vfio: Create device specific region info helper
[qemu/ar7.git] / migration / savevm.c
blob65ce0c61a3d7b8389b0d5b5544767b8f09c99c71
1 /*
2 * QEMU System Emulator
4 * Copyright (c) 2003-2008 Fabrice Bellard
5 * Copyright (c) 2009-2015 Red Hat Inc
7 * Authors:
8 * Juan Quintela <quintela@redhat.com>
10 * Permission is hereby granted, free of charge, to any person obtaining a copy
11 * of this software and associated documentation files (the "Software"), to deal
12 * in the Software without restriction, including without limitation the rights
13 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
14 * copies of the Software, and to permit persons to whom the Software is
15 * furnished to do so, subject to the following conditions:
17 * The above copyright notice and this permission notice shall be included in
18 * all copies or substantial portions of the Software.
20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
21 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
23 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
24 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
25 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 * THE SOFTWARE.
29 #include "qemu/osdep.h"
30 #include "cpu.h"
31 #include "hw/boards.h"
32 #include "hw/hw.h"
33 #include "hw/qdev.h"
34 #include "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"
53 #include "qemu/cutils.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 (int) state->len, 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 static bool enforce_config_section(void)
883 MachineState *machine = MACHINE(qdev_get_machine());
884 return machine->enforce_config_section;
887 void qemu_savevm_state_header(QEMUFile *f)
889 trace_savevm_state_header();
890 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
891 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
893 if (!savevm_state.skip_configuration || enforce_config_section()) {
894 qemu_put_byte(f, QEMU_VM_CONFIGURATION);
895 vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
900 void qemu_savevm_state_begin(QEMUFile *f,
901 const MigrationParams *params)
903 SaveStateEntry *se;
904 int ret;
906 trace_savevm_state_begin();
907 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
908 if (!se->ops || !se->ops->set_params) {
909 continue;
911 se->ops->set_params(params, se->opaque);
914 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
915 if (!se->ops || !se->ops->save_live_setup) {
916 continue;
918 if (se->ops && se->ops->is_active) {
919 if (!se->ops->is_active(se->opaque)) {
920 continue;
923 save_section_header(f, se, QEMU_VM_SECTION_START);
925 ret = se->ops->save_live_setup(f, se->opaque);
926 save_section_footer(f, se);
927 if (ret < 0) {
928 qemu_file_set_error(f, ret);
929 break;
935 * this function has three return values:
936 * negative: there was one error, and we have -errno.
937 * 0 : We haven't finished, caller have to go again
938 * 1 : We have finished, we can go to complete phase
940 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy)
942 SaveStateEntry *se;
943 int ret = 1;
945 trace_savevm_state_iterate();
946 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
947 if (!se->ops || !se->ops->save_live_iterate) {
948 continue;
950 if (se->ops && se->ops->is_active) {
951 if (!se->ops->is_active(se->opaque)) {
952 continue;
956 * In the postcopy phase, any device that doesn't know how to
957 * do postcopy should have saved it's state in the _complete
958 * call that's already run, it might get confused if we call
959 * iterate afterwards.
961 if (postcopy && !se->ops->save_live_complete_postcopy) {
962 continue;
964 if (qemu_file_rate_limit(f)) {
965 return 0;
967 trace_savevm_section_start(se->idstr, se->section_id);
969 save_section_header(f, se, QEMU_VM_SECTION_PART);
971 ret = se->ops->save_live_iterate(f, se->opaque);
972 trace_savevm_section_end(se->idstr, se->section_id, ret);
973 save_section_footer(f, se);
975 if (ret < 0) {
976 qemu_file_set_error(f, ret);
978 if (ret <= 0) {
979 /* Do not proceed to the next vmstate before this one reported
980 completion of the current stage. This serializes the migration
981 and reduces the probability that a faster changing state is
982 synchronized over and over again. */
983 break;
986 return ret;
989 static bool should_send_vmdesc(void)
991 MachineState *machine = MACHINE(qdev_get_machine());
992 bool in_postcopy = migration_in_postcopy(migrate_get_current());
993 return !machine->suppress_vmdesc && !in_postcopy;
997 * Calls the save_live_complete_postcopy methods
998 * causing the last few pages to be sent immediately and doing any associated
999 * cleanup.
1000 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1001 * all the other devices, but that happens at the point we switch to postcopy.
1003 void qemu_savevm_state_complete_postcopy(QEMUFile *f)
1005 SaveStateEntry *se;
1006 int ret;
1008 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1009 if (!se->ops || !se->ops->save_live_complete_postcopy) {
1010 continue;
1012 if (se->ops && se->ops->is_active) {
1013 if (!se->ops->is_active(se->opaque)) {
1014 continue;
1017 trace_savevm_section_start(se->idstr, se->section_id);
1018 /* Section type */
1019 qemu_put_byte(f, QEMU_VM_SECTION_END);
1020 qemu_put_be32(f, se->section_id);
1022 ret = se->ops->save_live_complete_postcopy(f, se->opaque);
1023 trace_savevm_section_end(se->idstr, se->section_id, ret);
1024 save_section_footer(f, se);
1025 if (ret < 0) {
1026 qemu_file_set_error(f, ret);
1027 return;
1031 qemu_put_byte(f, QEMU_VM_EOF);
1032 qemu_fflush(f);
1035 void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only)
1037 QJSON *vmdesc;
1038 int vmdesc_len;
1039 SaveStateEntry *se;
1040 int ret;
1041 bool in_postcopy = migration_in_postcopy(migrate_get_current());
1043 trace_savevm_state_complete_precopy();
1045 cpu_synchronize_all_states();
1047 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1048 if (!se->ops ||
1049 (in_postcopy && se->ops->save_live_complete_postcopy) ||
1050 (in_postcopy && !iterable_only) ||
1051 !se->ops->save_live_complete_precopy) {
1052 continue;
1055 if (se->ops && se->ops->is_active) {
1056 if (!se->ops->is_active(se->opaque)) {
1057 continue;
1060 trace_savevm_section_start(se->idstr, se->section_id);
1062 save_section_header(f, se, QEMU_VM_SECTION_END);
1064 ret = se->ops->save_live_complete_precopy(f, se->opaque);
1065 trace_savevm_section_end(se->idstr, se->section_id, ret);
1066 save_section_footer(f, se);
1067 if (ret < 0) {
1068 qemu_file_set_error(f, ret);
1069 return;
1073 if (iterable_only) {
1074 return;
1077 vmdesc = qjson_new();
1078 json_prop_int(vmdesc, "page_size", TARGET_PAGE_SIZE);
1079 json_start_array(vmdesc, "devices");
1080 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1082 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1083 continue;
1085 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1086 trace_savevm_section_skip(se->idstr, se->section_id);
1087 continue;
1090 trace_savevm_section_start(se->idstr, se->section_id);
1092 json_start_object(vmdesc, NULL);
1093 json_prop_str(vmdesc, "name", se->idstr);
1094 json_prop_int(vmdesc, "instance_id", se->instance_id);
1096 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1097 vmstate_save(f, se, vmdesc);
1098 trace_savevm_section_end(se->idstr, se->section_id, 0);
1099 save_section_footer(f, se);
1101 json_end_object(vmdesc);
1104 if (!in_postcopy) {
1105 /* Postcopy stream will still be going */
1106 qemu_put_byte(f, QEMU_VM_EOF);
1109 json_end_array(vmdesc);
1110 qjson_finish(vmdesc);
1111 vmdesc_len = strlen(qjson_get_str(vmdesc));
1113 if (should_send_vmdesc()) {
1114 qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
1115 qemu_put_be32(f, vmdesc_len);
1116 qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len);
1118 qjson_destroy(vmdesc);
1120 qemu_fflush(f);
1123 /* Give an estimate of the amount left to be transferred,
1124 * the result is split into the amount for units that can and
1125 * for units that can't do postcopy.
1127 void qemu_savevm_state_pending(QEMUFile *f, uint64_t max_size,
1128 uint64_t *res_non_postcopiable,
1129 uint64_t *res_postcopiable)
1131 SaveStateEntry *se;
1133 *res_non_postcopiable = 0;
1134 *res_postcopiable = 0;
1137 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1138 if (!se->ops || !se->ops->save_live_pending) {
1139 continue;
1141 if (se->ops && se->ops->is_active) {
1142 if (!se->ops->is_active(se->opaque)) {
1143 continue;
1146 se->ops->save_live_pending(f, se->opaque, max_size,
1147 res_non_postcopiable, res_postcopiable);
1151 void qemu_savevm_state_cleanup(void)
1153 SaveStateEntry *se;
1155 trace_savevm_state_cleanup();
1156 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1157 if (se->ops && se->ops->cleanup) {
1158 se->ops->cleanup(se->opaque);
1163 static int qemu_savevm_state(QEMUFile *f, Error **errp)
1165 int ret;
1166 MigrationParams params = {
1167 .blk = 0,
1168 .shared = 0
1170 MigrationState *ms = migrate_init(&params);
1171 ms->to_dst_file = f;
1173 if (migration_is_blocked(errp)) {
1174 return -EINVAL;
1177 qemu_mutex_unlock_iothread();
1178 qemu_savevm_state_header(f);
1179 qemu_savevm_state_begin(f, &params);
1180 qemu_mutex_lock_iothread();
1182 while (qemu_file_get_error(f) == 0) {
1183 if (qemu_savevm_state_iterate(f, false) > 0) {
1184 break;
1188 ret = qemu_file_get_error(f);
1189 if (ret == 0) {
1190 qemu_savevm_state_complete_precopy(f, false);
1191 ret = qemu_file_get_error(f);
1193 qemu_savevm_state_cleanup();
1194 if (ret != 0) {
1195 error_setg_errno(errp, -ret, "Error while writing VM state");
1197 return ret;
1200 static int qemu_save_device_state(QEMUFile *f)
1202 SaveStateEntry *se;
1204 qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
1205 qemu_put_be32(f, QEMU_VM_FILE_VERSION);
1207 cpu_synchronize_all_states();
1209 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1210 if (se->is_ram) {
1211 continue;
1213 if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
1214 continue;
1216 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
1217 continue;
1220 save_section_header(f, se, QEMU_VM_SECTION_FULL);
1222 vmstate_save(f, se, NULL);
1224 save_section_footer(f, se);
1227 qemu_put_byte(f, QEMU_VM_EOF);
1229 return qemu_file_get_error(f);
1232 static SaveStateEntry *find_se(const char *idstr, int instance_id)
1234 SaveStateEntry *se;
1236 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
1237 if (!strcmp(se->idstr, idstr) &&
1238 (instance_id == se->instance_id ||
1239 instance_id == se->alias_id))
1240 return se;
1241 /* Migrating from an older version? */
1242 if (strstr(se->idstr, idstr) && se->compat) {
1243 if (!strcmp(se->compat->idstr, idstr) &&
1244 (instance_id == se->compat->instance_id ||
1245 instance_id == se->alias_id))
1246 return se;
1249 return NULL;
1252 enum LoadVMExitCodes {
1253 /* Allow a command to quit all layers of nested loadvm loops */
1254 LOADVM_QUIT = 1,
1257 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis);
1259 /* ------ incoming postcopy messages ------ */
1260 /* 'advise' arrives before any transfers just to tell us that a postcopy
1261 * *might* happen - it might be skipped if precopy transferred everything
1262 * quickly.
1264 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis)
1266 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1267 uint64_t remote_hps, remote_tps;
1269 trace_loadvm_postcopy_handle_advise();
1270 if (ps != POSTCOPY_INCOMING_NONE) {
1271 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps);
1272 return -1;
1275 if (!postcopy_ram_supported_by_host()) {
1276 return -1;
1279 remote_hps = qemu_get_be64(mis->from_src_file);
1280 if (remote_hps != getpagesize()) {
1282 * Some combinations of mismatch are probably possible but it gets
1283 * a bit more complicated. In particular we need to place whole
1284 * host pages on the dest at once, and we need to ensure that we
1285 * handle dirtying to make sure we never end up sending part of
1286 * a hostpage on it's own.
1288 error_report("Postcopy needs matching host page sizes (s=%d d=%d)",
1289 (int)remote_hps, getpagesize());
1290 return -1;
1293 remote_tps = qemu_get_be64(mis->from_src_file);
1294 if (remote_tps != (1ul << qemu_target_page_bits())) {
1296 * Again, some differences could be dealt with, but for now keep it
1297 * simple.
1299 error_report("Postcopy needs matching target page sizes (s=%d d=%d)",
1300 (int)remote_tps, 1 << qemu_target_page_bits());
1301 return -1;
1304 if (ram_postcopy_incoming_init(mis)) {
1305 return -1;
1308 postcopy_state_set(POSTCOPY_INCOMING_ADVISE);
1310 return 0;
1313 /* After postcopy we will be told to throw some pages away since they're
1314 * dirty and will have to be demand fetched. Must happen before CPU is
1315 * started.
1316 * There can be 0..many of these messages, each encoding multiple pages.
1318 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis,
1319 uint16_t len)
1321 int tmp;
1322 char ramid[256];
1323 PostcopyState ps = postcopy_state_get();
1325 trace_loadvm_postcopy_ram_handle_discard();
1327 switch (ps) {
1328 case POSTCOPY_INCOMING_ADVISE:
1329 /* 1st discard */
1330 tmp = postcopy_ram_prepare_discard(mis);
1331 if (tmp) {
1332 return tmp;
1334 break;
1336 case POSTCOPY_INCOMING_DISCARD:
1337 /* Expected state */
1338 break;
1340 default:
1341 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1342 ps);
1343 return -1;
1345 /* We're expecting a
1346 * Version (0)
1347 * a RAM ID string (length byte, name, 0 term)
1348 * then at least 1 16 byte chunk
1350 if (len < (1 + 1 + 1 + 1 + 2 * 8)) {
1351 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1352 return -1;
1355 tmp = qemu_get_byte(mis->from_src_file);
1356 if (tmp != postcopy_ram_discard_version) {
1357 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp);
1358 return -1;
1361 if (!qemu_get_counted_string(mis->from_src_file, ramid)) {
1362 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1363 return -1;
1365 tmp = qemu_get_byte(mis->from_src_file);
1366 if (tmp != 0) {
1367 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp);
1368 return -1;
1371 len -= 3 + strlen(ramid);
1372 if (len % 16) {
1373 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len);
1374 return -1;
1376 trace_loadvm_postcopy_ram_handle_discard_header(ramid, len);
1377 while (len) {
1378 uint64_t start_addr, block_length;
1379 start_addr = qemu_get_be64(mis->from_src_file);
1380 block_length = qemu_get_be64(mis->from_src_file);
1382 len -= 16;
1383 int ret = ram_discard_range(mis, ramid, start_addr,
1384 block_length);
1385 if (ret) {
1386 return ret;
1389 trace_loadvm_postcopy_ram_handle_discard_end();
1391 return 0;
1395 * Triggered by a postcopy_listen command; this thread takes over reading
1396 * the input stream, leaving the main thread free to carry on loading the rest
1397 * of the device state (from RAM).
1398 * (TODO:This could do with being in a postcopy file - but there again it's
1399 * just another input loop, not that postcopy specific)
1401 static void *postcopy_ram_listen_thread(void *opaque)
1403 QEMUFile *f = opaque;
1404 MigrationIncomingState *mis = migration_incoming_get_current();
1405 int load_res;
1407 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
1408 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1409 qemu_sem_post(&mis->listen_thread_sem);
1410 trace_postcopy_ram_listen_thread_start();
1413 * Because we're a thread and not a coroutine we can't yield
1414 * in qemu_file, and thus we must be blocking now.
1416 qemu_file_set_blocking(f, true);
1417 load_res = qemu_loadvm_state_main(f, mis);
1418 /* And non-blocking again so we don't block in any cleanup */
1419 qemu_file_set_blocking(f, false);
1421 trace_postcopy_ram_listen_thread_exit();
1422 if (load_res < 0) {
1423 error_report("%s: loadvm failed: %d", __func__, load_res);
1424 qemu_file_set_error(f, load_res);
1425 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1426 MIGRATION_STATUS_FAILED);
1427 } else {
1429 * This looks good, but it's possible that the device loading in the
1430 * main thread hasn't finished yet, and so we might not be in 'RUN'
1431 * state yet; wait for the end of the main thread.
1433 qemu_event_wait(&mis->main_thread_load_event);
1435 postcopy_ram_incoming_cleanup(mis);
1437 if (load_res < 0) {
1439 * If something went wrong then we have a bad state so exit;
1440 * depending how far we got it might be possible at this point
1441 * to leave the guest running and fire MCEs for pages that never
1442 * arrived as a desperate recovery step.
1444 exit(EXIT_FAILURE);
1447 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1448 MIGRATION_STATUS_COMPLETED);
1450 * If everything has worked fine, then the main thread has waited
1451 * for us to start, and we're the last use of the mis.
1452 * (If something broke then qemu will have to exit anyway since it's
1453 * got a bad migration state).
1455 migration_incoming_state_destroy();
1458 return NULL;
1461 /* After this message we must be able to immediately receive postcopy data */
1462 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
1464 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING);
1465 trace_loadvm_postcopy_handle_listen();
1466 if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) {
1467 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps);
1468 return -1;
1470 if (ps == POSTCOPY_INCOMING_ADVISE) {
1472 * A rare case, we entered listen without having to do any discards,
1473 * so do the setup that's normally done at the time of the 1st discard.
1475 postcopy_ram_prepare_discard(mis);
1479 * Sensitise RAM - can now generate requests for blocks that don't exist
1480 * However, at this point the CPU shouldn't be running, and the IO
1481 * shouldn't be doing anything yet so don't actually expect requests
1483 if (postcopy_ram_enable_notify(mis)) {
1484 return -1;
1487 if (mis->have_listen_thread) {
1488 error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread");
1489 return -1;
1492 mis->have_listen_thread = true;
1493 /* Start up the listening thread and wait for it to signal ready */
1494 qemu_sem_init(&mis->listen_thread_sem, 0);
1495 qemu_thread_create(&mis->listen_thread, "postcopy/listen",
1496 postcopy_ram_listen_thread, mis->from_src_file,
1497 QEMU_THREAD_DETACHED);
1498 qemu_sem_wait(&mis->listen_thread_sem);
1499 qemu_sem_destroy(&mis->listen_thread_sem);
1501 return 0;
1505 typedef struct {
1506 QEMUBH *bh;
1507 } HandleRunBhData;
1509 static void loadvm_postcopy_handle_run_bh(void *opaque)
1511 Error *local_err = NULL;
1512 HandleRunBhData *data = opaque;
1514 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1515 * in migration.c
1517 cpu_synchronize_all_post_init();
1519 qemu_announce_self();
1521 /* Make sure all file formats flush their mutable metadata */
1522 bdrv_invalidate_cache_all(&local_err);
1523 if (local_err) {
1524 error_report_err(local_err);
1527 trace_loadvm_postcopy_handle_run_cpu_sync();
1528 cpu_synchronize_all_post_init();
1530 trace_loadvm_postcopy_handle_run_vmstart();
1532 if (autostart) {
1533 /* Hold onto your hats, starting the CPU */
1534 vm_start();
1535 } else {
1536 /* leave it paused and let management decide when to start the CPU */
1537 runstate_set(RUN_STATE_PAUSED);
1540 qemu_bh_delete(data->bh);
1541 g_free(data);
1544 /* After all discards we can start running and asking for pages */
1545 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis)
1547 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_RUNNING);
1548 HandleRunBhData *data;
1550 trace_loadvm_postcopy_handle_run();
1551 if (ps != POSTCOPY_INCOMING_LISTENING) {
1552 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps);
1553 return -1;
1556 data = g_new(HandleRunBhData, 1);
1557 data->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, data);
1558 qemu_bh_schedule(data->bh);
1560 /* We need to finish reading the stream from the package
1561 * and also stop reading anything more from the stream that loaded the
1562 * package (since it's now being read by the listener thread).
1563 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
1565 return LOADVM_QUIT;
1569 * Immediately following this command is a blob of data containing an embedded
1570 * chunk of migration stream; read it and load it.
1572 * @mis: Incoming state
1573 * @length: Length of packaged data to read
1575 * Returns: Negative values on error
1578 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis)
1580 int ret;
1581 uint8_t *buffer;
1582 uint32_t length;
1583 QEMUSizedBuffer *qsb;
1585 length = qemu_get_be32(mis->from_src_file);
1586 trace_loadvm_handle_cmd_packaged(length);
1588 if (length > MAX_VM_CMD_PACKAGED_SIZE) {
1589 error_report("Unreasonably large packaged state: %u", length);
1590 return -1;
1592 buffer = g_malloc0(length);
1593 ret = qemu_get_buffer(mis->from_src_file, buffer, (int)length);
1594 if (ret != length) {
1595 g_free(buffer);
1596 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%d",
1597 ret, length);
1598 return (ret < 0) ? ret : -EAGAIN;
1600 trace_loadvm_handle_cmd_packaged_received(ret);
1602 /* Setup a dummy QEMUFile that actually reads from the buffer */
1603 qsb = qsb_create(buffer, length);
1604 g_free(buffer); /* Because qsb_create copies */
1605 if (!qsb) {
1606 error_report("Unable to create qsb");
1608 QEMUFile *packf = qemu_bufopen("r", qsb);
1610 ret = qemu_loadvm_state_main(packf, mis);
1611 trace_loadvm_handle_cmd_packaged_main(ret);
1612 qemu_fclose(packf);
1613 qsb_free(qsb);
1615 return ret;
1619 * Process an incoming 'QEMU_VM_COMMAND'
1620 * 0 just a normal return
1621 * LOADVM_QUIT All good, but exit the loop
1622 * <0 Error
1624 static int loadvm_process_command(QEMUFile *f)
1626 MigrationIncomingState *mis = migration_incoming_get_current();
1627 uint16_t cmd;
1628 uint16_t len;
1629 uint32_t tmp32;
1631 cmd = qemu_get_be16(f);
1632 len = qemu_get_be16(f);
1634 trace_loadvm_process_command(cmd, len);
1635 if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) {
1636 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len);
1637 return -EINVAL;
1640 if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) {
1641 error_report("%s received with bad length - expecting %zu, got %d",
1642 mig_cmd_args[cmd].name,
1643 (size_t)mig_cmd_args[cmd].len, len);
1644 return -ERANGE;
1647 switch (cmd) {
1648 case MIG_CMD_OPEN_RETURN_PATH:
1649 if (mis->to_src_file) {
1650 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
1651 /* Not really a problem, so don't give up */
1652 return 0;
1654 mis->to_src_file = qemu_file_get_return_path(f);
1655 if (!mis->to_src_file) {
1656 error_report("CMD_OPEN_RETURN_PATH failed");
1657 return -1;
1659 break;
1661 case MIG_CMD_PING:
1662 tmp32 = qemu_get_be32(f);
1663 trace_loadvm_process_command_ping(tmp32);
1664 if (!mis->to_src_file) {
1665 error_report("CMD_PING (0x%x) received with no return path",
1666 tmp32);
1667 return -1;
1669 migrate_send_rp_pong(mis, tmp32);
1670 break;
1672 case MIG_CMD_PACKAGED:
1673 return loadvm_handle_cmd_packaged(mis);
1675 case MIG_CMD_POSTCOPY_ADVISE:
1676 return loadvm_postcopy_handle_advise(mis);
1678 case MIG_CMD_POSTCOPY_LISTEN:
1679 return loadvm_postcopy_handle_listen(mis);
1681 case MIG_CMD_POSTCOPY_RUN:
1682 return loadvm_postcopy_handle_run(mis);
1684 case MIG_CMD_POSTCOPY_RAM_DISCARD:
1685 return loadvm_postcopy_ram_handle_discard(mis, len);
1688 return 0;
1691 struct LoadStateEntry {
1692 QLIST_ENTRY(LoadStateEntry) entry;
1693 SaveStateEntry *se;
1694 int section_id;
1695 int version_id;
1699 * Read a footer off the wire and check that it matches the expected section
1701 * Returns: true if the footer was good
1702 * false if there is a problem (and calls error_report to say why)
1704 static bool check_section_footer(QEMUFile *f, LoadStateEntry *le)
1706 uint8_t read_mark;
1707 uint32_t read_section_id;
1709 if (skip_section_footers) {
1710 /* No footer to check */
1711 return true;
1714 read_mark = qemu_get_byte(f);
1716 if (read_mark != QEMU_VM_SECTION_FOOTER) {
1717 error_report("Missing section footer for %s", le->se->idstr);
1718 return false;
1721 read_section_id = qemu_get_be32(f);
1722 if (read_section_id != le->section_id) {
1723 error_report("Mismatched section id in footer for %s -"
1724 " read 0x%x expected 0x%x",
1725 le->se->idstr, read_section_id, le->section_id);
1726 return false;
1729 /* All good */
1730 return true;
1733 void loadvm_free_handlers(MigrationIncomingState *mis)
1735 LoadStateEntry *le, *new_le;
1737 QLIST_FOREACH_SAFE(le, &mis->loadvm_handlers, entry, new_le) {
1738 QLIST_REMOVE(le, entry);
1739 g_free(le);
1743 static int
1744 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis)
1746 uint32_t instance_id, version_id, section_id;
1747 SaveStateEntry *se;
1748 LoadStateEntry *le;
1749 char idstr[256];
1750 int ret;
1752 /* Read section start */
1753 section_id = qemu_get_be32(f);
1754 if (!qemu_get_counted_string(f, idstr)) {
1755 error_report("Unable to read ID string for section %u",
1756 section_id);
1757 return -EINVAL;
1759 instance_id = qemu_get_be32(f);
1760 version_id = qemu_get_be32(f);
1762 trace_qemu_loadvm_state_section_startfull(section_id, idstr,
1763 instance_id, version_id);
1764 /* Find savevm section */
1765 se = find_se(idstr, instance_id);
1766 if (se == NULL) {
1767 error_report("Unknown savevm section or instance '%s' %d",
1768 idstr, instance_id);
1769 return -EINVAL;
1772 /* Validate version */
1773 if (version_id > se->version_id) {
1774 error_report("savevm: unsupported version %d for '%s' v%d",
1775 version_id, idstr, se->version_id);
1776 return -EINVAL;
1779 /* Add entry */
1780 le = g_malloc0(sizeof(*le));
1782 le->se = se;
1783 le->section_id = section_id;
1784 le->version_id = version_id;
1785 QLIST_INSERT_HEAD(&mis->loadvm_handlers, le, entry);
1787 ret = vmstate_load(f, le->se, le->version_id);
1788 if (ret < 0) {
1789 error_report("error while loading state for instance 0x%x of"
1790 " device '%s'", instance_id, idstr);
1791 return ret;
1793 if (!check_section_footer(f, le)) {
1794 return -EINVAL;
1797 return 0;
1800 static int
1801 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis)
1803 uint32_t section_id;
1804 LoadStateEntry *le;
1805 int ret;
1807 section_id = qemu_get_be32(f);
1809 trace_qemu_loadvm_state_section_partend(section_id);
1810 QLIST_FOREACH(le, &mis->loadvm_handlers, entry) {
1811 if (le->section_id == section_id) {
1812 break;
1815 if (le == NULL) {
1816 error_report("Unknown savevm section %d", section_id);
1817 return -EINVAL;
1820 ret = vmstate_load(f, le->se, le->version_id);
1821 if (ret < 0) {
1822 error_report("error while loading state section id %d(%s)",
1823 section_id, le->se->idstr);
1824 return ret;
1826 if (!check_section_footer(f, le)) {
1827 return -EINVAL;
1830 return 0;
1833 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis)
1835 uint8_t section_type;
1836 int ret;
1838 while ((section_type = qemu_get_byte(f)) != QEMU_VM_EOF) {
1840 trace_qemu_loadvm_state_section(section_type);
1841 switch (section_type) {
1842 case QEMU_VM_SECTION_START:
1843 case QEMU_VM_SECTION_FULL:
1844 ret = qemu_loadvm_section_start_full(f, mis);
1845 if (ret < 0) {
1846 return ret;
1848 break;
1849 case QEMU_VM_SECTION_PART:
1850 case QEMU_VM_SECTION_END:
1851 ret = qemu_loadvm_section_part_end(f, mis);
1852 if (ret < 0) {
1853 return ret;
1855 break;
1856 case QEMU_VM_COMMAND:
1857 ret = loadvm_process_command(f);
1858 trace_qemu_loadvm_state_section_command(ret);
1859 if ((ret < 0) || (ret & LOADVM_QUIT)) {
1860 return ret;
1862 break;
1863 default:
1864 error_report("Unknown savevm section type %d", section_type);
1865 return -EINVAL;
1869 return 0;
1872 int qemu_loadvm_state(QEMUFile *f)
1874 MigrationIncomingState *mis = migration_incoming_get_current();
1875 Error *local_err = NULL;
1876 unsigned int v;
1877 int ret;
1879 if (qemu_savevm_state_blocked(&local_err)) {
1880 error_report_err(local_err);
1881 return -EINVAL;
1884 v = qemu_get_be32(f);
1885 if (v != QEMU_VM_FILE_MAGIC) {
1886 error_report("Not a migration stream");
1887 return -EINVAL;
1890 v = qemu_get_be32(f);
1891 if (v == QEMU_VM_FILE_VERSION_COMPAT) {
1892 error_report("SaveVM v2 format is obsolete and don't work anymore");
1893 return -ENOTSUP;
1895 if (v != QEMU_VM_FILE_VERSION) {
1896 error_report("Unsupported migration stream version");
1897 return -ENOTSUP;
1900 if (!savevm_state.skip_configuration || enforce_config_section()) {
1901 if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
1902 error_report("Configuration section missing");
1903 return -EINVAL;
1905 ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
1907 if (ret) {
1908 return ret;
1912 ret = qemu_loadvm_state_main(f, mis);
1913 qemu_event_set(&mis->main_thread_load_event);
1915 trace_qemu_loadvm_state_post_main(ret);
1917 if (mis->have_listen_thread) {
1918 /* Listen thread still going, can't clean up yet */
1919 return ret;
1922 if (ret == 0) {
1923 ret = qemu_file_get_error(f);
1927 * Try to read in the VMDESC section as well, so that dumping tools that
1928 * intercept our migration stream have the chance to see it.
1931 /* We've got to be careful; if we don't read the data and just shut the fd
1932 * then the sender can error if we close while it's still sending.
1933 * We also mustn't read data that isn't there; some transports (RDMA)
1934 * will stall waiting for that data when the source has already closed.
1936 if (ret == 0 && should_send_vmdesc()) {
1937 uint8_t *buf;
1938 uint32_t size;
1939 uint8_t section_type = qemu_get_byte(f);
1941 if (section_type != QEMU_VM_VMDESCRIPTION) {
1942 error_report("Expected vmdescription section, but got %d",
1943 section_type);
1945 * It doesn't seem worth failing at this point since
1946 * we apparently have an otherwise valid VM state
1948 } else {
1949 buf = g_malloc(0x1000);
1950 size = qemu_get_be32(f);
1952 while (size > 0) {
1953 uint32_t read_chunk = MIN(size, 0x1000);
1954 qemu_get_buffer(f, buf, read_chunk);
1955 size -= read_chunk;
1957 g_free(buf);
1961 cpu_synchronize_all_post_init();
1963 return ret;
1966 void hmp_savevm(Monitor *mon, const QDict *qdict)
1968 BlockDriverState *bs, *bs1;
1969 QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1;
1970 int ret;
1971 QEMUFile *f;
1972 int saved_vm_running;
1973 uint64_t vm_state_size;
1974 qemu_timeval tv;
1975 struct tm tm;
1976 const char *name = qdict_get_try_str(qdict, "name");
1977 Error *local_err = NULL;
1978 AioContext *aio_context;
1980 if (!bdrv_all_can_snapshot(&bs)) {
1981 monitor_printf(mon, "Device '%s' is writable but does not "
1982 "support snapshots.\n", bdrv_get_device_name(bs));
1983 return;
1986 /* Delete old snapshots of the same name */
1987 if (name && bdrv_all_delete_snapshot(name, &bs1, &local_err) < 0) {
1988 error_reportf_err(local_err,
1989 "Error while deleting snapshot on device '%s': ",
1990 bdrv_get_device_name(bs1));
1991 return;
1994 bs = bdrv_all_find_vmstate_bs();
1995 if (bs == NULL) {
1996 monitor_printf(mon, "No block device can accept snapshots\n");
1997 return;
1999 aio_context = bdrv_get_aio_context(bs);
2001 saved_vm_running = runstate_is_running();
2003 ret = global_state_store();
2004 if (ret) {
2005 monitor_printf(mon, "Error saving global state\n");
2006 return;
2008 vm_stop(RUN_STATE_SAVE_VM);
2010 aio_context_acquire(aio_context);
2012 memset(sn, 0, sizeof(*sn));
2014 /* fill auxiliary fields */
2015 qemu_gettimeofday(&tv);
2016 sn->date_sec = tv.tv_sec;
2017 sn->date_nsec = tv.tv_usec * 1000;
2018 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
2020 if (name) {
2021 ret = bdrv_snapshot_find(bs, old_sn, name);
2022 if (ret >= 0) {
2023 pstrcpy(sn->name, sizeof(sn->name), old_sn->name);
2024 pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str);
2025 } else {
2026 pstrcpy(sn->name, sizeof(sn->name), name);
2028 } else {
2029 /* cast below needed for OpenBSD where tv_sec is still 'long' */
2030 localtime_r((const time_t *)&tv.tv_sec, &tm);
2031 strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm);
2034 /* save the VM state */
2035 f = qemu_fopen_bdrv(bs, 1);
2036 if (!f) {
2037 monitor_printf(mon, "Could not open VM state file\n");
2038 goto the_end;
2040 ret = qemu_savevm_state(f, &local_err);
2041 vm_state_size = qemu_ftell(f);
2042 qemu_fclose(f);
2043 if (ret < 0) {
2044 error_report_err(local_err);
2045 goto the_end;
2048 ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, &bs);
2049 if (ret < 0) {
2050 monitor_printf(mon, "Error while creating snapshot on '%s'\n",
2051 bdrv_get_device_name(bs));
2054 the_end:
2055 aio_context_release(aio_context);
2056 if (saved_vm_running) {
2057 vm_start();
2061 void qmp_xen_save_devices_state(const char *filename, Error **errp)
2063 QEMUFile *f;
2064 int saved_vm_running;
2065 int ret;
2067 saved_vm_running = runstate_is_running();
2068 vm_stop(RUN_STATE_SAVE_VM);
2069 global_state_store_running();
2071 f = qemu_fopen(filename, "wb");
2072 if (!f) {
2073 error_setg_file_open(errp, errno, filename);
2074 goto the_end;
2076 ret = qemu_save_device_state(f);
2077 qemu_fclose(f);
2078 if (ret < 0) {
2079 error_setg(errp, QERR_IO_ERROR);
2082 the_end:
2083 if (saved_vm_running) {
2084 vm_start();
2088 int load_vmstate(const char *name)
2090 BlockDriverState *bs, *bs_vm_state;
2091 QEMUSnapshotInfo sn;
2092 QEMUFile *f;
2093 int ret;
2094 AioContext *aio_context;
2096 if (!bdrv_all_can_snapshot(&bs)) {
2097 error_report("Device '%s' is writable but does not support snapshots.",
2098 bdrv_get_device_name(bs));
2099 return -ENOTSUP;
2101 ret = bdrv_all_find_snapshot(name, &bs);
2102 if (ret < 0) {
2103 error_report("Device '%s' does not have the requested snapshot '%s'",
2104 bdrv_get_device_name(bs), name);
2105 return ret;
2108 bs_vm_state = bdrv_all_find_vmstate_bs();
2109 if (!bs_vm_state) {
2110 error_report("No block device supports snapshots");
2111 return -ENOTSUP;
2113 aio_context = bdrv_get_aio_context(bs_vm_state);
2115 /* Don't even try to load empty VM states */
2116 aio_context_acquire(aio_context);
2117 ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
2118 aio_context_release(aio_context);
2119 if (ret < 0) {
2120 return ret;
2121 } else if (sn.vm_state_size == 0) {
2122 error_report("This is a disk-only snapshot. Revert to it offline "
2123 "using qemu-img.");
2124 return -EINVAL;
2127 /* Flush all IO requests so they don't interfere with the new state. */
2128 bdrv_drain_all();
2130 ret = bdrv_all_goto_snapshot(name, &bs);
2131 if (ret < 0) {
2132 error_report("Error %d while activating snapshot '%s' on '%s'",
2133 ret, name, bdrv_get_device_name(bs));
2134 return ret;
2137 /* restore the VM state */
2138 f = qemu_fopen_bdrv(bs_vm_state, 0);
2139 if (!f) {
2140 error_report("Could not open VM state file");
2141 return -EINVAL;
2144 qemu_system_reset(VMRESET_SILENT);
2145 migration_incoming_state_new(f);
2147 aio_context_acquire(aio_context);
2148 ret = qemu_loadvm_state(f);
2149 qemu_fclose(f);
2150 aio_context_release(aio_context);
2152 migration_incoming_state_destroy();
2153 if (ret < 0) {
2154 error_report("Error %d while loading VM state", ret);
2155 return ret;
2158 return 0;
2161 void hmp_delvm(Monitor *mon, const QDict *qdict)
2163 BlockDriverState *bs;
2164 Error *err;
2165 const char *name = qdict_get_str(qdict, "name");
2167 if (bdrv_all_delete_snapshot(name, &bs, &err) < 0) {
2168 error_reportf_err(err,
2169 "Error while deleting snapshot on device '%s': ",
2170 bdrv_get_device_name(bs));
2174 void hmp_info_snapshots(Monitor *mon, const QDict *qdict)
2176 BlockDriverState *bs, *bs1;
2177 QEMUSnapshotInfo *sn_tab, *sn;
2178 int nb_sns, i;
2179 int total;
2180 int *available_snapshots;
2181 AioContext *aio_context;
2183 bs = bdrv_all_find_vmstate_bs();
2184 if (!bs) {
2185 monitor_printf(mon, "No available block device supports snapshots\n");
2186 return;
2188 aio_context = bdrv_get_aio_context(bs);
2190 aio_context_acquire(aio_context);
2191 nb_sns = bdrv_snapshot_list(bs, &sn_tab);
2192 aio_context_release(aio_context);
2194 if (nb_sns < 0) {
2195 monitor_printf(mon, "bdrv_snapshot_list: error %d\n", nb_sns);
2196 return;
2199 if (nb_sns == 0) {
2200 monitor_printf(mon, "There is no snapshot available.\n");
2201 return;
2204 available_snapshots = g_new0(int, nb_sns);
2205 total = 0;
2206 for (i = 0; i < nb_sns; i++) {
2207 if (bdrv_all_find_snapshot(sn_tab[i].id_str, &bs1) == 0) {
2208 available_snapshots[total] = i;
2209 total++;
2213 if (total > 0) {
2214 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, NULL);
2215 monitor_printf(mon, "\n");
2216 for (i = 0; i < total; i++) {
2217 sn = &sn_tab[available_snapshots[i]];
2218 bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, sn);
2219 monitor_printf(mon, "\n");
2221 } else {
2222 monitor_printf(mon, "There is no suitable snapshot available\n");
2225 g_free(sn_tab);
2226 g_free(available_snapshots);
2230 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
2232 qemu_ram_set_idstr(mr->ram_block,
2233 memory_region_name(mr), dev);
2236 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
2238 qemu_ram_unset_idstr(mr->ram_block);
2241 void vmstate_register_ram_global(MemoryRegion *mr)
2243 vmstate_register_ram(mr, NULL);