;
510 /* if this is a live_savem then set is_ram */
511 if (ops->save_live_setup != NULL) {
512 se->is_ram = 1;
515 if (dev) {
516 char *id = qdev_get_dev_path(dev);
517 if (id) {
518 pstrcpy(se->idstr, sizeof(se->idstr), id);
519 pstrcat(se->idstr, sizeof(se->idstr), "/");
520 g_free(id);
522 se->compat = g_new0(CompatEntry, 1);
523 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), idstr);
524 se->compat->instance_id = instance_id == -1 ?
525 calculate_compat_instance_id(idstr) : instance_id;
526 instance_id = -1;
529 pstrcat(se->idstr, sizeof(se->idstr), idstr);
531 if (instance_id == -1) {
532 se->instance_id = calculate_new_instance_id(se->idstr);
533 } else {
534 se->instance_id = instance_id;
536 assert(!se->compat || se->instance_id == 0);
537 /* add at the end of list */
538 QTAILQ_INSERT_TAIL(&savevm_state.handlers, se, entry);
539 return 0;
542 int register_savevm(DeviceState *dev,
543 const char *idstr,
544 int instance_id,
545 int version_id,
546 SaveStateHandler *save_state,
547 LoadStateHandler *load_state,
548 void *opaque)
550 SaveVMHandlers *ops = g_new0(SaveVMHandlers, 1);
551 ops->save_state = save_state;
552 ops->load_state = load_state;
553 return register_savevm_live(dev, idstr, instance_id, version_id,
554 ops, opaque);
557 void unregister_savevm(DeviceState *dev, const char *idstr, void *opaque)
559 SaveStateEntry *se, *new_se;
560 char id[256] = "";
562 if (dev) {
563 char *path = qdev_get_dev_path(dev);
564 if (path) {
565 pstrcpy(id, sizeof(id), path);
566 pstrcat(id, sizeof(id), "/");
567 g_free(path);
570 pstrcat(id, sizeof(id), idstr);
572 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
573 if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
574 QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
575 g_free(se->compat);
576 g_free(se->ops);
577 g_free(se);
582 int vmstate_register_with_alias_id(DeviceState *dev, int instance_id,
583 const VMStateDescription *vmsd,
584 void *opaque, int alias_id,
585 int required_for_version)
587 SaveStateEntry *se;
589 /* If this triggers, alias support can be dropped for the vmsd. */
590 assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
592 se = g_new0(SaveStateEntry, 1);
593 se->version_id = vmsd->version_id;
594 se->section_id = savevm_state.global_section_id++;
595 se->opaque = opaque;
596 se->vmsd = vmsd;
597 se->alias_id = alias_id;
599 if (dev) {
600 char *id = qdev_get_dev_path(dev);
601 if (id) {
602 pstrcpy(se->idstr, sizeof(se->idstr), id);
603 pstrcat(se->idstr, sizeof(se->idstr), "/");
604 g_free(id);
606 se->compat = g_new0(CompatEntry, 1);
607 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
608 se->compat->instance_id = instance_id == -1 ?
609 calculate_compat_instance_id(vmsd->name) : instance_id;
610 instance_id = -1;
613 pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
615 if (instance_id == -1) {
616 se->instance_id = calculate_new_instance_id(se->idstr);
617 } else {
618 se->instance_id = instance_id;
620 assert(!se->compat || se->instance_id == 0);
621 /* add at the end of list */
622 QTAILQ_INSERT_TAIL(&savevm_state.handlers, se, entry);
623 return 0;
626 void vmstate_unregister(DeviceState *dev, const VMStateDescription *vmsd,
627 void *opaque)
629 SaveStateEntry *se, *new_se;
631 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
632 if (se->vmsd == vmsd && se->opaque == opaque) {
633 QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
634 g_free(se->compat);
635 g_free(se);
640 static int vmstate_load(QEMUFile *f, SaveStateEntry *se, int version_id)
642 trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
643 if (!se->vmsd) { /* Old style */
644 return se->ops->load_state(f, se->opaque, version_id);
646 return vmstate_load_state(f, se->vmsd, se->opaque, version_id);
649 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
651 int64_t old_offset, size;
653 old_offset = qemu_ftell_fast(f);
654 se->ops->save_state(f, se->opaque);
655 size = qemu_ftell_fast(f) - old_offset;
657 if (vmdesc) {
658 json_prop_int(vmdesc, "size", size);
659 json_start_array(vmdesc, "fields");
660 json_start_object(vmdesc, NULL);
661 json_prop_str(vmdesc, "name", "data");
662 json_prop_int(vmdesc, "size", size);
663 json_prop_str(vmdesc, "type", "buffer");
664 json_end_object(vmdesc);
665 json_end_array(vmdesc);
669 static void vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
671 trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
672 if (!se->vmsd) {
673 vmstate_save_old_style(f, se, vmdesc);
674 return;
676 vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
679 void savevm_skip_section_footers(void)
681 skip_section_footers = true;
685 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
687 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
688 uint8_t section_type)
690 qemu_put_byte(f, section_type);
691 qemu_put_be32(f, se->section_id);
693 if (section_type == QEMU_VM_SECTION_FULL ||
694 section_type == QEMU_VM_SECTION_START) {
695 /* ID string */
696 size_t len = strlen(se->idstr);
697 qemu_put_byte(f, len);
698 qemu_put_buffer(f, (uint8_t *)se->idstr, len);
700 qemu_put_be32(f, se->instance_id);
701 qemu_put_be32(f, se->version_id);
706 * Write a footer onto device sections that catches cases misformatted device
707 * sections.
709 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
711 if (!skip_section_footers) {
712 qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
713 qemu_put_be32(f, se->section_id);
718 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
719 * command and associated data.
721 * @f: File to send command on
722 * @command: Command type to send
723 * @len: Length of associated data
724 * @data: Data associated with command.
726 void qemu_savevm_command_send(QEMUFile *f,
727 enum qemu_vm_cmd command,
728 uint16_t len,
729 uint8_t *data)
731 trace_savevm_command_send(command, len);
732 qemu_put_byte(f, QEMU_VM_COMMAND);
733 qemu_put_be16(f, (uint16_t)command);
734 qemu_put_be16(f, len);
735 qemu_put_buffer(f, data, len);
736 qemu_fflush(f);
739 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value)
741 uint32_t buf;
743 trace_savevm_send_ping(value);
744 buf = cpu_to_be32(value);
745 qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf);
748 void qemu_savevm_send_open_return_path(QEMUFile *f)
750 trace_savevm_send_open_return_path();
751 qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL);
754 /* We have a buffer of data to send; we don't want that all to be loaded
755 * by the command itself, so the command contains just the length of the
756 * extra buffer that we then send straight after it.
757 * TODO: Must be a better way to organise that
759 * Returns:
760 * 0 on success
761 * -ve on error
763 int qemu_savevm_send_packaged(QEMUFile *f, const QEMUSizedBuffer *qsb)
765 size_t cur_iov;
766 size_t len = qsb_get_length(qsb);
767 uint32_t tmp;
769 if (len > MAX_VM_CMD_PACKAGED_SIZE) {
770 error_report("%s: Unreasonably large packaged state: %zu",
771 __func__, len);
772 return -1;
775 tmp = cpu_to_be32(len);
777 trace_qemu_savevm_send_packaged();
778 qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp);
780 /* all the data follows (concatinating the iov's) */
781 for (cur_iov = 0; cur_iov < qsb->n_iov; cur_iov++) {
782 /* The iov entries are partially filled */
783 size_t towrite = MIN(qsb->iov[cur_iov].iov_len, len);
784 len -= towrite;
786 if (!towrite) {
787 break;
790 qemu_put_buffer(f, qsb->iov[cur_iov].iov_base, towrite);
793 return 0;
796 /* Send prior to any postcopy transfer */
797 void qemu_savevm_send_postcopy_advise(QEMUFile *f)
799 uint64_t tmp[2];
800 tmp[0] = cpu_to_be64(getpagesize());
801 tmp[1] = cpu_to_be64(1ul << qemu_target_page_bits());
803 trace_qemu_savevm_send_postcopy_advise();
804 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 16, (uint8_t *)tmp);
807 /* Sent prior to starting the destination running in postcopy, discard pages
808 * that have already been sent but redirtied on the source.
809 * CMD_POSTCOPY_RAM_DISCARD consist of:
810 * byte version (0)
811 * byte Length of name field (not including 0)
812 * n x byte RAM block name
813 * byte 0 terminator (just for safety)
814 * n x Byte ranges within the named RAMBlock
815 * be64 Start of the range
816 * be64 Length
818 * name: RAMBlock name that these entries are part of
819 * len: Number of page entries
820 * start_list: 'len' addresses
821 * length_list: 'len' addresses
824 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
825 uint16_t len,
826 uint64_t *start_list,
827 uint64_t *length_list)
829 uint8_t *buf;
830 uint16_t tmplen;
831 uint16_t t;
832 size_t name_len = strlen(name);
834 trace_qemu_savevm_send_postcopy_ram_discard(name, len);
835 assert(name_len < 256);
836 buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len);
837 buf[0] = postcopy_ram_discard_version;
838 buf[1] = name_len;
839 memcpy(buf + 2, name, name_len);
840 tmplen = 2 + name_len;
841 buf[tmplen++] = '\0';
843 for (t = 0; t < len; t++) {
844 cpu_to_be64w((uint64_t *)(buf + tmplen), start_list[t]);
845 tmplen += 8;
846 cpu_to_be64w((uint64_t *)(buf + tmplen), length_list[t]);
847 tmplen += 8;
849 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf);
850 g_free(buf);
853 /* Get the destination into a state where it can receive postcopy data. */
854 void qemu_savevm_send_postcopy_listen(QEMUFile *f)
856 trace_savevm_send_postcopy_listen();
857 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL);
860 /* Kick the destination into running */
861 void qemu_savevm_send_postcopy_run(QEMUFile *f)
863 trace_savevm_send_postcopy_run();
864 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL);
867 bool qemu_savevm_state_blocked(Error **errp)
869 SaveStateEntry *se;
871 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
872 if (se->vmsd && se->vmsd->unmigratable) {
873 error_setg(errp, "State blocked by non-migratable device '%s'",
874 se->idstr);
875 return true;
878 return false;
881 void qemu_savevm_state_header(QEMUFile *f)
883 trace_savevm_state_header();
884 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
885 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
887 if (!savevm_state.skip_configuration) {
888 qemu_put_byte(f, QEMU_VM_CONFIGURATION);
889 vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
894 void qemu_savevm_state_begin(QEMUFile *f,
895 const MigrationParams *params)
897 SaveStateEntry *se;
898 int ret;
900 trace_savevm_state_begin();
901 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
902 if (!se->ops || !se->ops->set_params) {
903 continue;
905 se->ops->set_params(params, se->opaque);
908 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
909 if (!se->ops || !se->ops->save_live_setup) {
910 continue;
912 if (se->ops && se->ops->is_active) {
913 if (!se->ops->is_active(se->opaque)) {
914 continue;
917 save_section_header(f, se, QEMU_VM_SECTION_START);
919 ret = se->ops->save_live_setup(f, se->opaque);
920 save_section_footer(f, se);
921 if (ret < 0) {
922 qemu_file_set_error(f, ret);
923 break;
929 * this function has three return values:
930 * negative: there was one error, and we have -errno.
931 * 0 : We haven't finished, caller have to go again
932 * 1 : We have finished, we can go to complete phase
934 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
936 SaveStateEntry *se;
937 int ret = 1;
939 trace_savevm_state_iterate();
940 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
941 if (!se->ops || !se->ops->save_live_iterate) {
942 continue;
944 if (se->ops && se->ops->is_active) {
945 if (!se->ops->is_active(se->opaque)) {
946 continue;
950 * In the postcopy phase, any device that doesn't know how to
951 * do postcopy should have saved it's state in the _complete
952 * call that's already run, it might get confused if we call
953 * iterate afterwards.
955 if (postcopy && !se->ops->save_live_complete_postcopy) {
956 continue;
958 if (qemu_file_rate_limit(f)) {
959 return 0;
961 trace_savevm_section_start(se->idstr, se->section_id);
963 save_section_header(f, se, QEMU_VM_SECTION_PART);
965 ret = se->ops->save_live_iterate(f, se->opaque);
966 trace_savevm_section_end(se->idstr, se->section_id, ret);
967 save_section_footer(f, se);
969 if (ret < 0) {
970 qemu_file_set_error(f, ret);
972 if (ret <= 0) {
973 /* Do not proceed to the next vmstate before this one reported
974 completion of the current stage. This serializes the migration
975 and reduces the probability that a faster changing state is
976 synchronized over and over again. */
977 break;
980 return ret;
983 static bool should_send_vmdesc(void)
985 MachineState *machine = MACHINE(qdev_get_machine());
986 bool in_postcopy = migration_in_postcopy(migrate_get_current());
987 return !machine->suppress_vmdesc && !in_postcopy;
991 * Calls the save_live_complete_postcopy methods
992 * causing the last few pages to be sent immediately and doing any associated
993 * cleanup.
994 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
995 * all the other devices, but that happens at the point we switch to postcopy.
997 void qemu_savevm_state_complete_postcopy(QEMUFile *f)
999 SaveStateEntry *se;
1000 int ret;
1002 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1003 if (!se->ops || !se->ops->save_live_complete_postcopy) {
1004 continue;
1006 if (se->ops && se->ops->is_active) {
1007 if (!se->ops->is_active(se->opaque)) {
1008 continue;
1011 trace_savevm_section_start(se->idstr, se->section_id);
1012 /* Section type */
1013 qemu_put_byte(f, QEMU_VM_SECTION_END);
1014 qemu_put_be32(f, se->section_id);
1016 ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1017 trace_savevm_section_end(se->idstr, se->section_id, ret);
1018 save_section_footer(f, se);
1019 if (ret < 0) {
1020 qemu_file_set_error(f, ret);
1021 return;
1025 qemu_put_byte(f, QEMU_VM_EOF);
1026 qemu_fflush(f);
1029 void qemu_savevm_state_complete_precopy(QEMUFile *f)
1031 QJSON *vmdesc;
1032 int vmdesc_len;
1033 SaveStateEntry *se;
1034 int ret;
1035 bool in_postcopy = migration_in_postcopy(migrate_get_current());
1037 trace_savevm_state_complete_precopy();
1039 cpu_synchronize_all_states();
1041 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1042 if (!se->ops ||
1043 (in_postcopy && se->ops->save_live_complete_postcopy) ||
1044 !se->ops->save_live_complete_precopy) {
1045 continue;
1047 if (se->ops && se->ops->is_active) {
1048 if (!se->ops->is_active(se->opaque)) {
1049 continue;
1052 trace_savevm_section_start(se->idstr, se->section_id);
1054 save_section_header(f, se, QEMU_VM_SECTION_END);
1056 ret = se->ops->save_live_complete_precopy(f, se->opaque);
1057 trace_savevm_section_end(se->idstr, se->section_id, ret);
1058 save_section_footer(f, se);
1059 if (ret < 0) {
1060 qemu_file_set_error(f, ret);
1061 return;
1065 vmdesc = qjson_new();
1066 json_prop_int(vmdesc, "page_size", TARGET_PAGE_SIZE);
1067 json_start_array(vmdesc, "devices");
1068 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1070 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1071 continue;
1073 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1074 trace_savevm_section_skip(se->idstr, se->section_id);
1075 continue;
1078 trace_savevm_section_start(se->idstr, se->section_id);
1080 json_start_object(vmdesc, NULL);
1081 json_prop_str(vmdesc, "name", se->idstr);
1082 json_prop_int(vmdesc, "instance_id", se->instance_id);
1084 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1086 vmstate_save(f, se, vmdesc);
1088 json_end_object(vmdesc);
1089 trace_savevm_section_end(se->idstr, se->section_id, 0);
1090 save_section_footer(f, se);
1093 if (!in_postcopy) {
1094 /* Postcopy stream will still be going */
1095 qemu_put_byte(f, QEMU_VM_EOF);
1098 json_end_array(vmdesc);
1099 qjson_finish(vmdesc);
1100 vmdesc_len = strlen(qjson_get_str(vmdesc));
1102 if (should_send_vmdesc()) {
1103 qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1104 qemu_put_be32(f, vmdesc_len);
1105 qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len);
1107 object_unref(OBJECT(vmdesc));
1109 qemu_fflush(f);
1112 /* Give an estimate of the amount left to be transferred,
1113 * the result is split into the amount for units that can and
1114 * for units that can't do postcopy.
1116 void qemu_savevm_state_pending(QEMUFile *f, uint64_t max_size,
1117 uint64_t *res_non_postcopiable,
1118 uint64_t *res_postcopiable)
1120 SaveStateEntry *se;
1122 *res_non_postcopiable = 0;
1123 *res_postcopiable = 0;
1126 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1127 if (!se->ops || !se->ops->save_live_pending) {
1128 continue;
1130 if (se->ops && se->ops->is_active) {
1131 if (!se->ops->is_active(se->opaque)) {
1132 continue;
1135 se->ops->save_live_pending(f, se->opaque, max_size,
1136 res_non_postcopiable, res_postcopiable);
1140 void qemu_savevm_state_cleanup(void)
1142 SaveStateEntry *se;
1144 trace_savevm_state_cleanup();
1145 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1146 if (se->ops && se->ops->cleanup) {
1147 se->ops->cleanup(se->opaque);
1152 static int qemu_savevm_state(QEMUFile *f, Error **errp)
1154 int ret;
1155 MigrationParams params = {
1156 .blk = 0,
1157 .shared = 0
1159 MigrationState *ms = migrate_init(&params);
1160 ms->file = f;
1162 if (qemu_savevm_state_blocked(errp)) {
1163 return -EINVAL;
1166 qemu_mutex_unlock_iothread();
1167 qemu_savevm_state_header(f);
1168 qemu_savevm_state_begin(f, &params);
1169 qemu_mutex_lock_iothread();
1171 while (qemu_file_get_error(f) == 0) {
1172 if (qemu_savevm_state_iterate(f, false) > 0) {
1173 break;
1177 ret = qemu_file_get_error(f);
1178 if (ret == 0) {
1179 qemu_savevm_state_complete_precopy(f);
1180 ret = qemu_file_get_error(f);
1182 qemu_savevm_state_cleanup();
1183 if (ret != 0) {
1184 error_setg_errno(errp, -ret, "Error while writing VM state");
1186 return ret;
1189 static int qemu_save_device_state(QEMUFile *f)
1191 SaveStateEntry *se;
1193 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1194 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1196 cpu_synchronize_all_states();
1198 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1199 if (se->is_ram) {
1200 continue;
1202 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1203 continue;
1205 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1206 continue;
1209 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1211 vmstate_save(f, se, NULL);
1213 save_section_footer(f, se);
1216 qemu_put_byte(f, QEMU_VM_EOF);
1218 return qemu_file_get_error(f);
1221 static SaveStateEntry *find_se(const char *idstr, int instance_id)
1223 SaveStateEntry *se;
1225 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1226 if (!strcmp(se->idstr, idstr) &&
1227 (instance_id == se->instance_id ||
1228 instance_id == se->alias_id))
1229 return se;
1230 /* Migrating from an older version? */
1231 if (strstr(se->idstr, idstr) && se->compat) {
1232 if (!strcmp(se->compat->idstr, idstr) &&
1233 (instance_id == se->compat->instance_id ||
1234 instance_id == se->alias_id))
1235 return se;
1238 return NULL;
1241 enum LoadVMExitCodes {
1242 /* Allow a command to quit all layers of nested loadvm loops */
1243 LOADVM_QUIT = 1,
1246 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis);
1248 /* ------ incoming postcopy messages ------ */
1249 /* 'advise' arrives before any transfers just to tell us that a postcopy
1250 * *might* happen - it might be skipped if precopy transferred everything
1251 * quickly.
1253 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis)
1255 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1256 uint64_t remote_hps, remote_tps;
1258 trace_loadvm_postcopy_handle_advise();
1259 if (ps != POSTCOPY_INCOMING_NONE) {
1260 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1261 return -1;
1264 if (!postcopy_ram_supported_by_host()) {
1265 return -1;
1268 remote_hps = qemu_get_be64(mis->from_src_file);
1269 if (remote_hps != getpagesize()) {
1271 * Some combinations of mismatch are probably possible but it gets
1272 * a bit more complicated. In particular we need to place whole
1273 * host pages on the dest at once, and we need to ensure that we
1274 * handle dirtying to make sure we never end up sending part of
1275 * a hostpage on it's own.
1277 error_report("Postcopy needs matching host page sizes (s=%d d=%d)",
1278 (int)remote_hps, getpagesize());
1279 return -1;
1282 remote_tps = qemu_get_be64(mis->from_src_file);
1283 if (remote_tps != (1ul << qemu_target_page_bits())) {
1285 * Again, some differences could be dealt with, but for now keep it
1286 * simple.
1288 error_report("Postcopy needs matching target page sizes (s=%d d=%d)",
1289 (int)remote_tps, 1 << qemu_target_page_bits());
1290 return -1;
1293 if (ram_postcopy_incoming_init(mis)) {
1294 return -1;
1297 postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1299 return 0;
1302 /* After postcopy we will be told to throw some pages away since they're
1303 * dirty and will have to be demand fetched. Must happen before CPU is
1304 * started.
1305 * There can be 0..many of these messages, each encoding multiple pages.
1307 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1308 uint16_t len)
1310 int tmp;
1311 char ramid[256];
1312 PostcopyState ps = postcopy_state_get();
1314 trace_loadvm_postcopy_ram_handle_discard();
1316 switch (ps) {
1317 case POSTCOPY_INCOMING_ADVISE:
1318 /* 1st discard */
1319 tmp = postcopy_ram_prepare_discard(mis);
1320 if (tmp) {
1321 return tmp;
1323 break;
1325 case POSTCOPY_INCOMING_DISCARD:
1326 /* Expected state */
1327 break;
1329 default:
1330 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1331 ps);
1332 return -1;
1334 /* We're expecting a
1335 * Version (0)
1336 * a RAM ID string (length byte, name, 0 term)
1337 * then at least 1 16 byte chunk
1339 if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1340 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1341 return -1;
1344 tmp = qemu_get_byte(mis->from_src_file);
1345 if (tmp != postcopy_ram_discard_version) {
1346 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1347 return -1;
1350 if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1351 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1352 return -1;
1354 tmp = qemu_get_byte(mis->from_src_file);
1355 if (tmp != 0) {
1356 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1357 return -1;
1360 len -= 3 + strlen(ramid);
1361 if (len % 16) {
1362 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1363 return -1;
1365 trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1366 while (len) {
1367 uint64_t start_addr, block_length;
1368 start_addr = qemu_get_be64(mis->from_src_file);
1369 block_length = qemu_get_be64(mis->from_src_file);
1371 len -= 16;
1372 int ret = ram_discard_range(mis, ramid, start_addr,
1373 block_length);
1374 if (ret) {
1375 return ret;
1378 trace_loadvm_postcopy_ram_handle_discard_end();
1380 return 0;
1384 * Triggered by a postcopy_listen command; this thread takes over reading
1385 * the input stream, leaving the main thread free to carry on loading the rest
1386 * of the device state (from RAM).
1387 * (TODO:This could do with being in a postcopy file - but there again it's
1388 * just another input loop, not that postcopy specific)
1390 static void *postcopy_ram_listen_thread(void *opaque)
1392 QEMUFile *f = opaque;
1393 MigrationIncomingState *mis = migration_incoming_get_current();
1394 int load_res;
1396 qemu_sem_post(&mis->listen_thread_sem);
1397 trace_postcopy_ram_listen_thread_start();
1400 * Because we're a thread and not a coroutine we can't yield
1401 * in qemu_file, and thus we must be blocking now.
1403 qemu_file_set_blocking(f, true);
1404 load_res = qemu_loadvm_state_main(f, mis);
1405 /* And non-blocking again so we don't block in any cleanup */
1406 qemu_file_set_blocking(f, false);
1408 trace_postcopy_ram_listen_thread_exit();
1409 if (load_res < 0) {
1410 error_report("%s: loadvm failed: %d", __func__, load_res);
1411 qemu_file_set_error(f, load_res);
1412 } else {
1414 * This looks good, but it's possible that the device loading in the
1415 * main thread hasn't finished yet, and so we might not be in 'RUN'
1416 * state yet; wait for the end of the main thread.
1418 qemu_event_wait(&mis->main_thread_load_event);
1420 postcopy_ram_incoming_cleanup(mis);
1422 * If everything has worked fine, then the main thread has waited
1423 * for us to start, and we're the last use of the mis.
1424 * (If something broke then qemu will have to exit anyway since it's
1425 * got a bad migration state).
1427 migration_incoming_state_destroy();
1429 if (load_res < 0) {
1431 * If something went wrong then we have a bad state so exit;
1432 * depending how far we got it might be possible at this point
1433 * to leave the guest running and fire MCEs for pages that never
1434 * arrived as a desperate recovery step.
1436 exit(EXIT_FAILURE);
1439 return NULL;
1442 /* After this message we must be able to immediately receive postcopy data */
1443 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
1445 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
1446 trace_loadvm_postcopy_handle_listen();
1447 if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
1448 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
1449 return -1;
1451 if (ps == POSTCOPY_INCOMING_ADVISE) {
1453 * A rare case, we entered listen without having to do any discards,
1454 * so do the setup that's normally done at the time of the 1st discard.
1456 postcopy_ram_prepare_discard(mis);
1460 * Sensitise RAM - can now generate requests for blocks that don't exist
1461 * However, at this point the CPU shouldn't be running, and the IO
1462 * shouldn't be doing anything yet so don't actually expect requests
1464 if (postcopy_ram_enable_notify(mis)) {
1465 return -1;
1468 if (mis->have_listen_thread) {
1469 error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread");
1470 return -1;
1473 mis->have_listen_thread = true;
1474 /* Start up the listening thread and wait for it to signal ready */
1475 qemu_sem_init(&mis->listen_thread_sem, 0);
1476 qemu_thread_create(&mis->listen_thread, "postcopy/listen",
1477 postcopy_ram_listen_thread, mis->from_src_file,
1478 QEMU_THREAD_JOINABLE);
1479 qemu_sem_wait(&mis->listen_thread_sem);
1480 qemu_sem_destroy(&mis->listen_thread_sem);
1482 return 0;
1485 /* After all discards we can start running and asking for pages */
1486 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
1488 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
1489 Error *local_err = NULL;
1491 trace_loadvm_postcopy_handle_run();
1492 if (ps != POSTCOPY_INCOMING_LISTENING) {
1493 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
1494 return -1;
1497 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1498 * in migration.c
1500 cpu_synchronize_all_post_init();
1502 qemu_announce_self();
1504 /* Make sure all file formats flush their mutable metadata */
1505 bdrv_invalidate_cache_all(&local_err);
1506 if (local_err) {
1507 error_report_err(local_err);
1508 return -1;
1511 trace_loadvm_postcopy_handle_run_cpu_sync();
1512 cpu_synchronize_all_post_init();
1514 trace_loadvm_postcopy_handle_run_vmstart();
1516 if (autostart) {
1517 /* Hold onto your hats, starting the CPU */
1518 vm_start();
1519 } else {
1520 /* leave it paused and let management decide when to start the CPU */
1521 runstate_set(RUN_STATE_PAUSED);
1524 /* We need to finish reading the stream from the package
1525 * and also stop reading anything more from the stream that loaded the
1526 * package (since it's now being read by the listener thread).
1527 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
1529 return LOADVM_QUIT;
1533 * Immediately following this command is a blob of data containing an embedded
1534 * chunk of migration stream; read it and load it.
1536 * @mis: Incoming state
1537 * @length: Length of packaged data to read
1539 * Returns: Negative values on error
1542 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
1544 int ret;
1545 uint8_t *buffer;
1546 uint32_t length;
1547 QEMUSizedBuffer *qsb;
1549 length = qemu_get_be32(mis->from_src_file);
1550 trace_loadvm_handle_cmd_packaged(length);
1552 if (length > MAX_VM_CMD_PACKAGED_SIZE) {
1553 error_report("Unreasonably large packaged state: %u", length);
1554 return -1;
1556 buffer = g_malloc0(length);
1557 ret = qemu_get_buffer(mis->from_src_file, buffer, (int)length);
1558 if (ret != length) {
1559 g_free(buffer);
1560 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%d\n",
1561 ret, length);
1562 return (ret < 0) ? ret : -EAGAIN;
1564 trace_loadvm_handle_cmd_packaged_received(ret);
1566 /* Setup a dummy QEMUFile that actually reads from the buffer */
1567 qsb = qsb_create(buffer, length);
1568 g_free(buffer); /* Because qsb_create copies */
1569 if (!qsb) {
1570 error_report("Unable to create qsb");
1572 QEMUFile *packf = qemu_bufopen("r", qsb);
1574 ret = qemu_loadvm_state_main(packf, mis);
1575 trace_loadvm_handle_cmd_packaged_main(ret);
1576 qemu_fclose(packf);
1577 qsb_free(qsb);
1579 return ret;
1583 * Process an incoming 'QEMU_VM_COMMAND'
1584 * 0 just a normal return
1585 * LOADVM_QUIT All good, but exit the loop
1586 * <0 Error
1588 static int loadvm_process_command(QEMUFile *f)
1590 MigrationIncomingState *mis = migration_incoming_get_current();
1591 uint16_t cmd;
1592 uint16_t len;
1593 uint32_t tmp32;
1595 cmd = qemu_get_be16(f);
1596 len = qemu_get_be16(f);
1598 trace_loadvm_process_command(cmd, len);
1599 if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
1600 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
1601 return -EINVAL;
1604 if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
1605 error_report("%s received with bad length - expecting %zu, got %d",
1606 mig_cmd_args[cmd].name,
1607 (size_t)mig_cmd_args[cmd].len, len);
1608 return -ERANGE;
1611 switch (cmd) {
1612 case MIG_CMD_OPEN_RETURN_PATH:
1613 if (mis->to_src_file) {
1614 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
1615 /* Not really a problem, so don't give up */
1616 return 0;
1618 mis->to_src_file = qemu_file_get_return_path(f);
1619 if (!mis->to_src_file) {
1620 error_report("CMD_OPEN_RETURN_PATH failed");
1621 return -1;
1623 break;
1625 case MIG_CMD_PING:
1626 tmp32 = qemu_get_be32(f);
1627 trace_loadvm_process_command_ping(tmp32);
1628 if (!mis->to_src_file) {
1629 error_report("CMD_PING (0x%x) received with no return path",
1630 tmp32);
1631 return -1;
1633 migrate_send_rp_pong(mis, tmp32);
1634 break;
1636 case MIG_CMD_PACKAGED:
1637 return loadvm_handle_cmd_packaged(mis);
1639 case MIG_CMD_POSTCOPY_ADVISE:
1640 return loadvm_postcopy_handle_advise(mis);
1642 case MIG_CMD_POSTCOPY_LISTEN:
1643 return loadvm_postcopy_handle_listen(mis);
1645 case MIG_CMD_POSTCOPY_RUN:
1646 return loadvm_postcopy_handle_run(mis);
1648 case MIG_CMD_POSTCOPY_RAM_DISCARD:
1649 return loadvm_postcopy_ram_handle_discard(mis, len);
1652 return 0;
1655 struct LoadStateEntry {
1656 QLIST_ENTRY(LoadStateEntry) entry;
1657 SaveStateEntry *se;
1658 int section_id;
1659 int version_id;
1663 * Read a footer off the wire and check that it matches the expected section
1665 * Returns: true if the footer was good
1666 * false if there is a problem (and calls error_report to say why)
1668 static bool check_section_footer(QEMUFile *f, LoadStateEntry *le)
1670 uint8_t read_mark;
1671 uint32_t read_section_id;
1673 if (skip_section_footers) {
1674 /* No footer to check */
1675 return true;
1678 read_mark = qemu_get_byte(f);
1680 if (read_mark != QEMU_VM_SECTION_FOOTER) {
1681 error_report("Missing section footer for %s", le->se->idstr);
1682 return false;
1685 read_section_id = qemu_get_be32(f);
1686 if (read_section_id != le->section_id) {
1687 error_report("Mismatched section id in footer for %s -"
1688 " read 0x%x expected 0x%x",
1689 le->se->idstr, read_section_id, le->section_id);
1690 return false;
1693 /* All good */
1694 return true;
1697 void loadvm_free_handlers(MigrationIncomingState *mis)
1699 LoadStateEntry *le, *new_le;
1701 QLIST_FOREACH_SAFE(le, &mis->loadvm_handlers, entry, new_le) {
1702 QLIST_REMOVE(le, entry);
1703 g_free(le);
1707 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
1709 uint8_t section_type;
1710 int ret;
1712 while ((section_type = qemu_get_byte(f)) != QEMU_VM_EOF) {
1713 uint32_t instance_id, version_id, section_id;
1714 SaveStateEntry *se;
1715 LoadStateEntry *le;
1716 char idstr[256];
1718 trace_qemu_loadvm_state_section(section_type);
1719 switch (section_type) {
1720 case QEMU_VM_SECTION_START:
1721 case QEMU_VM_SECTION_FULL:
1722 /* Read section start */
1723 section_id = qemu_get_be32(f);
1724 if (!qemu_get_counted_string(f, idstr)) {
1725 error_report("Unable to read ID string for section %u",
1726 section_id);
1727 return -EINVAL;
1729 instance_id = qemu_get_be32(f);
1730 version_id = qemu_get_be32(f);
1732 trace_qemu_loadvm_state_section_startfull(section_id, idstr,
1733 instance_id, version_id);
1734 /* Find savevm section */
1735 se = find_se(idstr, instance_id);
1736 if (se == NULL) {
1737 error_report("Unknown savevm section or instance '%s' %d",
1738 idstr, instance_id);
1739 return -EINVAL;
1742 /* Validate version */
1743 if (version_id > se->version_id) {
1744 error_report("savevm: unsupported version %d for '%s' v%d",
1745 version_id, idstr, se->version_id);
1746 return -EINVAL;
1749 /* Add entry */
1750 le = g_malloc0(sizeof(*le));
1752 le->se = se;
1753 le->section_id = section_id;
1754 le->version_id = version_id;
1755 QLIST_INSERT_HEAD(&mis->loadvm_handlers, le, entry);
1757 ret = vmstate_load(f, le->se, le->version_id);
1758 if (ret < 0) {
1759 error_report("error while loading state for instance 0x%x of"
1760 " device '%s'", instance_id, idstr);
1761 return ret;
1763 if (!check_section_footer(f, le)) {
1764 return -EINVAL;
1766 break;
1767 case QEMU_VM_SECTION_PART:
1768 case QEMU_VM_SECTION_END:
1769 section_id = qemu_get_be32(f);
1771 trace_qemu_loadvm_state_section_partend(section_id);
1772 QLIST_FOREACH(le, &mis->loadvm_handlers, entry) {
1773 if (le->section_id == section_id) {
1774 break;
1777 if (le == NULL) {
1778 error_report("Unknown savevm section %d", section_id);
1779 return -EINVAL;
1782 ret = vmstate_load(f, le->se, le->version_id);
1783 if (ret < 0) {
1784 error_report("error while loading state section id %d(%s)",
1785 section_id, le->se->idstr);
1786 return ret;
1788 if (!check_section_footer(f, le)) {
1789 return -EINVAL;
1791 break;
1792 case QEMU_VM_COMMAND:
1793 ret = loadvm_process_command(f);
1794 trace_qemu_loadvm_state_section_command(ret);
1795 if ((ret < 0) || (ret & LOADVM_QUIT)) {
1796 return ret;
1798 break;
1799 default:
1800 error_report("Unknown savevm section type %d", section_type);
1801 return -EINVAL;
1805 return 0;
1808 int qemu_loadvm_state(QEMUFile *f)
1810 MigrationIncomingState *mis = migration_incoming_get_current();
1811 Error *local_err = NULL;
1812 unsigned int v;
1813 int ret;
1815 if (qemu_savevm_state_blocked(&local_err)) {
1816 error_report_err(local_err);
1817 return -EINVAL;
1820 v = qemu_get_be32(f);
1821 if (v != QEMU_VM_FILE_MAGIC) {
1822 error_report("Not a migration stream");
1823 return -EINVAL;
1826 v = qemu_get_be32(f);
1827 if (v == QEMU_VM_FILE_VERSION_COMPAT) {
1828 error_report("SaveVM v2 format is obsolete and don't work anymore");
1829 return -ENOTSUP;
1831 if (v != QEMU_VM_FILE_VERSION) {
1832 error_report("Unsupported migration stream version");
1833 return -ENOTSUP;
1836 if (!savevm_state.skip_configuration) {
1837 if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
1838 error_report("Configuration section missing");
1839 return -EINVAL;
1841 ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
1843 if (ret) {
1844 return ret;
1848 ret = qemu_loadvm_state_main(f, mis);
1849 qemu_event_set(&mis->main_thread_load_event);
1851 trace_qemu_loadvm_state_post_main(ret);
1853 if (mis->have_listen_thread) {
1854 /* Listen thread still going, can't clean up yet */
1855 return ret;
1858 if (ret == 0) {
1859 ret = qemu_file_get_error(f);
1863 * Try to read in the VMDESC section as well, so that dumping tools that
1864 * intercept our migration stream have the chance to see it.
1867 /* We've got to be careful; if we don't read the data and just shut the fd
1868 * then the sender can error if we close while it's still sending.
1869 * We also mustn't read data that isn't there; some transports (RDMA)
1870 * will stall waiting for that data when the source has already closed.
1872 if (ret == 0 && should_send_vmdesc()) {
1873 uint8_t *buf;
1874 uint32_t size;
1875 uint8_t section_type = qemu_get_byte(f);
1877 if (section_type != QEMU_VM_VMDESCRIPTION) {
1878 error_report("Expected vmdescription section, but got %d",
1879 section_type);
1881 * It doesn't seem worth failing at this point since
1882 * we apparently have an otherwise valid VM state
1884 } else {
1885 buf = g_malloc(0x1000);
1886 size = qemu_get_be32(f);
1888 while (size > 0) {
1889 uint32_t read_chunk = MIN(size, 0x1000);
1890 qemu_get_buffer(f, buf, read_chunk);
1891 size -= read_chunk;
1893 g_free(buf);
1897 cpu_synchronize_all_post_init();
1899 return ret;
1902 static BlockDriverState *find_vmstate_bs(void)
1904 BlockDriverState *bs = NULL;
1905 while ((bs = bdrv_next(bs))) {
1906 if (bdrv_can_snapshot(bs)) {
1907 return bs;
1910 return NULL;
1914 * Deletes snapshots of a given name in all opened images.
1916 static int del_existing_snapshots(Monitor *mon, const char *name)
1918 BlockDriverState *bs;
1919 QEMUSnapshotInfo sn1, *snapshot = &sn1;
1920 Error *err = NULL;
1922 bs = NULL;
1923 while ((bs = bdrv_next(bs))) {
1924 if (bdrv_can_snapshot(bs) &&
1925 bdrv_snapshot_find(bs, snapshot, name) >= 0) {
1926 bdrv_snapshot_delete_by_id_or_name(bs, name, &err);
1927 if (err) {
1928 monitor_printf(mon,
1929 "Error while deleting snapshot on device '%s':"
1930 " %s\n",
1931 bdrv_get_device_name(bs),
1932 error_get_pretty(err));
1933 error_free(err);
1934 return -1;
1939 return 0;
1942 void hmp_savevm(Monitor *mon, const QDict *qdict)
1944 BlockDriverState *bs, *bs1;
1945 QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1;
1946 int ret;
1947 QEMUFile *f;
1948 int saved_vm_running;
1949 uint64_t vm_state_size;
1950 qemu_timeval tv;
1951 struct tm tm;
1952 const char *name = qdict_get_try_str(qdict, "name");
1953 Error *local_err = NULL;
1955 /* Verify if there is a device that doesn't support snapshots and is writable */
1956 bs = NULL;
1957 while ((bs = bdrv_next(bs))) {
1959 if (!bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
1960 continue;
1963 if (!bdrv_can_snapshot(bs)) {
1964 monitor_printf(mon, "Device '%s' is writable but does not support snapshots.\n",
1965 bdrv_get_device_name(bs));
1966 return;
1970 bs = find_vmstate_bs();
1971 if (!bs) {
1972 monitor_printf(mon, "No block device can accept snapshots\n");
1973 return;
1976 saved_vm_running = runstate_is_running();
1978 ret = global_state_store();
1979 if (ret) {
1980 monitor_printf(mon, "Error saving global state\n");
1981 return;
1983 vm_stop(RUN_STATE_SAVE_VM);
1985 memset(sn, 0, sizeof(*sn));
1987 /* fill auxiliary fields */
1988 qemu_gettimeofday(&tv);
1989 sn->date_sec = tv.tv_sec;
1990 sn->date_nsec = tv.tv_usec * 1000;
1991 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1993 if (name) {
1994 ret = bdrv_snapshot_find(bs, old_sn, name);
1995 if (ret >= 0) {
1996 pstrcpy(sn->name, sizeof(sn->name), old_sn->name);
1997 pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str);
1998 } else {
1999 pstrcpy(sn->name, sizeof(sn->name), name);
2001 } else {
2002 /* cast below needed for OpenBSD where tv_sec is still 'long' */
2003 localtime_r((const time_t *)&tv.tv_sec, &tm);
2004 strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm);
2007 /* Delete old snapshots of the same name */
2008 if (name && del_existing_snapshots(mon, name) < 0) {
2009 goto the_end;
2012 /* save the VM state */
2013 f = qemu_fopen_bdrv(bs, 1);
2014 if (!f) {
2015 monitor_printf(mon, "Could not open VM state file\n");
2016 goto the_end;
2018 ret = qemu_savevm_state(f, &local_err);
2019 vm_state_size = qemu_ftell(f);
2020 qemu_fclose(f);
2021 if (ret < 0) {
2022 monitor_printf(mon, "%s\n", error_get_pretty(local_err));
2023 error_free(local_err);
2024 goto the_end;
2027 /* create the snapshots */
2029 bs1 = NULL;
2030 while ((bs1 = bdrv_next(bs1))) {
2031 if (bdrv_can_snapshot(bs1)) {
2032 /* Write VM state size only to the image that contains the state */
2033 sn->vm_state_size = (bs == bs1 ? vm_state_size : 0);
2034 ret = bdrv_snapshot_create(bs1, sn);
2035 if (ret < 0) {
2036 monitor_printf(mon, "Error while creating snapshot on '%s'\n",
2037 bdrv_get_device_name(bs1));
2042 the_end:
2043 if (saved_vm_running) {
2044 vm_start();
2048 void qmp_xen_save_devices_state(const char *filename, Error **errp)
2050 QEMUFile *f;
2051 int saved_vm_running;
2052 int ret;
2054 saved_vm_running = runstate_is_running();
2055 vm_stop(RUN_STATE_SAVE_VM);
2056 global_state_store_running();
2058 f = qemu_fopen(filename, "wb");
2059 if (!f) {
2060 error_setg_file_open(errp, errno, filename);
2061 goto the_end;
2063 ret = qemu_save_device_state(f);
2064 qemu_fclose(f);
2065 if (ret < 0) {
2066 error_setg(errp, QERR_IO_ERROR);
2069 the_end:
2070 if (saved_vm_running) {
2071 vm_start();
2075 int load_vmstate(const char *name)
2077 BlockDriverState *bs, *bs_vm_state;
2078 QEMUSnapshotInfo sn;
2079 QEMUFile *f;
2080 int ret;
2082 bs_vm_state = find_vmstate_bs();
2083 if (!bs_vm_state) {
2084 error_report("No block device supports snapshots");
2085 return -ENOTSUP;
2088 /* Don't even try to load empty VM states */
2089 ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
2090 if (ret < 0) {
2091 return ret;
2092 } else if (sn.vm_state_size == 0) {
2093 error_report("This is a disk-only snapshot. Revert to it offline "
2094 "using qemu-img.");
2095 return -EINVAL;
2098 /* Verify if there is any device that doesn't support snapshots and is
2099 writable and check if the requested snapshot is available too. */
2100 bs = NULL;
2101 while ((bs = bdrv_next(bs))) {
2103 if (!bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
2104 continue;
2107 if (!bdrv_can_snapshot(bs)) {
2108 error_report("Device '%s' is writable but does not support snapshots.",
2109 bdrv_get_device_name(bs));
2110 return -ENOTSUP;
2113 ret = bdrv_snapshot_find(bs, &sn, name);
2114 if (ret < 0) {
2115 error_report("Device '%s' does not have the requested snapshot '%s'",
2116 bdrv_get_device_name(bs), name);
2117 return ret;
2121 /* Flush all IO requests so they don't interfere with the new state. */
2122 bdrv_drain_all();
2124 bs = NULL;
2125 while ((bs = bdrv_next(bs))) {
2126 if (bdrv_can_snapshot(bs)) {
2127 ret = bdrv_snapshot_goto(bs, name);
2128 if (ret < 0) {
2129 error_report("Error %d while activating snapshot '%s' on '%s'",
2130 ret, name, bdrv_get_device_name(bs));
2131 return ret;
2136 /* restore the VM state */
2137 f = qemu_fopen_bdrv(bs_vm_state, 0);
2138 if (!f) {
2139 error_report("Could not open VM state file");
2140 return -EINVAL;
2143 qemu_system_reset(VMRESET_SILENT);
2144 migration_incoming_state_new(f);
2145 ret = qemu_loadvm_state(f);
2147 qemu_fclose(f);
2148 migration_incoming_state_destroy();
2149 if (ret < 0) {
2150 error_report("Error %d while loading VM state", ret);
2151 return ret;
2154 return 0;
2157 void hmp_delvm(Monitor *mon, const QDict *qdict)
2159 BlockDriverState *bs;
2160 Error *err;
2161 const char *name = qdict_get_str(qdict, "name");
2163 if (!find_vmstate_bs()) {
2164 monitor_printf(mon, "No block device supports snapshots\n");
2165 return;
2168 bs = NULL;
2169 while ((bs = bdrv_next(bs))) {
2170 if (bdrv_can_snapshot(bs)) {
2171 err = NULL;
2172 bdrv_snapshot_delete_by_id_or_name(bs, name, &err);
2173 if (err) {
2174 monitor_printf(mon,
2175 "Error while deleting snapshot on device '%s':"
2176 " %s\n",
2177 bdrv_get_device_name(bs),
2178 error_get_pretty(err));
2179 error_free(err);
2185 void hmp_info_snapshots(Monitor *mon, const QDict *qdict)
2187 BlockDriverState *bs, *bs1;
2188 QEMUSnapshotInfo *sn_tab, *sn, s, *sn_info = &s;
2189 int nb_sns, i, ret, available;
2190 int total;
2191 int *available_snapshots;
2193 bs = find_vmstate_bs();
2194 if (!bs) {
2195 monitor_printf(mon, "No available block device supports snapshots\n");
2196 return;
2199 nb_sns = bdrv_snapshot_list(bs, &sn_tab);
2200 if (nb_sns < 0) {
2201 monitor_printf(mon, "bdrv_snapshot_list: error %d\n", nb_sns);
2202 return;
2205 if (nb_sns == 0) {
2206 monitor_printf(mon, "There is no snapshot available.\n");
2207 return;
2210 available_snapshots = g_new0(int, nb_sns);
2211 total = 0;
2212 for (i = 0; i < nb_sns; i++) {
2213 sn = &sn_tab[i];
2214 available = 1;
2215 bs1 = NULL;
2217 while ((bs1 = bdrv_next(bs1))) {
2218 if (bdrv_can_snapshot(bs1) && bs1 != bs) {
2219 ret = bdrv_snapshot_find(bs1, sn_info, sn->id_str);
2220 if (ret < 0) {
2221 available = 0;
2222 break;
2227 if (available) {
2228 available_snapshots[total] = i;
2229 total++;
2233 if (total > 0) {
2234 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, NULL);
2235 monitor_printf(mon, "\n");
2236 for (i = 0; i < total; i++) {
2237 sn = &sn_tab[available_snapshots[i]];
2238 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, sn);
2239 monitor_printf(mon, "\n");
2241 } else {
2242 monitor_printf(mon, "There is no suitable snapshot available\n");
2245 g_free(sn_tab);
2246 g_free(available_snapshots);
2250 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
2252 qemu_ram_set_idstr(memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK,
2253 memory_region_name(mr), dev);
2256 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
2258 qemu_ram_unset_idstr(memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK);
2261 void vmstate_register_ram_global(MemoryRegion *mr)
2263 vmstate_register_ram(mr, NULL);